[ { "path": ".gitignore", "content": "# OS X files\n.DS_Store\n\n# Node.js files\nnode_modules\n" }, { "path": "README.md", "content": "# Netscope CNN Analyzer\n\navailable here: http://dgschwend.github.io/netscope \n\nThis is a CNN Analyzer tool, based on Netscope by [ethereon](https://github.com/ethereon).\nNetscope is a web-based tool for visualizing neural network topologies. It currently supports UC Berkeley's [Caffe framework](https://github.com/bvlc/caffe).\n\nThis fork adds analysis capabilities, enabling the computation of network complexity (number of operations) and network size (number of parameters) for easy comparison of different networks.\n\n### Documentation\n- Netscope [Quick Start Guide](http://dgschwend.github.io/netscope/quickstart.html)\n\n### Demo\n- :new: [Visualization of ZynqNet CNN](http://dgschwend.github.io/netscope/#/preset/zynqnet)\n- [Visualization of the Deep Convolutional Neural Network \"SqueezeNet\"](http://dgschwend.github.io/netscope/#/preset/squeezenet)\n\n### License\n\nReleased under the MIT license.\nAll included network models provided under their respective licenses.\n" }, { "path": "assets/css/codemirror.css", "content": "/* BASICS */\n\n.CodeMirror {\n /* Set height, width, borders, and global font properties here */\n font-family: monospace;\n height: 300px;\n color: black;\n}\n\n/* PADDING */\n\n.CodeMirror-lines {\n padding: 4px 0; /* Vertical padding around content */\n}\n.CodeMirror pre {\n padding: 0 4px; /* Horizontal padding of content */\n}\n\n.CodeMirror-scrollbar-filler, .CodeMirror-gutter-filler {\n background-color: white; /* The little square between H and V scrollbars */\n}\n\n/* GUTTER */\n\n.CodeMirror-gutters {\n border-right: 1px solid #ddd;\n background-color: #f7f7f7;\n white-space: nowrap;\n}\n.CodeMirror-linenumbers {}\n.CodeMirror-linenumber {\n padding: 0 3px 0 5px;\n min-width: 20px;\n text-align: right;\n color: #999;\n white-space: nowrap;\n}\n\n.CodeMirror-guttermarker { color: black; }\n.CodeMirror-guttermarker-subtle { color: #999; }\n\n/* CURSOR */\n\n.CodeMirror div.CodeMirror-cursor {\n border-left: 1px solid black;\n}\n/* Shown when moving in bi-directional text */\n.CodeMirror div.CodeMirror-secondarycursor {\n border-left: 1px solid silver;\n}\n.CodeMirror.cm-fat-cursor div.CodeMirror-cursor {\n width: auto;\n border: 0;\n background: #7e7;\n}\n.CodeMirror.cm-fat-cursor div.CodeMirror-cursors {\n z-index: 1;\n}\n\n.cm-animate-fat-cursor {\n width: auto;\n border: 0;\n -webkit-animation: blink 1.06s steps(1) infinite;\n -moz-animation: blink 1.06s steps(1) infinite;\n animation: blink 1.06s steps(1) infinite;\n background-color: #7e7;\n}\n@-moz-keyframes blink {\n 0% {}\n 50% { background-color: transparent; }\n 100% {}\n}\n@-webkit-keyframes blink {\n 0% {}\n 50% { background-color: transparent; }\n 100% {}\n}\n@keyframes blink {\n 0% {}\n 50% { background-color: transparent; }\n 100% {}\n}\n\n/* Can style cursor different in overwrite (non-insert) mode */\ndiv.CodeMirror-overwrite div.CodeMirror-cursor {}\n\n.cm-tab { display: inline-block; text-decoration: inherit; }\n\n.CodeMirror-ruler {\n border-left: 1px solid #ccc;\n position: absolute;\n}\n\n/* DEFAULT THEME */\n\n.CodeMirror .CodeMirror-scroll {\n background-color: #2c2c2c;\n color: #E6E1DC;\n}\n\n.CodeMirror .CodeMirror-gutters {\n background-color: #3B3B3B;\n border-right: none;\n}\n\n.CodeMirror-focused .CodeMirror-activeline-background {\n background: #232323 !important;\n}\n\n.show-line-padding .CodeMirror-focused .CodeMirror-activeline-background {\n box-shadow: inset 15px 0 0 0 #000;\n}\n\n.CodeMirror-focused .CodeMirror-activeline .CodeMirror-gutter-elt {\n background: #232323;\n color: #fff;\n}\n\n.CodeMirror-focused .CodeMirror-activeline .inline-widget .CodeMirror-gutter-elt {\n color: #787878;\n}\n\n.CodeMirror.find-highlighting div.CodeMirror-selected,\n.CodeMirror .CodeMirror.find-highlighting div.CodeMirror-selected {\n background: red;\n}\n\n.CodeMirror-cursor {\n border-left: 1px solid #c5c8c6 !important;\n}\n\n.CodeMirror-overwrite .CodeMirror-cursor {\n border-left: none !important;\n border-bottom: 1px solid #c5c8c6;\n}\n\n.CodeMirror-focused .CodeMirror-activeline .CodeMirror-gutter-elt {\n color: #fff;\n}\n\n.CodeMirror .CodeMirror-selected {\n background: #333333;\n}\n.CodeMirror-focused .CodeMirror-selected {\n background: #3c3c3c;\n}\n\n.CodeMirror-matchingbracket, .CodeMirror-matchingtag {\n /* Ensure visibility against gray inline editor background */\n background-color: #000000;\n color: #8bff04 !important;\n}\n\n.cm-atom { color: #E0E2E4; }\n.cm-qualifier { color: #FFC66D; }\n.cm-def { color: #D1E0DD; }\n.cm-keyword { color: #EE786E; }\n.cm-comment { color: #9D9B96; }\n.cm-string {color: #CA99C2;}\n.cm-string-2 {color: #A5C261;}\n.cm-variable { color: #D0D0FF; }\n.cm-variable-2 { color: #D0D0FF; }\n.cm-variable-3 { color: #D0D0FF; }\n.cm-tag { color: #E8BF6A; }\n.cm-attribute { color: #FFC66D; }\n.cm-property { color: #6E9CBE; }\n.cm-number {color: #FDCD50; }\n.cm-meta {color: #FFC66D; }\n.cm-operator {color: #CC7833; }\n.cm-builtin { color: #CC7833; }\n.cm-hr { color: #A5C261; }\n.cm-plus { color: #FFC66D; }\n.cm-bracket { color: #9D9B96; }\n.cm-error { color: #bc322f; }\n.cm-minus { color: #bc322f; }\n.cm-quote { color: #CA99C2; }\n\n/* Non-editor styling */\n\n#image-holder,\n#not-editor {\n background-color: #2c2c2c;\n}\n\n#image-holder {\n color: #E6E1DC;\n}\n\n\n.CodeMirror-composing { border-bottom: 2px solid; }\n\n/* Default styles for common addons */\n\ndiv.CodeMirror span.CodeMirror-matchingbracket {color: #0f0;}\ndiv.CodeMirror span.CodeMirror-nonmatchingbracket {color: #f22;}\n.CodeMirror-matchingtag { background: rgba(255, 150, 0, .3); }\n.CodeMirror-activeline-background {background: #e8f2ff;}\n\n/* STOP */\n\n/* The rest of this file contains styles related to the mechanics of\n the editor. You probably shouldn't touch them. */\n\n.CodeMirror {\n position: relative;\n overflow: hidden;\n background: white;\n}\n\n.CodeMirror-scroll {\n overflow: scroll !important; /* Things will break if this is overridden */\n /* 30px is the magic margin used to hide the element's real scrollbars */\n /* See overflow: hidden in .CodeMirror */\n margin-bottom: -30px; margin-right: -30px;\n padding-bottom: 30px;\n height: 100%;\n outline: none; /* Prevent dragging from highlighting the element */\n position: relative;\n}\n.CodeMirror-sizer {\n position: relative;\n border-right: 30px solid transparent;\n}\n\n/* The fake, visible scrollbars. Used to force redraw during scrolling\n before actuall scrolling happens, thus preventing shaking and\n flickering artifacts. */\n.CodeMirror-vscrollbar, .CodeMirror-hscrollbar, .CodeMirror-scrollbar-filler, .CodeMirror-gutter-filler {\n position: absolute;\n z-index: 6;\n display: none;\n}\n.CodeMirror-vscrollbar {\n right: 0; top: 0;\n overflow-x: hidden;\n overflow-y: scroll;\n}\n.CodeMirror-hscrollbar {\n bottom: 0; left: 0;\n overflow-y: hidden;\n overflow-x: scroll;\n}\n.CodeMirror-scrollbar-filler {\n right: 0; bottom: 0;\n}\n.CodeMirror-gutter-filler {\n left: 0; bottom: 0;\n}\n\n.CodeMirror-gutters {\n position: absolute; left: 0; top: 0;\n z-index: 3;\n}\n.CodeMirror-gutter {\n white-space: normal;\n height: 100%;\n display: inline-block;\n margin-bottom: -30px;\n /* Hack to make IE7 behave */\n *zoom:1;\n *display:inline;\n}\n.CodeMirror-gutter-wrapper {\n position: absolute;\n z-index: 4;\n background: none !important;\n border: none !important;\n}\n.CodeMirror-gutter-background {\n position: absolute;\n top: 0; bottom: 0;\n z-index: 4;\n}\n.CodeMirror-gutter-elt {\n position: absolute;\n cursor: default;\n z-index: 4;\n}\n.CodeMirror-gutter-wrapper {\n -webkit-user-select: none;\n -moz-user-select: none;\n user-select: none;\n}\n\n.CodeMirror-lines {\n cursor: text;\n min-height: 1px; /* prevents collapsing before first draw */\n}\n.CodeMirror pre {\n /* Reset some styles that the rest of the page might have set */\n -moz-border-radius: 0; -webkit-border-radius: 0; border-radius: 0;\n border-width: 0;\n background: transparent;\n font-family: inherit;\n font-size: inherit;\n margin: 0;\n white-space: pre;\n word-wrap: normal;\n line-height: inherit;\n color: inherit;\n z-index: 2;\n position: relative;\n overflow: visible;\n -webkit-tap-highlight-color: transparent;\n}\n.CodeMirror-wrap pre {\n word-wrap: break-word;\n white-space: pre-wrap;\n word-break: normal;\n}\n\n.CodeMirror-linebackground {\n position: absolute;\n left: 0; right: 0; top: 0; bottom: 0;\n z-index: 0;\n}\n\n.CodeMirror-linewidget {\n position: relative;\n z-index: 2;\n overflow: auto;\n}\n\n.CodeMirror-widget {}\n\n.CodeMirror-code {\n outline: none;\n}\n\n/* Force content-box sizing for the elements where we expect it */\n.CodeMirror-scroll,\n.CodeMirror-sizer,\n.CodeMirror-gutter,\n.CodeMirror-gutters,\n.CodeMirror-linenumber {\n -moz-box-sizing: content-box;\n box-sizing: content-box;\n}\n\n.CodeMirror-measure {\n position: absolute;\n width: 100%;\n height: 0;\n overflow: hidden;\n visibility: hidden;\n}\n.CodeMirror-measure pre { position: static; }\n\n.CodeMirror div.CodeMirror-cursor {\n position: absolute;\n border-right: none;\n width: 0;\n}\n\ndiv.CodeMirror-cursors {\n visibility: hidden;\n position: relative;\n z-index: 3;\n}\n.CodeMirror-focused div.CodeMirror-cursors {\n visibility: visible;\n}\n\n.CodeMirror-selected { background: #d9d9d9; }\n.CodeMirror-focused .CodeMirror-selected { background: #d7d4f0; }\n.CodeMirror-crosshair { cursor: crosshair; }\n.CodeMirror-line::selection, .CodeMirror-line > span::selection, .CodeMirror-line > span > span::selection { background: #d7d4f0; }\n.CodeMirror-line::-moz-selection, .CodeMirror-line > span::-moz-selection, .CodeMirror-line > span > span::-moz-selection { background: #d7d4f0; }\n\n.cm-searching {\n background: #ffa;\n background: rgba(255, 255, 0, .4);\n}\n\n/* IE7 hack to prevent it from returning funny offsetTops on the spans */\n.CodeMirror span { *vertical-align: text-bottom; }\n\n/* Used to force a border model for a node */\n.cm-force-border { padding-right: .1px; }\n\n@media print {\n /* Hide the cursor when printing */\n .CodeMirror div.CodeMirror-cursors {\n visibility: hidden;\n }\n}\n\n/* See issue #2901 */\n.cm-tab-wrap-hack:after { content: ''; }\n\n/* Help users use markselection to safely style text background */\nspan.CodeMirror-selectedtext { background: none; }\n" }, { "path": "assets/css/doc.css", "content": "* { box-sizing: border-box; }\n\nhtml, body {\n font-family: \"Roboto\", \"Helvetica Neue\", Helvetica, Arial, sans-serf;\n background-color: #F7EBE1;\n margin:0;\n padding:0;\n}\n\nh1, h2, h3 {\n color: #554037;\n font-weight: 300;\n}\n\n.container {\n width: 650px;\n margin:25px auto 50px auto;\n color: #87746C;\n line-height: 1.5em;\n text-rendering: optimizelegibility;\n letter-spacing: 0.01rem;\n}\n\n.container a {\n color: #804A3A;\n text-decoration: none;\n}\n\n.container a:hover {\n color: #B13254;\n}\n\ncode, .code-box {\n font-family: Menlo,Monaco,Consolas,\"Courier New\",monospace;\n}\n\ncode {\n padding: 2px 4px;\n font-size: 90%;\n color: #654D37;\n background-color: #DBC8B8;\n border-radius: 4px;\n}\n\n.code-box {\n padding: 9px 14px;\n margin-bottom: 14px;\n color: #654D37;\n background-color: #F2E5DA;\n border: 1px solid #B39A8F;\n border-radius: 4px;\n}\n\n.code-box .highlight {\n color: #B13254;\n}\n\n.btn {\n display: inline-block;\n padding: 6px 12px;\n margin-bottom: 0px;\n font-size: 14px;\n font-weight: 400;\n line-height: 1.42857;\n text-align: center;\n white-space: nowrap;\n vertical-align: middle;\n cursor: pointer;\n -moz-user-select: none;\n background-image: none;\n border: 1px solid #B39A8F;\n border-radius: 4px;\n}\n\n.btn:hover {\n border: 1px solid transparent;\n background-color: #B13254;\n color: #FFF !important;\n}\n\n.preset {\n margin-bottom: 15px;\n}" }, { "path": "assets/css/netscope.css", "content": "* { box-sizing: border-box; }\n\nhtml, body {\n font-family: \"Roboto\", \"Helvetica Neue\", Helvetica, Arial, sans-serf;\n background-color: #F7EBE1;\n margin:0;\n padding:0;\n height: 100%;\n}\n\nh1, h2, h3 {\n color: #554037;\n font-weight: 300;\n}\n/* Layout */\n\n#master-container {\n /*overflow: hidden;\n height: 100%;\n width: 100%;*/\n}\n\n#net-column {\n width: 100%;\n}\n\n.column {\n float: left;\n width: 30%;\n height: 100%;\n overflow: scroll;\n}\n\n/* Editor */\n\n.CodeMirror {\n height: 100% !important;\n font-family: Menlo,Monaco,Consolas,\"Courier New\",monospace;\n font-size: 14px;\n}\n\n/* NetScope Viewer */\n\n#net-spinner {\n padding: 50px;\n text-align: center;\n}\n\n#net-container {\n display: none;\n text-align: center;\n}\n\n#table-container {\n display: none;\n text-align: center; \n padding-top:20px; \n}\n\n#table-header {\n padding-top: 20px;\n border-top: 1px solid #554037;\n min-width: 400px;\n text-align: center;\n}\n\n#table-content {\n text-align: center;\n}\n\n#table-content table {\n display: inline-block;\n text-align: left;\n width:auto;\n \n}\n\n#net-group {\n display: inline-block;\n}\n\n#net-svg {\n}\n\n#net-title {\n padding:20px;\n border-bottom: 1px solid #554037;\n min-width: 400px;\n text-align: center;\n}\n\n#net-title .editlink {\n font-size: 0.3em;\n padding-left: 10px;\n color: #554037;\n cursor: pointer;\n}\n\n#net-error {\n background-color: #C23739;\n color: #FFF;\n padding: 20px;\n border: 1px solid transparent;\n border-radius: 5px;\n margin: 20px auto 20px auto;\n width: 400px;\n font-size: 12px;\n font-family: Menlo,Monaco,Consolas,\"Courier New\",monospace;\n}\n\n#net-error .title {\n border-bottom: 1px solid #FFABAC;\n padding-bottom: 10px;\n margin-bottom: 10px;\n}\n\n#net-error .msg {\n}\n\n.node-label {\n font-size: 14px;\n color: #563621;\n padding: 1px;\n}\n\n.node-label:not(:first-child) {\n padding-top:5px;\n margin-top:5px;\n border-top: 1px solid rgba(255,255,255,0.5);\n}\n\n.node {\n fill: #FFB84D;\n}\n\n.node-type-convolution, .node-type-deconvolution {\n fill: #CA5237;\n}\n\n.node-type-convolution .node-label, .node-type-deconvolution .node-label {\n color: #FFF;\n}\n\n.node-type-inner-product, .node-type-innerproduct {\n fill: #bd0026;\n}\n\n.node-type-inner-product .node-label, .node-type-innerproduct .node-label {\n color: #FFF;\n}\n\n.node-type-batchnorm {\n fill: #84AC7E; /* dirty light green */\n}\n\n.node-type-lrn {\n fill: #A265AC; /* dirty pink */\n fill: #bd0026; /* deep red */\n}\n\n.node-type-lrn .node-label {\n color: #fff;\n}\n\n.node-type-source {\n fill: #37A058;\n}\n\n.node-type-source .node-label {\n color: #FFF;\n}\n\n.node-type-sink {\n fill: #7F6A9B;\n}\n\n.node-type-sink .node-label {\n color: #FFF;\n}\n\n@-webkit-keyframes highlight {\n from {color: #fff;}\n to {color: #000;}\n}\n@keyframes highlight {\n from {color: #fff;}\n to {color: #000;}\n}\n.node-highlight {\n animation: highlight 0.3s linear 0s 4 alternate\n}\n\n.edgePath path {\n stroke: #554037;\n fill: #554037;\n stroke-width: 1.5px;\n}\n.node-info-header {\n}\n.node-param-row {\n}\n.node-param-section {\n border-left: 1px solid #52637D;\n padding-left: 10px;\n}\n.node-param-key {\n color: #91A5BD;\n}\n.node-param-value {\n color: #26BF8C;\n}\n.node-info-title {\n color:#F7BF57;\n}\n.node-info-type {\n color: #C499FF;\n}\n.node-info-annotation {\n color: #69BFFF;\n}\n.node-info-group:not(:first-child) {\n padding-top:5px;\n margin-top:10px;\n border-top: 1px dotted #52637D;\n}" }, { "path": "assets/css/tablesorter.css", "content": ".tablesorter-default{width:100%;color:#333;border-spacing:0;margin:10px 0 15px;text-align:left}.tablesorter-default th,.tablesorter-default thead td{font-weight:700;color:#000;border-collapse:collapse;border-bottom:#ccc 2px solid;padding:0}.tablesorter-default tfoot td,.tablesorter-default tfoot th{border:0}.tablesorter-default .header,.tablesorter-default .tablesorter-header{background-image:url(data:image/gif;base64,R0lGODlhFQAJAIAAACMtMP///yH5BAEAAAEALAAAAAAVAAkAAAIXjI+AywnaYnhUMoqt3gZXPmVg94yJVQAAOw==);background-position:center right;background-repeat:no-repeat;cursor:pointer;white-space:normal;padding:4px 20px 4px 4px}.tablesorter-default thead .headerSortUp,.tablesorter-default thead .tablesorter-headerAsc,.tablesorter-default thead .tablesorter-headerSortUp{background-image:url(data:image/gif;base64,R0lGODlhFQAEAIAAACMtMP///yH5BAEAAAEALAAAAAAVAAQAAAINjI8Bya2wnINUMopZAQA7);border-bottom:#000 2px solid}.tablesorter-default thead .headerSortDown,.tablesorter-default thead .tablesorter-headerDesc,.tablesorter-default thead .tablesorter-headerSortDown{background-image:url(data:image/gif;base64,R0lGODlhFQAEAIAAACMtMP///yH5BAEAAAEALAAAAAAVAAQAAAINjB+gC+jP2ptn0WskLQA7);border-bottom:#000 2px solid}.tablesorter-default thead .sorter-false{background-image:none;cursor:default;padding:4px}.tablesorter-default tfoot .tablesorter-headerAsc,.tablesorter-default tfoot .tablesorter-headerDesc,.tablesorter-default tfoot .tablesorter-headerSortDown,.tablesorter-default tfoot .tablesorter-headerSortUp{border-top:#000 2px solid}.tablesorter-default td{background-color:#fff;border-bottom:#ccc 1px solid;padding:4px;vertical-align:top}.tablesorter-default tbody>tr.even:hover>td,.tablesorter-default tbody>tr.hover>td,.tablesorter-default tbody>tr.odd:hover>td,.tablesorter-default tbody>tr:hover>td{background-color:#fff;color:#000}.tablesorter-default .tablesorter-processing{background-position:center center!important;background-repeat:no-repeat!important;background-image:url(data:image/gif;base64,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)!important}.tablesorter-default tr.odd>td{background-color:#dfdfdf}.tablesorter-default tr.even>td{background-color:#efefef}.tablesorter-default tr.odd td.primary{background-color:#bfbfbf}.tablesorter-default td.primary,.tablesorter-default tr.even td.primary,.tablesorter-default tr.odd td.secondary{background-color:#d9d9d9}.tablesorter-default td.secondary,.tablesorter-default tr.even td.secondary,.tablesorter-default tr.odd td.tertiary{background-color:#e6e6e6}.tablesorter-default td.tertiary,.tablesorter-default tr.even td.tertiary{background-color:#f2f2f2}caption{background-color:#fff}.tablesorter-default .tablesorter-filter-row{background-color:#eee}.tablesorter-default .tablesorter-filter-row td{background-color:#eee;border-bottom:#ccc 1px solid;line-height:normal;text-align:center;-webkit-transition:line-height .1s ease;-moz-transition:line-height .1s ease;-o-transition:line-height .1s ease;transition:line-height .1s ease}.tablesorter-default .tablesorter-filter-row .disabled{opacity:.5;filter:alpha(opacity=50);cursor:not-allowed}.tablesorter-default .tablesorter-filter-row.hideme td{padding:2px;margin:0;line-height:0;cursor:pointer}.tablesorter-default .tablesorter-filter-row.hideme *{height:1px;min-height:0;border:0;padding:0;margin:0;opacity:0;filter:alpha(opacity=0)}.tablesorter-default input.tablesorter-filter,.tablesorter-default select.tablesorter-filter{width:95%;height:auto;margin:4px auto;padding:4px;background-color:#fff;border:1px solid #bbb;color:#333;-webkit-box-sizing:border-box;-moz-box-sizing:border-box;box-sizing:border-box;-webkit-transition:height .1s ease;-moz-transition:height .1s ease;-o-transition:height .1s ease;transition:height .1s ease}.tablesorter .filtered{display:none}.tablesorter .tablesorter-errorRow td{text-align:center;cursor:pointer;background-color:#e6bf99}" }, { "path": "assets/css/tooltip.css", "content": "/*\n * qTip2 - Pretty powerful tooltips - v2.2.1\n * http://qtip2.com\n *\n * Copyright (c) 2014\n * Released under the MIT licenses\n * http://jquery.org/license\n *\n * Date: Sat Sep 6 2014 11:12 GMT+0100+0100\n * Plugins: tips modal viewport svg imagemap ie6\n * Styles: core basic css3\n */\n.qtip{\n\tposition: absolute;\n\tleft: -28000px;\n\ttop: -28000px;\n\tdisplay: none;\n\n\tmax-width: 280px;\n\tmin-width: 50px;\n\n\tfont-size: 10.5px;\n\tline-height: 12px;\n\n\tdirection: ltr;\n\n\tbox-shadow: none;\n\tpadding: 0;\n}\n\n\t.qtip-content{\n\t\tposition: relative;\n\t\tpadding: 5px 9px;\n\t\toverflow: hidden;\n\n\t\ttext-align: left;\n\t\tword-wrap: break-word;\n\t}\n\n\t.qtip-titlebar{\n\t\tposition: relative;\n\t\tpadding: 5px 35px 5px 10px;\n\t\toverflow: hidden;\n\n\t\tborder-width: 0 0 1px;\n\t\tfont-weight: bold;\n\t}\n\n\t.qtip-titlebar + .qtip-content{ border-top-width: 0 !important; }\n\n\t/* Default close button class */\n\t.qtip-close{\n\t\tposition: absolute;\n\t\tright: -9px; top: -9px;\n\t\tz-index: 11; /* Overlap .qtip-tip */\n\n\t\tcursor: pointer;\n\t\toutline: medium none;\n\n\t\tborder: 1px solid transparent;\n\t}\n\n\t\t.qtip-titlebar .qtip-close{\n\t\t\tright: 4px; top: 50%;\n\t\t\tmargin-top: -9px;\n\t\t}\n\n\t\t* html .qtip-titlebar .qtip-close{ top: 16px; } /* IE fix */\n\n\t\t.qtip-titlebar .ui-icon,\n\t\t.qtip-icon .ui-icon{\n\t\t\tdisplay: block;\n\t\t\ttext-indent: -1000em;\n\t\t\tdirection: ltr;\n\t\t}\n\n\t\t.qtip-icon, .qtip-icon .ui-icon{\n\t\t\t-moz-border-radius: 3px;\n\t\t\t-webkit-border-radius: 3px;\n\t\t\tborder-radius: 3px;\n\t\t\ttext-decoration: none;\n\t\t}\n\n\t\t\t.qtip-icon .ui-icon{\n\t\t\t\twidth: 18px;\n\t\t\t\theight: 14px;\n\n\t\t\t\tline-height: 14px;\n\t\t\t\ttext-align: center;\n\t\t\t\ttext-indent: 0;\n\t\t\t\tfont: normal bold 10px/13px Tahoma,sans-serif;\n\n\t\t\t\tcolor: inherit;\n\t\t\t\tbackground: transparent none no-repeat -100em -100em;\n\t\t\t}\n\n/* Applied to 'focused' tooltips e.g. most recently displayed/interacted with */\n.qtip-focus{}\n\n/* Applied on hover of tooltips i.e. added/removed on mouseenter/mouseleave respectively */\n.qtip-hover{}\n\n/* Default tooltip style */\n.qtip-default{\n\tborder: 1px solid #31343F;\n\tborder-radius: 5px;\n\tbackground-color: #31343F;\n\tcolor: #52637D;\n\n\tfont-family: 'Roboto';\n\tfont-size: 12px;\n\tline-height: 18px;\n}\n\n\t.qtip-default .qtip-icon{\n\t\tborder-color: #CCC;\n\t\tbackground: #F1F1F1;\n\t\tcolor: #777;\n\t}\n\n\t.qtip-default .qtip-titlebar .qtip-close{\n\t\tborder-color: #AAA;\n\t\tcolor: #111;\n\t}\n\n.qtip-shadow{\n\t-webkit-box-shadow: 1px 1px 3px 1px rgba(0, 0, 0, 0.15);\n\t-moz-box-shadow: 1px 1px 3px 1px rgba(0, 0, 0, 0.15);\n\tbox-shadow: 1px 1px 3px 1px rgba(0, 0, 0, 0.15);\n}\n\n.qtip-rounded .qtip-titlebar{\n\t-moz-border-radius: 4px 4px 0 0;\n\t-webkit-border-radius: 4px 4px 0 0;\n\tborder-radius: 4px 4px 0 0;\n}\n\n/* IE9 fix - removes all filters */\n.qtip:not(.ie9haxors) div.qtip-content,\n.qtip:not(.ie9haxors) div.qtip-titlebar{\n\tfilter: none;\n\t-ms-filter: none;\n}\n\n\n.qtip .qtip-tip{\n\tmargin: 0 auto;\n\toverflow: hidden;\n\tz-index: 10;\n\n}\n\n\t/* Opera bug #357 - Incorrect tip position\n\thttps://github.com/Craga89/qTip2/issues/367 */\n\tx:-o-prefocus, .qtip .qtip-tip{\n\t\tvisibility: hidden;\n\t}\n\n\t.qtip .qtip-tip,\n\t.qtip .qtip-tip .qtip-vml,\n\t.qtip .qtip-tip canvas{\n\t\tposition: absolute;\n\n\t\tcolor: #123456;\n\t\tbackground: transparent;\n\t\tborder: 0 dashed transparent;\n\t}\n\n\t.qtip .qtip-tip canvas{ top: 0; left: 0; }\n\n\t.qtip .qtip-tip .qtip-vml{\n\t\tbehavior: url(#default#VML);\n\t\tdisplay: inline-block;\n\t\tvisibility: visible;\n\t}\n\n\n#qtip-overlay{\n\tposition: fixed;\n\tleft: 0; top: 0;\n\twidth: 100%; height: 100%;\n}\n\n\t/* Applied to modals with show.modal.blur set to true */\n\t#qtip-overlay.blurs{ cursor: pointer; }\n\n\t/* Change opacity of overlay here */\n\t#qtip-overlay div{\n\t\tposition: absolute;\n\t\tleft: 0; top: 0;\n\t\twidth: 100%; height: 100%;\n\n\t\tbackground-color: black;\n\n\t\topacity: 0.7;\n\t\tfilter:alpha(opacity=70);\n\t\t-ms-filter:\"progid:DXImageTransform.Microsoft.Alpha(Opacity=70)\";\n\t}\n\n\n.qtipmodal-ie6fix{\n\tposition: absolute !important;\n}" }, { "path": "assets/js/netscope.js", "content": "(function e(t,n,r){function s(o,u){if(!n[o]){if(!t[o]){var a=typeof require==\"function\"&&require;if(!u&&a)return a(o,!0);if(i)return i(o,!0);var f=new Error(\"Cannot find module '\"+o+\"'\");throw f.code=\"MODULE_NOT_FOUND\",f}var l=n[o]={exports:{}};t[o][0].call(l.exports,function(e){var n=t[o][1][e];return s(n?n:e)},l,l.exports,e,t,n,r)}return n[o].exports}var i=typeof require==\"function\"&&require;for(var o=0;o= 0 && j < len ? [ this[ j ] ] : [] );\n\t},\n\n\tend: function() {\n\t\treturn this.prevObject || this.constructor();\n\t},\n\n\t// For internal use only.\n\t// Behaves like an Array's method, not like a jQuery method.\n\tpush: push,\n\tsort: arr.sort,\n\tsplice: arr.splice\n};\n\njQuery.extend = jQuery.fn.extend = function() {\n\tvar options, name, src, copy, copyIsArray, clone,\n\t\ttarget = arguments[ 0 ] || {},\n\t\ti = 1,\n\t\tlength = arguments.length,\n\t\tdeep = false;\n\n\t// Handle a deep copy situation\n\tif ( typeof target === \"boolean\" ) {\n\t\tdeep = target;\n\n\t\t// Skip the boolean and the target\n\t\ttarget = arguments[ i ] || {};\n\t\ti++;\n\t}\n\n\t// Handle case when target is a string or something (possible in deep copy)\n\tif ( typeof target !== \"object\" && !jQuery.isFunction( target ) ) {\n\t\ttarget = {};\n\t}\n\n\t// Extend jQuery itself if only one argument is passed\n\tif ( i === length ) {\n\t\ttarget = this;\n\t\ti--;\n\t}\n\n\tfor ( ; i < length; i++ ) {\n\n\t\t// Only deal with non-null/undefined values\n\t\tif ( ( options = arguments[ i ] ) != null ) {\n\n\t\t\t// Extend the base object\n\t\t\tfor ( name in options ) {\n\t\t\t\tsrc = target[ name ];\n\t\t\t\tcopy = options[ name ];\n\n\t\t\t\t// Prevent never-ending loop\n\t\t\t\tif ( target === copy ) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\n\t\t\t\t// Recurse if we're merging plain objects or arrays\n\t\t\t\tif ( deep && copy && ( jQuery.isPlainObject( copy ) ||\n\t\t\t\t\t( copyIsArray = Array.isArray( copy ) ) ) ) {\n\n\t\t\t\t\tif ( copyIsArray ) {\n\t\t\t\t\t\tcopyIsArray = false;\n\t\t\t\t\t\tclone = src && Array.isArray( src ) ? src : [];\n\n\t\t\t\t\t} else {\n\t\t\t\t\t\tclone = src && jQuery.isPlainObject( src ) ? src : {};\n\t\t\t\t\t}\n\n\t\t\t\t\t// Never move original objects, clone them\n\t\t\t\t\ttarget[ name ] = jQuery.extend( deep, clone, copy );\n\n\t\t\t\t// Don't bring in undefined values\n\t\t\t\t} else if ( copy !== undefined ) {\n\t\t\t\t\ttarget[ name ] = copy;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Return the modified object\n\treturn target;\n};\n\njQuery.extend( {\n\n\t// Unique for each copy of jQuery on the page\n\texpando: \"jQuery\" + ( version + Math.random() ).replace( /\\D/g, \"\" ),\n\n\t// Assume jQuery is ready without the ready module\n\tisReady: true,\n\n\terror: function( msg ) {\n\t\tthrow new Error( msg );\n\t},\n\n\tnoop: function() {},\n\n\tisFunction: function( obj ) {\n\t\treturn jQuery.type( obj ) === \"function\";\n\t},\n\n\tisWindow: function( obj ) {\n\t\treturn obj != null && obj === obj.window;\n\t},\n\n\tisNumeric: function( obj ) {\n\n\t\t// As of jQuery 3.0, isNumeric is limited to\n\t\t// strings and numbers (primitives or objects)\n\t\t// that can be coerced to finite numbers (gh-2662)\n\t\tvar type = jQuery.type( obj );\n\t\treturn ( type === \"number\" || type === \"string\" ) &&\n\n\t\t\t// parseFloat NaNs numeric-cast false positives (\"\")\n\t\t\t// ...but misinterprets leading-number strings, particularly hex literals (\"0x...\")\n\t\t\t// subtraction forces infinities to NaN\n\t\t\t!isNaN( obj - parseFloat( obj ) );\n\t},\n\n\tisPlainObject: function( obj ) {\n\t\tvar proto, Ctor;\n\n\t\t// Detect obvious negatives\n\t\t// Use toString instead of jQuery.type to catch host objects\n\t\tif ( !obj || toString.call( obj ) !== \"[object Object]\" ) {\n\t\t\treturn false;\n\t\t}\n\n\t\tproto = getProto( obj );\n\n\t\t// Objects with no prototype (e.g., `Object.create( null )`) are plain\n\t\tif ( !proto ) {\n\t\t\treturn true;\n\t\t}\n\n\t\t// Objects with prototype are plain iff they were constructed by a global Object function\n\t\tCtor = hasOwn.call( proto, \"constructor\" ) && proto.constructor;\n\t\treturn typeof Ctor === \"function\" && fnToString.call( Ctor ) === ObjectFunctionString;\n\t},\n\n\tisEmptyObject: function( obj ) {\n\n\t\t/* eslint-disable no-unused-vars */\n\t\t// See https://github.com/eslint/eslint/issues/6125\n\t\tvar name;\n\n\t\tfor ( name in obj ) {\n\t\t\treturn false;\n\t\t}\n\t\treturn true;\n\t},\n\n\ttype: function( obj ) {\n\t\tif ( obj == null ) {\n\t\t\treturn obj + \"\";\n\t\t}\n\n\t\t// Support: Android <=2.3 only (functionish RegExp)\n\t\treturn typeof obj === \"object\" || typeof obj === \"function\" ?\n\t\t\tclass2type[ toString.call( obj ) ] || \"object\" :\n\t\t\ttypeof obj;\n\t},\n\n\t// Evaluates a script in a global context\n\tglobalEval: function( code ) {\n\t\tDOMEval( code );\n\t},\n\n\t// Convert dashed to camelCase; used by the css and data modules\n\t// Support: IE <=9 - 11, Edge 12 - 13\n\t// Microsoft forgot to hump their vendor prefix (#9572)\n\tcamelCase: function( string ) {\n\t\treturn string.replace( rmsPrefix, \"ms-\" ).replace( rdashAlpha, fcamelCase );\n\t},\n\n\teach: function( obj, callback ) {\n\t\tvar length, i = 0;\n\n\t\tif ( isArrayLike( obj ) ) {\n\t\t\tlength = obj.length;\n\t\t\tfor ( ; i < length; i++ ) {\n\t\t\t\tif ( callback.call( obj[ i ], i, obj[ i ] ) === false ) {\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tfor ( i in obj ) {\n\t\t\t\tif ( callback.call( obj[ i ], i, obj[ i ] ) === false ) {\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn obj;\n\t},\n\n\t// Support: Android <=4.0 only\n\ttrim: function( text ) {\n\t\treturn text == null ?\n\t\t\t\"\" :\n\t\t\t( text + \"\" ).replace( rtrim, \"\" );\n\t},\n\n\t// results is for internal usage only\n\tmakeArray: function( arr, results ) {\n\t\tvar ret = results || [];\n\n\t\tif ( arr != null ) {\n\t\t\tif ( isArrayLike( Object( arr ) ) ) {\n\t\t\t\tjQuery.merge( ret,\n\t\t\t\t\ttypeof arr === \"string\" ?\n\t\t\t\t\t[ arr ] : arr\n\t\t\t\t);\n\t\t\t} else {\n\t\t\t\tpush.call( ret, arr );\n\t\t\t}\n\t\t}\n\n\t\treturn ret;\n\t},\n\n\tinArray: function( elem, arr, i ) {\n\t\treturn arr == null ? -1 : indexOf.call( arr, elem, i );\n\t},\n\n\t// Support: Android <=4.0 only, PhantomJS 1 only\n\t// push.apply(_, arraylike) throws on ancient WebKit\n\tmerge: function( first, second ) {\n\t\tvar len = +second.length,\n\t\t\tj = 0,\n\t\t\ti = first.length;\n\n\t\tfor ( ; j < len; j++ ) {\n\t\t\tfirst[ i++ ] = second[ j ];\n\t\t}\n\n\t\tfirst.length = i;\n\n\t\treturn first;\n\t},\n\n\tgrep: function( elems, callback, invert ) {\n\t\tvar callbackInverse,\n\t\t\tmatches = [],\n\t\t\ti = 0,\n\t\t\tlength = elems.length,\n\t\t\tcallbackExpect = !invert;\n\n\t\t// Go through the array, only saving the items\n\t\t// that pass the validator function\n\t\tfor ( ; i < length; i++ ) {\n\t\t\tcallbackInverse = !callback( elems[ i ], i );\n\t\t\tif ( callbackInverse !== callbackExpect ) {\n\t\t\t\tmatches.push( elems[ i ] );\n\t\t\t}\n\t\t}\n\n\t\treturn matches;\n\t},\n\n\t// arg is for internal usage only\n\tmap: function( elems, callback, arg ) {\n\t\tvar length, value,\n\t\t\ti = 0,\n\t\t\tret = [];\n\n\t\t// Go through the array, translating each of the items to their new values\n\t\tif ( isArrayLike( elems ) ) {\n\t\t\tlength = elems.length;\n\t\t\tfor ( ; i < length; i++ ) {\n\t\t\t\tvalue = callback( elems[ i ], i, arg );\n\n\t\t\t\tif ( value != null ) {\n\t\t\t\t\tret.push( value );\n\t\t\t\t}\n\t\t\t}\n\n\t\t// Go through every key on the object,\n\t\t} else {\n\t\t\tfor ( i in elems ) {\n\t\t\t\tvalue = callback( elems[ i ], i, arg );\n\n\t\t\t\tif ( value != null ) {\n\t\t\t\t\tret.push( value );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Flatten any nested arrays\n\t\treturn concat.apply( [], ret );\n\t},\n\n\t// A global GUID counter for objects\n\tguid: 1,\n\n\t// Bind a function to a context, optionally partially applying any\n\t// arguments.\n\tproxy: function( fn, context ) {\n\t\tvar tmp, args, proxy;\n\n\t\tif ( typeof context === \"string\" ) {\n\t\t\ttmp = fn[ context ];\n\t\t\tcontext = fn;\n\t\t\tfn = tmp;\n\t\t}\n\n\t\t// Quick check to determine if target is callable, in the spec\n\t\t// this throws a TypeError, but we will just return undefined.\n\t\tif ( !jQuery.isFunction( fn ) ) {\n\t\t\treturn undefined;\n\t\t}\n\n\t\t// Simulated bind\n\t\targs = slice.call( arguments, 2 );\n\t\tproxy = function() {\n\t\t\treturn fn.apply( context || this, args.concat( slice.call( arguments ) ) );\n\t\t};\n\n\t\t// Set the guid of unique handler to the same of original handler, so it can be removed\n\t\tproxy.guid = fn.guid = fn.guid || jQuery.guid++;\n\n\t\treturn proxy;\n\t},\n\n\tnow: Date.now,\n\n\t// jQuery.support is not used in Core but other projects attach their\n\t// properties to it so it needs to exist.\n\tsupport: support\n} );\n\nif ( typeof Symbol === \"function\" ) {\n\tjQuery.fn[ Symbol.iterator ] = arr[ Symbol.iterator ];\n}\n\n// Populate the class2type map\njQuery.each( \"Boolean Number String Function Array Date RegExp Object Error Symbol\".split( \" \" ),\nfunction( i, name ) {\n\tclass2type[ \"[object \" + name + \"]\" ] = name.toLowerCase();\n} );\n\nfunction isArrayLike( obj ) {\n\n\t// Support: real iOS 8.2 only (not reproducible in simulator)\n\t// `in` check used to prevent JIT error (gh-2145)\n\t// hasOwn isn't used here due to false negatives\n\t// regarding Nodelist length in IE\n\tvar length = !!obj && \"length\" in obj && obj.length,\n\t\ttype = jQuery.type( obj );\n\n\tif ( type === \"function\" || jQuery.isWindow( obj ) ) {\n\t\treturn false;\n\t}\n\n\treturn type === \"array\" || length === 0 ||\n\t\ttypeof length === \"number\" && length > 0 && ( length - 1 ) in obj;\n}\nvar Sizzle =\n/*!\n * Sizzle CSS Selector Engine v2.3.3\n * https://sizzlejs.com/\n *\n * Copyright jQuery Foundation and other contributors\n * Released under the MIT license\n * http://jquery.org/license\n *\n * Date: 2016-08-08\n */\n(function( window ) {\n\nvar i,\n\tsupport,\n\tExpr,\n\tgetText,\n\tisXML,\n\ttokenize,\n\tcompile,\n\tselect,\n\toutermostContext,\n\tsortInput,\n\thasDuplicate,\n\n\t// Local document vars\n\tsetDocument,\n\tdocument,\n\tdocElem,\n\tdocumentIsHTML,\n\trbuggyQSA,\n\trbuggyMatches,\n\tmatches,\n\tcontains,\n\n\t// Instance-specific data\n\texpando = \"sizzle\" + 1 * new Date(),\n\tpreferredDoc = window.document,\n\tdirruns = 0,\n\tdone = 0,\n\tclassCache = createCache(),\n\ttokenCache = createCache(),\n\tcompilerCache = createCache(),\n\tsortOrder = function( a, b ) {\n\t\tif ( a === b ) {\n\t\t\thasDuplicate = true;\n\t\t}\n\t\treturn 0;\n\t},\n\n\t// Instance methods\n\thasOwn = ({}).hasOwnProperty,\n\tarr = [],\n\tpop = arr.pop,\n\tpush_native = arr.push,\n\tpush = arr.push,\n\tslice = arr.slice,\n\t// Use a stripped-down indexOf as it's faster than native\n\t// https://jsperf.com/thor-indexof-vs-for/5\n\tindexOf = function( list, elem ) {\n\t\tvar i = 0,\n\t\t\tlen = list.length;\n\t\tfor ( ; i < len; i++ ) {\n\t\t\tif ( list[i] === elem ) {\n\t\t\t\treturn i;\n\t\t\t}\n\t\t}\n\t\treturn -1;\n\t},\n\n\tbooleans = \"checked|selected|async|autofocus|autoplay|controls|defer|disabled|hidden|ismap|loop|multiple|open|readonly|required|scoped\",\n\n\t// Regular expressions\n\n\t// http://www.w3.org/TR/css3-selectors/#whitespace\n\twhitespace = \"[\\\\x20\\\\t\\\\r\\\\n\\\\f]\",\n\n\t// http://www.w3.org/TR/CSS21/syndata.html#value-def-identifier\n\tidentifier = \"(?:\\\\\\\\.|[\\\\w-]|[^\\0-\\\\xa0])+\",\n\n\t// Attribute selectors: http://www.w3.org/TR/selectors/#attribute-selectors\n\tattributes = \"\\\\[\" + whitespace + \"*(\" + identifier + \")(?:\" + whitespace +\n\t\t// Operator (capture 2)\n\t\t\"*([*^$|!~]?=)\" + whitespace +\n\t\t// \"Attribute values must be CSS identifiers [capture 5] or strings [capture 3 or capture 4]\"\n\t\t\"*(?:'((?:\\\\\\\\.|[^\\\\\\\\'])*)'|\\\"((?:\\\\\\\\.|[^\\\\\\\\\\\"])*)\\\"|(\" + identifier + \"))|)\" + whitespace +\n\t\t\"*\\\\]\",\n\n\tpseudos = \":(\" + identifier + \")(?:\\\\((\" +\n\t\t// To reduce the number of selectors needing tokenize in the preFilter, prefer arguments:\n\t\t// 1. quoted (capture 3; capture 4 or capture 5)\n\t\t\"('((?:\\\\\\\\.|[^\\\\\\\\'])*)'|\\\"((?:\\\\\\\\.|[^\\\\\\\\\\\"])*)\\\")|\" +\n\t\t// 2. simple (capture 6)\n\t\t\"((?:\\\\\\\\.|[^\\\\\\\\()[\\\\]]|\" + attributes + \")*)|\" +\n\t\t// 3. anything else (capture 2)\n\t\t\".*\" +\n\t\t\")\\\\)|)\",\n\n\t// Leading and non-escaped trailing whitespace, capturing some non-whitespace characters preceding the latter\n\trwhitespace = new RegExp( whitespace + \"+\", \"g\" ),\n\trtrim = new RegExp( \"^\" + whitespace + \"+|((?:^|[^\\\\\\\\])(?:\\\\\\\\.)*)\" + whitespace + \"+$\", \"g\" ),\n\n\trcomma = new RegExp( \"^\" + whitespace + \"*,\" + whitespace + \"*\" ),\n\trcombinators = new RegExp( \"^\" + whitespace + \"*([>+~]|\" + whitespace + \")\" + whitespace + \"*\" ),\n\n\trattributeQuotes = new RegExp( \"=\" + whitespace + \"*([^\\\\]'\\\"]*?)\" + whitespace + \"*\\\\]\", \"g\" ),\n\n\trpseudo = new RegExp( pseudos ),\n\tridentifier = new RegExp( \"^\" + identifier + \"$\" ),\n\n\tmatchExpr = {\n\t\t\"ID\": new RegExp( \"^#(\" + identifier + \")\" ),\n\t\t\"CLASS\": new RegExp( \"^\\\\.(\" + identifier + \")\" ),\n\t\t\"TAG\": new RegExp( \"^(\" + identifier + \"|[*])\" ),\n\t\t\"ATTR\": new RegExp( \"^\" + attributes ),\n\t\t\"PSEUDO\": new RegExp( \"^\" + pseudos ),\n\t\t\"CHILD\": new RegExp( \"^:(only|first|last|nth|nth-last)-(child|of-type)(?:\\\\(\" + whitespace +\n\t\t\t\"*(even|odd|(([+-]|)(\\\\d*)n|)\" + whitespace + \"*(?:([+-]|)\" + whitespace +\n\t\t\t\"*(\\\\d+)|))\" + whitespace + \"*\\\\)|)\", \"i\" ),\n\t\t\"bool\": new RegExp( \"^(?:\" + booleans + \")$\", \"i\" ),\n\t\t// For use in libraries implementing .is()\n\t\t// We use this for POS matching in `select`\n\t\t\"needsContext\": new RegExp( \"^\" + whitespace + \"*[>+~]|:(even|odd|eq|gt|lt|nth|first|last)(?:\\\\(\" +\n\t\t\twhitespace + \"*((?:-\\\\d)?\\\\d*)\" + whitespace + \"*\\\\)|)(?=[^-]|$)\", \"i\" )\n\t},\n\n\trinputs = /^(?:input|select|textarea|button)$/i,\n\trheader = /^h\\d$/i,\n\n\trnative = /^[^{]+\\{\\s*\\[native \\w/,\n\n\t// Easily-parseable/retrievable ID or TAG or CLASS selectors\n\trquickExpr = /^(?:#([\\w-]+)|(\\w+)|\\.([\\w-]+))$/,\n\n\trsibling = /[+~]/,\n\n\t// CSS escapes\n\t// http://www.w3.org/TR/CSS21/syndata.html#escaped-characters\n\trunescape = new RegExp( \"\\\\\\\\([\\\\da-f]{1,6}\" + whitespace + \"?|(\" + whitespace + \")|.)\", \"ig\" ),\n\tfunescape = function( _, escaped, escapedWhitespace ) {\n\t\tvar high = \"0x\" + escaped - 0x10000;\n\t\t// NaN means non-codepoint\n\t\t// Support: Firefox<24\n\t\t// Workaround erroneous numeric interpretation of +\"0x\"\n\t\treturn high !== high || escapedWhitespace ?\n\t\t\tescaped :\n\t\t\thigh < 0 ?\n\t\t\t\t// BMP codepoint\n\t\t\t\tString.fromCharCode( high + 0x10000 ) :\n\t\t\t\t// Supplemental Plane codepoint (surrogate pair)\n\t\t\t\tString.fromCharCode( high >> 10 | 0xD800, high & 0x3FF | 0xDC00 );\n\t},\n\n\t// CSS string/identifier serialization\n\t// https://drafts.csswg.org/cssom/#common-serializing-idioms\n\trcssescape = /([\\0-\\x1f\\x7f]|^-?\\d)|^-$|[^\\0-\\x1f\\x7f-\\uFFFF\\w-]/g,\n\tfcssescape = function( ch, asCodePoint ) {\n\t\tif ( asCodePoint ) {\n\n\t\t\t// U+0000 NULL becomes U+FFFD REPLACEMENT CHARACTER\n\t\t\tif ( ch === \"\\0\" ) {\n\t\t\t\treturn \"\\uFFFD\";\n\t\t\t}\n\n\t\t\t// Control characters and (dependent upon position) numbers get escaped as code points\n\t\t\treturn ch.slice( 0, -1 ) + \"\\\\\" + ch.charCodeAt( ch.length - 1 ).toString( 16 ) + \" \";\n\t\t}\n\n\t\t// Other potentially-special ASCII characters get backslash-escaped\n\t\treturn \"\\\\\" + ch;\n\t},\n\n\t// Used for iframes\n\t// See setDocument()\n\t// Removing the function wrapper causes a \"Permission Denied\"\n\t// error in IE\n\tunloadHandler = function() {\n\t\tsetDocument();\n\t},\n\n\tdisabledAncestor = addCombinator(\n\t\tfunction( elem ) {\n\t\t\treturn elem.disabled === true && (\"form\" in elem || \"label\" in elem);\n\t\t},\n\t\t{ dir: \"parentNode\", next: \"legend\" }\n\t);\n\n// Optimize for push.apply( _, NodeList )\ntry {\n\tpush.apply(\n\t\t(arr = slice.call( preferredDoc.childNodes )),\n\t\tpreferredDoc.childNodes\n\t);\n\t// Support: Android<4.0\n\t// Detect silently failing push.apply\n\tarr[ preferredDoc.childNodes.length ].nodeType;\n} catch ( e ) {\n\tpush = { apply: arr.length ?\n\n\t\t// Leverage slice if possible\n\t\tfunction( target, els ) {\n\t\t\tpush_native.apply( target, slice.call(els) );\n\t\t} :\n\n\t\t// Support: IE<9\n\t\t// Otherwise append directly\n\t\tfunction( target, els ) {\n\t\t\tvar j = target.length,\n\t\t\t\ti = 0;\n\t\t\t// Can't trust NodeList.length\n\t\t\twhile ( (target[j++] = els[i++]) ) {}\n\t\t\ttarget.length = j - 1;\n\t\t}\n\t};\n}\n\nfunction Sizzle( selector, context, results, seed ) {\n\tvar m, i, elem, nid, match, groups, newSelector,\n\t\tnewContext = context && context.ownerDocument,\n\n\t\t// nodeType defaults to 9, since context defaults to document\n\t\tnodeType = context ? context.nodeType : 9;\n\n\tresults = results || [];\n\n\t// Return early from calls with invalid selector or context\n\tif ( typeof selector !== \"string\" || !selector ||\n\t\tnodeType !== 1 && nodeType !== 9 && nodeType !== 11 ) {\n\n\t\treturn results;\n\t}\n\n\t// Try to shortcut find operations (as opposed to filters) in HTML documents\n\tif ( !seed ) {\n\n\t\tif ( ( context ? context.ownerDocument || context : preferredDoc ) !== document ) {\n\t\t\tsetDocument( context );\n\t\t}\n\t\tcontext = context || document;\n\n\t\tif ( documentIsHTML ) {\n\n\t\t\t// If the selector is sufficiently simple, try using a \"get*By*\" DOM method\n\t\t\t// (excepting DocumentFragment context, where the methods don't exist)\n\t\t\tif ( nodeType !== 11 && (match = rquickExpr.exec( selector )) ) {\n\n\t\t\t\t// ID selector\n\t\t\t\tif ( (m = match[1]) ) {\n\n\t\t\t\t\t// Document context\n\t\t\t\t\tif ( nodeType === 9 ) {\n\t\t\t\t\t\tif ( (elem = context.getElementById( m )) ) {\n\n\t\t\t\t\t\t\t// Support: IE, Opera, Webkit\n\t\t\t\t\t\t\t// TODO: identify versions\n\t\t\t\t\t\t\t// getElementById can match elements by name instead of ID\n\t\t\t\t\t\t\tif ( elem.id === m ) {\n\t\t\t\t\t\t\t\tresults.push( elem );\n\t\t\t\t\t\t\t\treturn results;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\treturn results;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t// Element context\n\t\t\t\t\t} else {\n\n\t\t\t\t\t\t// Support: IE, Opera, Webkit\n\t\t\t\t\t\t// TODO: identify versions\n\t\t\t\t\t\t// getElementById can match elements by name instead of ID\n\t\t\t\t\t\tif ( newContext && (elem = newContext.getElementById( m )) &&\n\t\t\t\t\t\t\tcontains( context, elem ) &&\n\t\t\t\t\t\t\telem.id === m ) {\n\n\t\t\t\t\t\t\tresults.push( elem );\n\t\t\t\t\t\t\treturn results;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t// Type selector\n\t\t\t\t} else if ( match[2] ) {\n\t\t\t\t\tpush.apply( results, context.getElementsByTagName( selector ) );\n\t\t\t\t\treturn results;\n\n\t\t\t\t// Class selector\n\t\t\t\t} else if ( (m = match[3]) && support.getElementsByClassName &&\n\t\t\t\t\tcontext.getElementsByClassName ) {\n\n\t\t\t\t\tpush.apply( results, context.getElementsByClassName( m ) );\n\t\t\t\t\treturn results;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Take advantage of querySelectorAll\n\t\t\tif ( support.qsa &&\n\t\t\t\t!compilerCache[ selector + \" \" ] &&\n\t\t\t\t(!rbuggyQSA || !rbuggyQSA.test( selector )) ) {\n\n\t\t\t\tif ( nodeType !== 1 ) {\n\t\t\t\t\tnewContext = context;\n\t\t\t\t\tnewSelector = selector;\n\n\t\t\t\t// qSA looks outside Element context, which is not what we want\n\t\t\t\t// Thanks to Andrew Dupont for this workaround technique\n\t\t\t\t// Support: IE <=8\n\t\t\t\t// Exclude object elements\n\t\t\t\t} else if ( context.nodeName.toLowerCase() !== \"object\" ) {\n\n\t\t\t\t\t// Capture the context ID, setting it first if necessary\n\t\t\t\t\tif ( (nid = context.getAttribute( \"id\" )) ) {\n\t\t\t\t\t\tnid = nid.replace( rcssescape, fcssescape );\n\t\t\t\t\t} else {\n\t\t\t\t\t\tcontext.setAttribute( \"id\", (nid = expando) );\n\t\t\t\t\t}\n\n\t\t\t\t\t// Prefix every selector in the list\n\t\t\t\t\tgroups = tokenize( selector );\n\t\t\t\t\ti = groups.length;\n\t\t\t\t\twhile ( i-- ) {\n\t\t\t\t\t\tgroups[i] = \"#\" + nid + \" \" + toSelector( groups[i] );\n\t\t\t\t\t}\n\t\t\t\t\tnewSelector = groups.join( \",\" );\n\n\t\t\t\t\t// Expand context for sibling selectors\n\t\t\t\t\tnewContext = rsibling.test( selector ) && testContext( context.parentNode ) ||\n\t\t\t\t\t\tcontext;\n\t\t\t\t}\n\n\t\t\t\tif ( newSelector ) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tpush.apply( results,\n\t\t\t\t\t\t\tnewContext.querySelectorAll( newSelector )\n\t\t\t\t\t\t);\n\t\t\t\t\t\treturn results;\n\t\t\t\t\t} catch ( qsaError ) {\n\t\t\t\t\t} finally {\n\t\t\t\t\t\tif ( nid === expando ) {\n\t\t\t\t\t\t\tcontext.removeAttribute( \"id\" );\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// All others\n\treturn select( selector.replace( rtrim, \"$1\" ), context, results, seed );\n}\n\n/**\n * Create key-value caches of limited size\n * @returns {function(string, object)} Returns the Object data after storing it on itself with\n *\tproperty name the (space-suffixed) string and (if the cache is larger than Expr.cacheLength)\n *\tdeleting the oldest entry\n */\nfunction createCache() {\n\tvar keys = [];\n\n\tfunction cache( key, value ) {\n\t\t// Use (key + \" \") to avoid collision with native prototype properties (see Issue #157)\n\t\tif ( keys.push( key + \" \" ) > Expr.cacheLength ) {\n\t\t\t// Only keep the most recent entries\n\t\t\tdelete cache[ keys.shift() ];\n\t\t}\n\t\treturn (cache[ key + \" \" ] = value);\n\t}\n\treturn cache;\n}\n\n/**\n * Mark a function for special use by Sizzle\n * @param {Function} fn The function to mark\n */\nfunction markFunction( fn ) {\n\tfn[ expando ] = true;\n\treturn fn;\n}\n\n/**\n * Support testing using an element\n * @param {Function} fn Passed the created element and returns a boolean result\n */\nfunction assert( fn ) {\n\tvar el = document.createElement(\"fieldset\");\n\n\ttry {\n\t\treturn !!fn( el );\n\t} catch (e) {\n\t\treturn false;\n\t} finally {\n\t\t// Remove from its parent by default\n\t\tif ( el.parentNode ) {\n\t\t\tel.parentNode.removeChild( el );\n\t\t}\n\t\t// release memory in IE\n\t\tel = null;\n\t}\n}\n\n/**\n * Adds the same handler for all of the specified attrs\n * @param {String} attrs Pipe-separated list of attributes\n * @param {Function} handler The method that will be applied\n */\nfunction addHandle( attrs, handler ) {\n\tvar arr = attrs.split(\"|\"),\n\t\ti = arr.length;\n\n\twhile ( i-- ) {\n\t\tExpr.attrHandle[ arr[i] ] = handler;\n\t}\n}\n\n/**\n * Checks document order of two siblings\n * @param {Element} a\n * @param {Element} b\n * @returns {Number} Returns less than 0 if a precedes b, greater than 0 if a follows b\n */\nfunction siblingCheck( a, b ) {\n\tvar cur = b && a,\n\t\tdiff = cur && a.nodeType === 1 && b.nodeType === 1 &&\n\t\t\ta.sourceIndex - b.sourceIndex;\n\n\t// Use IE sourceIndex if available on both nodes\n\tif ( diff ) {\n\t\treturn diff;\n\t}\n\n\t// Check if b follows a\n\tif ( cur ) {\n\t\twhile ( (cur = cur.nextSibling) ) {\n\t\t\tif ( cur === b ) {\n\t\t\t\treturn -1;\n\t\t\t}\n\t\t}\n\t}\n\n\treturn a ? 1 : -1;\n}\n\n/**\n * Returns a function to use in pseudos for input types\n * @param {String} type\n */\nfunction createInputPseudo( type ) {\n\treturn function( elem ) {\n\t\tvar name = elem.nodeName.toLowerCase();\n\t\treturn name === \"input\" && elem.type === type;\n\t};\n}\n\n/**\n * Returns a function to use in pseudos for buttons\n * @param {String} type\n */\nfunction createButtonPseudo( type ) {\n\treturn function( elem ) {\n\t\tvar name = elem.nodeName.toLowerCase();\n\t\treturn (name === \"input\" || name === \"button\") && elem.type === type;\n\t};\n}\n\n/**\n * Returns a function to use in pseudos for :enabled/:disabled\n * @param {Boolean} disabled true for :disabled; false for :enabled\n */\nfunction createDisabledPseudo( disabled ) {\n\n\t// Known :disabled false positives: fieldset[disabled] > legend:nth-of-type(n+2) :can-disable\n\treturn function( elem ) {\n\n\t\t// Only certain elements can match :enabled or :disabled\n\t\t// https://html.spec.whatwg.org/multipage/scripting.html#selector-enabled\n\t\t// https://html.spec.whatwg.org/multipage/scripting.html#selector-disabled\n\t\tif ( \"form\" in elem ) {\n\n\t\t\t// Check for inherited disabledness on relevant non-disabled elements:\n\t\t\t// * listed form-associated elements in a disabled fieldset\n\t\t\t// https://html.spec.whatwg.org/multipage/forms.html#category-listed\n\t\t\t// https://html.spec.whatwg.org/multipage/forms.html#concept-fe-disabled\n\t\t\t// * option elements in a disabled optgroup\n\t\t\t// https://html.spec.whatwg.org/multipage/forms.html#concept-option-disabled\n\t\t\t// All such elements have a \"form\" property.\n\t\t\tif ( elem.parentNode && elem.disabled === false ) {\n\n\t\t\t\t// Option elements defer to a parent optgroup if present\n\t\t\t\tif ( \"label\" in elem ) {\n\t\t\t\t\tif ( \"label\" in elem.parentNode ) {\n\t\t\t\t\t\treturn elem.parentNode.disabled === disabled;\n\t\t\t\t\t} else {\n\t\t\t\t\t\treturn elem.disabled === disabled;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Support: IE 6 - 11\n\t\t\t\t// Use the isDisabled shortcut property to check for disabled fieldset ancestors\n\t\t\t\treturn elem.isDisabled === disabled ||\n\n\t\t\t\t\t// Where there is no isDisabled, check manually\n\t\t\t\t\t/* jshint -W018 */\n\t\t\t\t\telem.isDisabled !== !disabled &&\n\t\t\t\t\t\tdisabledAncestor( elem ) === disabled;\n\t\t\t}\n\n\t\t\treturn elem.disabled === disabled;\n\n\t\t// Try to winnow out elements that can't be disabled before trusting the disabled property.\n\t\t// Some victims get caught in our net (label, legend, menu, track), but it shouldn't\n\t\t// even exist on them, let alone have a boolean value.\n\t\t} else if ( \"label\" in elem ) {\n\t\t\treturn elem.disabled === disabled;\n\t\t}\n\n\t\t// Remaining elements are neither :enabled nor :disabled\n\t\treturn false;\n\t};\n}\n\n/**\n * Returns a function to use in pseudos for positionals\n * @param {Function} fn\n */\nfunction createPositionalPseudo( fn ) {\n\treturn markFunction(function( argument ) {\n\t\targument = +argument;\n\t\treturn markFunction(function( seed, matches ) {\n\t\t\tvar j,\n\t\t\t\tmatchIndexes = fn( [], seed.length, argument ),\n\t\t\t\ti = matchIndexes.length;\n\n\t\t\t// Match elements found at the specified indexes\n\t\t\twhile ( i-- ) {\n\t\t\t\tif ( seed[ (j = matchIndexes[i]) ] ) {\n\t\t\t\t\tseed[j] = !(matches[j] = seed[j]);\n\t\t\t\t}\n\t\t\t}\n\t\t});\n\t});\n}\n\n/**\n * Checks a node for validity as a Sizzle context\n * @param {Element|Object=} context\n * @returns {Element|Object|Boolean} The input node if acceptable, otherwise a falsy value\n */\nfunction testContext( context ) {\n\treturn context && typeof context.getElementsByTagName !== \"undefined\" && context;\n}\n\n// Expose support vars for convenience\nsupport = Sizzle.support = {};\n\n/**\n * Detects XML nodes\n * @param {Element|Object} elem An element or a document\n * @returns {Boolean} True iff elem is a non-HTML XML node\n */\nisXML = Sizzle.isXML = function( elem ) {\n\t// documentElement is verified for cases where it doesn't yet exist\n\t// (such as loading iframes in IE - #4833)\n\tvar documentElement = elem && (elem.ownerDocument || elem).documentElement;\n\treturn documentElement ? documentElement.nodeName !== \"HTML\" : false;\n};\n\n/**\n * Sets document-related variables once based on the current document\n * @param {Element|Object} [doc] An element or document object to use to set the document\n * @returns {Object} Returns the current document\n */\nsetDocument = Sizzle.setDocument = function( node ) {\n\tvar hasCompare, subWindow,\n\t\tdoc = node ? node.ownerDocument || node : preferredDoc;\n\n\t// Return early if doc is invalid or already selected\n\tif ( doc === document || doc.nodeType !== 9 || !doc.documentElement ) {\n\t\treturn document;\n\t}\n\n\t// Update global variables\n\tdocument = doc;\n\tdocElem = document.documentElement;\n\tdocumentIsHTML = !isXML( document );\n\n\t// Support: IE 9-11, Edge\n\t// Accessing iframe documents after unload throws \"permission denied\" errors (jQuery #13936)\n\tif ( preferredDoc !== document &&\n\t\t(subWindow = document.defaultView) && subWindow.top !== subWindow ) {\n\n\t\t// Support: IE 11, Edge\n\t\tif ( subWindow.addEventListener ) {\n\t\t\tsubWindow.addEventListener( \"unload\", unloadHandler, false );\n\n\t\t// Support: IE 9 - 10 only\n\t\t} else if ( subWindow.attachEvent ) {\n\t\t\tsubWindow.attachEvent( \"onunload\", unloadHandler );\n\t\t}\n\t}\n\n\t/* Attributes\n\t---------------------------------------------------------------------- */\n\n\t// Support: IE<8\n\t// Verify that getAttribute really returns attributes and not properties\n\t// (excepting IE8 booleans)\n\tsupport.attributes = assert(function( el ) {\n\t\tel.className = \"i\";\n\t\treturn !el.getAttribute(\"className\");\n\t});\n\n\t/* getElement(s)By*\n\t---------------------------------------------------------------------- */\n\n\t// Check if getElementsByTagName(\"*\") returns only elements\n\tsupport.getElementsByTagName = assert(function( el ) {\n\t\tel.appendChild( document.createComment(\"\") );\n\t\treturn !el.getElementsByTagName(\"*\").length;\n\t});\n\n\t// Support: IE<9\n\tsupport.getElementsByClassName = rnative.test( document.getElementsByClassName );\n\n\t// Support: IE<10\n\t// Check if getElementById returns elements by name\n\t// The broken getElementById methods don't pick up programmatically-set names,\n\t// so use a roundabout getElementsByName test\n\tsupport.getById = assert(function( el ) {\n\t\tdocElem.appendChild( el ).id = expando;\n\t\treturn !document.getElementsByName || !document.getElementsByName( expando ).length;\n\t});\n\n\t// ID filter and find\n\tif ( support.getById ) {\n\t\tExpr.filter[\"ID\"] = function( id ) {\n\t\t\tvar attrId = id.replace( runescape, funescape );\n\t\t\treturn function( elem ) {\n\t\t\t\treturn elem.getAttribute(\"id\") === attrId;\n\t\t\t};\n\t\t};\n\t\tExpr.find[\"ID\"] = function( id, context ) {\n\t\t\tif ( typeof context.getElementById !== \"undefined\" && documentIsHTML ) {\n\t\t\t\tvar elem = context.getElementById( id );\n\t\t\t\treturn elem ? [ elem ] : [];\n\t\t\t}\n\t\t};\n\t} else {\n\t\tExpr.filter[\"ID\"] = function( id ) {\n\t\t\tvar attrId = id.replace( runescape, funescape );\n\t\t\treturn function( elem ) {\n\t\t\t\tvar node = typeof elem.getAttributeNode !== \"undefined\" &&\n\t\t\t\t\telem.getAttributeNode(\"id\");\n\t\t\t\treturn node && node.value === attrId;\n\t\t\t};\n\t\t};\n\n\t\t// Support: IE 6 - 7 only\n\t\t// getElementById is not reliable as a find shortcut\n\t\tExpr.find[\"ID\"] = function( id, context ) {\n\t\t\tif ( typeof context.getElementById !== \"undefined\" && documentIsHTML ) {\n\t\t\t\tvar node, i, elems,\n\t\t\t\t\telem = context.getElementById( id );\n\n\t\t\t\tif ( elem ) {\n\n\t\t\t\t\t// Verify the id attribute\n\t\t\t\t\tnode = elem.getAttributeNode(\"id\");\n\t\t\t\t\tif ( node && node.value === id ) {\n\t\t\t\t\t\treturn [ elem ];\n\t\t\t\t\t}\n\n\t\t\t\t\t// Fall back on getElementsByName\n\t\t\t\t\telems = context.getElementsByName( id );\n\t\t\t\t\ti = 0;\n\t\t\t\t\twhile ( (elem = elems[i++]) ) {\n\t\t\t\t\t\tnode = elem.getAttributeNode(\"id\");\n\t\t\t\t\t\tif ( node && node.value === id ) {\n\t\t\t\t\t\t\treturn [ elem ];\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\treturn [];\n\t\t\t}\n\t\t};\n\t}\n\n\t// Tag\n\tExpr.find[\"TAG\"] = support.getElementsByTagName ?\n\t\tfunction( tag, context ) {\n\t\t\tif ( typeof context.getElementsByTagName !== \"undefined\" ) {\n\t\t\t\treturn context.getElementsByTagName( tag );\n\n\t\t\t// DocumentFragment nodes don't have gEBTN\n\t\t\t} else if ( support.qsa ) {\n\t\t\t\treturn context.querySelectorAll( tag );\n\t\t\t}\n\t\t} :\n\n\t\tfunction( tag, context ) {\n\t\t\tvar elem,\n\t\t\t\ttmp = [],\n\t\t\t\ti = 0,\n\t\t\t\t// By happy coincidence, a (broken) gEBTN appears on DocumentFragment nodes too\n\t\t\t\tresults = context.getElementsByTagName( tag );\n\n\t\t\t// Filter out possible comments\n\t\t\tif ( tag === \"*\" ) {\n\t\t\t\twhile ( (elem = results[i++]) ) {\n\t\t\t\t\tif ( elem.nodeType === 1 ) {\n\t\t\t\t\t\ttmp.push( elem );\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\treturn tmp;\n\t\t\t}\n\t\t\treturn results;\n\t\t};\n\n\t// Class\n\tExpr.find[\"CLASS\"] = support.getElementsByClassName && function( className, context ) {\n\t\tif ( typeof context.getElementsByClassName !== \"undefined\" && documentIsHTML ) {\n\t\t\treturn context.getElementsByClassName( className );\n\t\t}\n\t};\n\n\t/* QSA/matchesSelector\n\t---------------------------------------------------------------------- */\n\n\t// QSA and matchesSelector support\n\n\t// matchesSelector(:active) reports false when true (IE9/Opera 11.5)\n\trbuggyMatches = [];\n\n\t// qSa(:focus) reports false when true (Chrome 21)\n\t// We allow this because of a bug in IE8/9 that throws an error\n\t// whenever `document.activeElement` is accessed on an iframe\n\t// So, we allow :focus to pass through QSA all the time to avoid the IE error\n\t// See https://bugs.jquery.com/ticket/13378\n\trbuggyQSA = [];\n\n\tif ( (support.qsa = rnative.test( document.querySelectorAll )) ) {\n\t\t// Build QSA regex\n\t\t// Regex strategy adopted from Diego Perini\n\t\tassert(function( el ) {\n\t\t\t// Select is set to empty string on purpose\n\t\t\t// This is to test IE's treatment of not explicitly\n\t\t\t// setting a boolean content attribute,\n\t\t\t// since its presence should be enough\n\t\t\t// https://bugs.jquery.com/ticket/12359\n\t\t\tdocElem.appendChild( el ).innerHTML = \"\" +\n\t\t\t\t\"\";\n\n\t\t\t// Support: IE8, Opera 11-12.16\n\t\t\t// Nothing should be selected when empty strings follow ^= or $= or *=\n\t\t\t// The test attribute must be unknown in Opera but \"safe\" for WinRT\n\t\t\t// https://msdn.microsoft.com/en-us/library/ie/hh465388.aspx#attribute_section\n\t\t\tif ( el.querySelectorAll(\"[msallowcapture^='']\").length ) {\n\t\t\t\trbuggyQSA.push( \"[*^$]=\" + whitespace + \"*(?:''|\\\"\\\")\" );\n\t\t\t}\n\n\t\t\t// Support: IE8\n\t\t\t// Boolean attributes and \"value\" are not treated correctly\n\t\t\tif ( !el.querySelectorAll(\"[selected]\").length ) {\n\t\t\t\trbuggyQSA.push( \"\\\\[\" + whitespace + \"*(?:value|\" + booleans + \")\" );\n\t\t\t}\n\n\t\t\t// Support: Chrome<29, Android<4.4, Safari<7.0+, iOS<7.0+, PhantomJS<1.9.8+\n\t\t\tif ( !el.querySelectorAll( \"[id~=\" + expando + \"-]\" ).length ) {\n\t\t\t\trbuggyQSA.push(\"~=\");\n\t\t\t}\n\n\t\t\t// Webkit/Opera - :checked should return selected option elements\n\t\t\t// http://www.w3.org/TR/2011/REC-css3-selectors-20110929/#checked\n\t\t\t// IE8 throws error here and will not see later tests\n\t\t\tif ( !el.querySelectorAll(\":checked\").length ) {\n\t\t\t\trbuggyQSA.push(\":checked\");\n\t\t\t}\n\n\t\t\t// Support: Safari 8+, iOS 8+\n\t\t\t// https://bugs.webkit.org/show_bug.cgi?id=136851\n\t\t\t// In-page `selector#id sibling-combinator selector` fails\n\t\t\tif ( !el.querySelectorAll( \"a#\" + expando + \"+*\" ).length ) {\n\t\t\t\trbuggyQSA.push(\".#.+[+~]\");\n\t\t\t}\n\t\t});\n\n\t\tassert(function( el ) {\n\t\t\tel.innerHTML = \"\" +\n\t\t\t\t\"\";\n\n\t\t\t// Support: Windows 8 Native Apps\n\t\t\t// The type and name attributes are restricted during .innerHTML assignment\n\t\t\tvar input = document.createElement(\"input\");\n\t\t\tinput.setAttribute( \"type\", \"hidden\" );\n\t\t\tel.appendChild( input ).setAttribute( \"name\", \"D\" );\n\n\t\t\t// Support: IE8\n\t\t\t// Enforce case-sensitivity of name attribute\n\t\t\tif ( el.querySelectorAll(\"[name=d]\").length ) {\n\t\t\t\trbuggyQSA.push( \"name\" + whitespace + \"*[*^$|!~]?=\" );\n\t\t\t}\n\n\t\t\t// FF 3.5 - :enabled/:disabled and hidden elements (hidden elements are still enabled)\n\t\t\t// IE8 throws error here and will not see later tests\n\t\t\tif ( el.querySelectorAll(\":enabled\").length !== 2 ) {\n\t\t\t\trbuggyQSA.push( \":enabled\", \":disabled\" );\n\t\t\t}\n\n\t\t\t// Support: IE9-11+\n\t\t\t// IE's :disabled selector does not pick up the children of disabled fieldsets\n\t\t\tdocElem.appendChild( el ).disabled = true;\n\t\t\tif ( el.querySelectorAll(\":disabled\").length !== 2 ) {\n\t\t\t\trbuggyQSA.push( \":enabled\", \":disabled\" );\n\t\t\t}\n\n\t\t\t// Opera 10-11 does not throw on post-comma invalid pseudos\n\t\t\tel.querySelectorAll(\"*,:x\");\n\t\t\trbuggyQSA.push(\",.*:\");\n\t\t});\n\t}\n\n\tif ( (support.matchesSelector = rnative.test( (matches = docElem.matches ||\n\t\tdocElem.webkitMatchesSelector ||\n\t\tdocElem.mozMatchesSelector ||\n\t\tdocElem.oMatchesSelector ||\n\t\tdocElem.msMatchesSelector) )) ) {\n\n\t\tassert(function( el ) {\n\t\t\t// Check to see if it's possible to do matchesSelector\n\t\t\t// on a disconnected node (IE 9)\n\t\t\tsupport.disconnectedMatch = matches.call( el, \"*\" );\n\n\t\t\t// This should fail with an exception\n\t\t\t// Gecko does not error, returns false instead\n\t\t\tmatches.call( el, \"[s!='']:x\" );\n\t\t\trbuggyMatches.push( \"!=\", pseudos );\n\t\t});\n\t}\n\n\trbuggyQSA = rbuggyQSA.length && new RegExp( rbuggyQSA.join(\"|\") );\n\trbuggyMatches = rbuggyMatches.length && new RegExp( rbuggyMatches.join(\"|\") );\n\n\t/* Contains\n\t---------------------------------------------------------------------- */\n\thasCompare = rnative.test( docElem.compareDocumentPosition );\n\n\t// Element contains another\n\t// Purposefully self-exclusive\n\t// As in, an element does not contain itself\n\tcontains = hasCompare || rnative.test( docElem.contains ) ?\n\t\tfunction( a, b ) {\n\t\t\tvar adown = a.nodeType === 9 ? a.documentElement : a,\n\t\t\t\tbup = b && b.parentNode;\n\t\t\treturn a === bup || !!( bup && bup.nodeType === 1 && (\n\t\t\t\tadown.contains ?\n\t\t\t\t\tadown.contains( bup ) :\n\t\t\t\t\ta.compareDocumentPosition && a.compareDocumentPosition( bup ) & 16\n\t\t\t));\n\t\t} :\n\t\tfunction( a, b ) {\n\t\t\tif ( b ) {\n\t\t\t\twhile ( (b = b.parentNode) ) {\n\t\t\t\t\tif ( b === a ) {\n\t\t\t\t\t\treturn true;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn false;\n\t\t};\n\n\t/* Sorting\n\t---------------------------------------------------------------------- */\n\n\t// Document order sorting\n\tsortOrder = hasCompare ?\n\tfunction( a, b ) {\n\n\t\t// Flag for duplicate removal\n\t\tif ( a === b ) {\n\t\t\thasDuplicate = true;\n\t\t\treturn 0;\n\t\t}\n\n\t\t// Sort on method existence if only one input has compareDocumentPosition\n\t\tvar compare = !a.compareDocumentPosition - !b.compareDocumentPosition;\n\t\tif ( compare ) {\n\t\t\treturn compare;\n\t\t}\n\n\t\t// Calculate position if both inputs belong to the same document\n\t\tcompare = ( a.ownerDocument || a ) === ( b.ownerDocument || b ) ?\n\t\t\ta.compareDocumentPosition( b ) :\n\n\t\t\t// Otherwise we know they are disconnected\n\t\t\t1;\n\n\t\t// Disconnected nodes\n\t\tif ( compare & 1 ||\n\t\t\t(!support.sortDetached && b.compareDocumentPosition( a ) === compare) ) {\n\n\t\t\t// Choose the first element that is related to our preferred document\n\t\t\tif ( a === document || a.ownerDocument === preferredDoc && contains(preferredDoc, a) ) {\n\t\t\t\treturn -1;\n\t\t\t}\n\t\t\tif ( b === document || b.ownerDocument === preferredDoc && contains(preferredDoc, b) ) {\n\t\t\t\treturn 1;\n\t\t\t}\n\n\t\t\t// Maintain original order\n\t\t\treturn sortInput ?\n\t\t\t\t( indexOf( sortInput, a ) - indexOf( sortInput, b ) ) :\n\t\t\t\t0;\n\t\t}\n\n\t\treturn compare & 4 ? -1 : 1;\n\t} :\n\tfunction( a, b ) {\n\t\t// Exit early if the nodes are identical\n\t\tif ( a === b ) {\n\t\t\thasDuplicate = true;\n\t\t\treturn 0;\n\t\t}\n\n\t\tvar cur,\n\t\t\ti = 0,\n\t\t\taup = a.parentNode,\n\t\t\tbup = b.parentNode,\n\t\t\tap = [ a ],\n\t\t\tbp = [ b ];\n\n\t\t// Parentless nodes are either documents or disconnected\n\t\tif ( !aup || !bup ) {\n\t\t\treturn a === document ? -1 :\n\t\t\t\tb === document ? 1 :\n\t\t\t\taup ? -1 :\n\t\t\t\tbup ? 1 :\n\t\t\t\tsortInput ?\n\t\t\t\t( indexOf( sortInput, a ) - indexOf( sortInput, b ) ) :\n\t\t\t\t0;\n\n\t\t// If the nodes are siblings, we can do a quick check\n\t\t} else if ( aup === bup ) {\n\t\t\treturn siblingCheck( a, b );\n\t\t}\n\n\t\t// Otherwise we need full lists of their ancestors for comparison\n\t\tcur = a;\n\t\twhile ( (cur = cur.parentNode) ) {\n\t\t\tap.unshift( cur );\n\t\t}\n\t\tcur = b;\n\t\twhile ( (cur = cur.parentNode) ) {\n\t\t\tbp.unshift( cur );\n\t\t}\n\n\t\t// Walk down the tree looking for a discrepancy\n\t\twhile ( ap[i] === bp[i] ) {\n\t\t\ti++;\n\t\t}\n\n\t\treturn i ?\n\t\t\t// Do a sibling check if the nodes have a common ancestor\n\t\t\tsiblingCheck( ap[i], bp[i] ) :\n\n\t\t\t// Otherwise nodes in our document sort first\n\t\t\tap[i] === preferredDoc ? -1 :\n\t\t\tbp[i] === preferredDoc ? 1 :\n\t\t\t0;\n\t};\n\n\treturn document;\n};\n\nSizzle.matches = function( expr, elements ) {\n\treturn Sizzle( expr, null, null, elements );\n};\n\nSizzle.matchesSelector = function( elem, expr ) {\n\t// Set document vars if needed\n\tif ( ( elem.ownerDocument || elem ) !== document ) {\n\t\tsetDocument( elem );\n\t}\n\n\t// Make sure that attribute selectors are quoted\n\texpr = expr.replace( rattributeQuotes, \"='$1']\" );\n\n\tif ( support.matchesSelector && documentIsHTML &&\n\t\t!compilerCache[ expr + \" \" ] &&\n\t\t( !rbuggyMatches || !rbuggyMatches.test( expr ) ) &&\n\t\t( !rbuggyQSA || !rbuggyQSA.test( expr ) ) ) {\n\n\t\ttry {\n\t\t\tvar ret = matches.call( elem, expr );\n\n\t\t\t// IE 9's matchesSelector returns false on disconnected nodes\n\t\t\tif ( ret || support.disconnectedMatch ||\n\t\t\t\t\t// As well, disconnected nodes are said to be in a document\n\t\t\t\t\t// fragment in IE 9\n\t\t\t\t\telem.document && elem.document.nodeType !== 11 ) {\n\t\t\t\treturn ret;\n\t\t\t}\n\t\t} catch (e) {}\n\t}\n\n\treturn Sizzle( expr, document, null, [ elem ] ).length > 0;\n};\n\nSizzle.contains = function( context, elem ) {\n\t// Set document vars if needed\n\tif ( ( context.ownerDocument || context ) !== document ) {\n\t\tsetDocument( context );\n\t}\n\treturn contains( context, elem );\n};\n\nSizzle.attr = function( elem, name ) {\n\t// Set document vars if needed\n\tif ( ( elem.ownerDocument || elem ) !== document ) {\n\t\tsetDocument( elem );\n\t}\n\n\tvar fn = Expr.attrHandle[ name.toLowerCase() ],\n\t\t// Don't get fooled by Object.prototype properties (jQuery #13807)\n\t\tval = fn && hasOwn.call( Expr.attrHandle, name.toLowerCase() ) ?\n\t\t\tfn( elem, name, !documentIsHTML ) :\n\t\t\tundefined;\n\n\treturn val !== undefined ?\n\t\tval :\n\t\tsupport.attributes || !documentIsHTML ?\n\t\t\telem.getAttribute( name ) :\n\t\t\t(val = elem.getAttributeNode(name)) && val.specified ?\n\t\t\t\tval.value :\n\t\t\t\tnull;\n};\n\nSizzle.escape = function( sel ) {\n\treturn (sel + \"\").replace( rcssescape, fcssescape );\n};\n\nSizzle.error = function( msg ) {\n\tthrow new Error( \"Syntax error, unrecognized expression: \" + msg );\n};\n\n/**\n * Document sorting and removing duplicates\n * @param {ArrayLike} results\n */\nSizzle.uniqueSort = function( results ) {\n\tvar elem,\n\t\tduplicates = [],\n\t\tj = 0,\n\t\ti = 0;\n\n\t// Unless we *know* we can detect duplicates, assume their presence\n\thasDuplicate = !support.detectDuplicates;\n\tsortInput = !support.sortStable && results.slice( 0 );\n\tresults.sort( sortOrder );\n\n\tif ( hasDuplicate ) {\n\t\twhile ( (elem = results[i++]) ) {\n\t\t\tif ( elem === results[ i ] ) {\n\t\t\t\tj = duplicates.push( i );\n\t\t\t}\n\t\t}\n\t\twhile ( j-- ) {\n\t\t\tresults.splice( duplicates[ j ], 1 );\n\t\t}\n\t}\n\n\t// Clear input after sorting to release objects\n\t// See https://github.com/jquery/sizzle/pull/225\n\tsortInput = null;\n\n\treturn results;\n};\n\n/**\n * Utility function for retrieving the text value of an array of DOM nodes\n * @param {Array|Element} elem\n */\ngetText = Sizzle.getText = function( elem ) {\n\tvar node,\n\t\tret = \"\",\n\t\ti = 0,\n\t\tnodeType = elem.nodeType;\n\n\tif ( !nodeType ) {\n\t\t// If no nodeType, this is expected to be an array\n\t\twhile ( (node = elem[i++]) ) {\n\t\t\t// Do not traverse comment nodes\n\t\t\tret += getText( node );\n\t\t}\n\t} else if ( nodeType === 1 || nodeType === 9 || nodeType === 11 ) {\n\t\t// Use textContent for elements\n\t\t// innerText usage removed for consistency of new lines (jQuery #11153)\n\t\tif ( typeof elem.textContent === \"string\" ) {\n\t\t\treturn elem.textContent;\n\t\t} else {\n\t\t\t// Traverse its children\n\t\t\tfor ( elem = elem.firstChild; elem; elem = elem.nextSibling ) {\n\t\t\t\tret += getText( elem );\n\t\t\t}\n\t\t}\n\t} else if ( nodeType === 3 || nodeType === 4 ) {\n\t\treturn elem.nodeValue;\n\t}\n\t// Do not include comment or processing instruction nodes\n\n\treturn ret;\n};\n\nExpr = Sizzle.selectors = {\n\n\t// Can be adjusted by the user\n\tcacheLength: 50,\n\n\tcreatePseudo: markFunction,\n\n\tmatch: matchExpr,\n\n\tattrHandle: {},\n\n\tfind: {},\n\n\trelative: {\n\t\t\">\": { dir: \"parentNode\", first: true },\n\t\t\" \": { dir: \"parentNode\" },\n\t\t\"+\": { dir: \"previousSibling\", first: true },\n\t\t\"~\": { dir: \"previousSibling\" }\n\t},\n\n\tpreFilter: {\n\t\t\"ATTR\": function( match ) {\n\t\t\tmatch[1] = match[1].replace( runescape, funescape );\n\n\t\t\t// Move the given value to match[3] whether quoted or unquoted\n\t\t\tmatch[3] = ( match[3] || match[4] || match[5] || \"\" ).replace( runescape, funescape );\n\n\t\t\tif ( match[2] === \"~=\" ) {\n\t\t\t\tmatch[3] = \" \" + match[3] + \" \";\n\t\t\t}\n\n\t\t\treturn match.slice( 0, 4 );\n\t\t},\n\n\t\t\"CHILD\": function( match ) {\n\t\t\t/* matches from matchExpr[\"CHILD\"]\n\t\t\t\t1 type (only|nth|...)\n\t\t\t\t2 what (child|of-type)\n\t\t\t\t3 argument (even|odd|\\d*|\\d*n([+-]\\d+)?|...)\n\t\t\t\t4 xn-component of xn+y argument ([+-]?\\d*n|)\n\t\t\t\t5 sign of xn-component\n\t\t\t\t6 x of xn-component\n\t\t\t\t7 sign of y-component\n\t\t\t\t8 y of y-component\n\t\t\t*/\n\t\t\tmatch[1] = match[1].toLowerCase();\n\n\t\t\tif ( match[1].slice( 0, 3 ) === \"nth\" ) {\n\t\t\t\t// nth-* requires argument\n\t\t\t\tif ( !match[3] ) {\n\t\t\t\t\tSizzle.error( match[0] );\n\t\t\t\t}\n\n\t\t\t\t// numeric x and y parameters for Expr.filter.CHILD\n\t\t\t\t// remember that false/true cast respectively to 0/1\n\t\t\t\tmatch[4] = +( match[4] ? match[5] + (match[6] || 1) : 2 * ( match[3] === \"even\" || match[3] === \"odd\" ) );\n\t\t\t\tmatch[5] = +( ( match[7] + match[8] ) || match[3] === \"odd\" );\n\n\t\t\t// other types prohibit arguments\n\t\t\t} else if ( match[3] ) {\n\t\t\t\tSizzle.error( match[0] );\n\t\t\t}\n\n\t\t\treturn match;\n\t\t},\n\n\t\t\"PSEUDO\": function( match ) {\n\t\t\tvar excess,\n\t\t\t\tunquoted = !match[6] && match[2];\n\n\t\t\tif ( matchExpr[\"CHILD\"].test( match[0] ) ) {\n\t\t\t\treturn null;\n\t\t\t}\n\n\t\t\t// Accept quoted arguments as-is\n\t\t\tif ( match[3] ) {\n\t\t\t\tmatch[2] = match[4] || match[5] || \"\";\n\n\t\t\t// Strip excess characters from unquoted arguments\n\t\t\t} else if ( unquoted && rpseudo.test( unquoted ) &&\n\t\t\t\t// Get excess from tokenize (recursively)\n\t\t\t\t(excess = tokenize( unquoted, true )) &&\n\t\t\t\t// advance to the next closing parenthesis\n\t\t\t\t(excess = unquoted.indexOf( \")\", unquoted.length - excess ) - unquoted.length) ) {\n\n\t\t\t\t// excess is a negative index\n\t\t\t\tmatch[0] = match[0].slice( 0, excess );\n\t\t\t\tmatch[2] = unquoted.slice( 0, excess );\n\t\t\t}\n\n\t\t\t// Return only captures needed by the pseudo filter method (type and argument)\n\t\t\treturn match.slice( 0, 3 );\n\t\t}\n\t},\n\n\tfilter: {\n\n\t\t\"TAG\": function( nodeNameSelector ) {\n\t\t\tvar nodeName = nodeNameSelector.replace( runescape, funescape ).toLowerCase();\n\t\t\treturn nodeNameSelector === \"*\" ?\n\t\t\t\tfunction() { return true; } :\n\t\t\t\tfunction( elem ) {\n\t\t\t\t\treturn elem.nodeName && elem.nodeName.toLowerCase() === nodeName;\n\t\t\t\t};\n\t\t},\n\n\t\t\"CLASS\": function( className ) {\n\t\t\tvar pattern = classCache[ className + \" \" ];\n\n\t\t\treturn pattern ||\n\t\t\t\t(pattern = new RegExp( \"(^|\" + whitespace + \")\" + className + \"(\" + whitespace + \"|$)\" )) &&\n\t\t\t\tclassCache( className, function( elem ) {\n\t\t\t\t\treturn pattern.test( typeof elem.className === \"string\" && elem.className || typeof elem.getAttribute !== \"undefined\" && elem.getAttribute(\"class\") || \"\" );\n\t\t\t\t});\n\t\t},\n\n\t\t\"ATTR\": function( name, operator, check ) {\n\t\t\treturn function( elem ) {\n\t\t\t\tvar result = Sizzle.attr( elem, name );\n\n\t\t\t\tif ( result == null ) {\n\t\t\t\t\treturn operator === \"!=\";\n\t\t\t\t}\n\t\t\t\tif ( !operator ) {\n\t\t\t\t\treturn true;\n\t\t\t\t}\n\n\t\t\t\tresult += \"\";\n\n\t\t\t\treturn operator === \"=\" ? result === check :\n\t\t\t\t\toperator === \"!=\" ? result !== check :\n\t\t\t\t\toperator === \"^=\" ? check && result.indexOf( check ) === 0 :\n\t\t\t\t\toperator === \"*=\" ? check && result.indexOf( check ) > -1 :\n\t\t\t\t\toperator === \"$=\" ? check && result.slice( -check.length ) === check :\n\t\t\t\t\toperator === \"~=\" ? ( \" \" + result.replace( rwhitespace, \" \" ) + \" \" ).indexOf( check ) > -1 :\n\t\t\t\t\toperator === \"|=\" ? result === check || result.slice( 0, check.length + 1 ) === check + \"-\" :\n\t\t\t\t\tfalse;\n\t\t\t};\n\t\t},\n\n\t\t\"CHILD\": function( type, what, argument, first, last ) {\n\t\t\tvar simple = type.slice( 0, 3 ) !== \"nth\",\n\t\t\t\tforward = type.slice( -4 ) !== \"last\",\n\t\t\t\tofType = what === \"of-type\";\n\n\t\t\treturn first === 1 && last === 0 ?\n\n\t\t\t\t// Shortcut for :nth-*(n)\n\t\t\t\tfunction( elem ) {\n\t\t\t\t\treturn !!elem.parentNode;\n\t\t\t\t} :\n\n\t\t\t\tfunction( elem, context, xml ) {\n\t\t\t\t\tvar cache, uniqueCache, outerCache, node, nodeIndex, start,\n\t\t\t\t\t\tdir = simple !== forward ? \"nextSibling\" : \"previousSibling\",\n\t\t\t\t\t\tparent = elem.parentNode,\n\t\t\t\t\t\tname = ofType && elem.nodeName.toLowerCase(),\n\t\t\t\t\t\tuseCache = !xml && !ofType,\n\t\t\t\t\t\tdiff = false;\n\n\t\t\t\t\tif ( parent ) {\n\n\t\t\t\t\t\t// :(first|last|only)-(child|of-type)\n\t\t\t\t\t\tif ( simple ) {\n\t\t\t\t\t\t\twhile ( dir ) {\n\t\t\t\t\t\t\t\tnode = elem;\n\t\t\t\t\t\t\t\twhile ( (node = node[ dir ]) ) {\n\t\t\t\t\t\t\t\t\tif ( ofType ?\n\t\t\t\t\t\t\t\t\t\tnode.nodeName.toLowerCase() === name :\n\t\t\t\t\t\t\t\t\t\tnode.nodeType === 1 ) {\n\n\t\t\t\t\t\t\t\t\t\treturn false;\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t// Reverse direction for :only-* (if we haven't yet done so)\n\t\t\t\t\t\t\t\tstart = dir = type === \"only\" && !start && \"nextSibling\";\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\treturn true;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tstart = [ forward ? parent.firstChild : parent.lastChild ];\n\n\t\t\t\t\t\t// non-xml :nth-child(...) stores cache data on `parent`\n\t\t\t\t\t\tif ( forward && useCache ) {\n\n\t\t\t\t\t\t\t// Seek `elem` from a previously-cached index\n\n\t\t\t\t\t\t\t// ...in a gzip-friendly way\n\t\t\t\t\t\t\tnode = parent;\n\t\t\t\t\t\t\touterCache = node[ expando ] || (node[ expando ] = {});\n\n\t\t\t\t\t\t\t// Support: IE <9 only\n\t\t\t\t\t\t\t// Defend against cloned attroperties (jQuery gh-1709)\n\t\t\t\t\t\t\tuniqueCache = outerCache[ node.uniqueID ] ||\n\t\t\t\t\t\t\t\t(outerCache[ node.uniqueID ] = {});\n\n\t\t\t\t\t\t\tcache = uniqueCache[ type ] || [];\n\t\t\t\t\t\t\tnodeIndex = cache[ 0 ] === dirruns && cache[ 1 ];\n\t\t\t\t\t\t\tdiff = nodeIndex && cache[ 2 ];\n\t\t\t\t\t\t\tnode = nodeIndex && parent.childNodes[ nodeIndex ];\n\n\t\t\t\t\t\t\twhile ( (node = ++nodeIndex && node && node[ dir ] ||\n\n\t\t\t\t\t\t\t\t// Fallback to seeking `elem` from the start\n\t\t\t\t\t\t\t\t(diff = nodeIndex = 0) || start.pop()) ) {\n\n\t\t\t\t\t\t\t\t// When found, cache indexes on `parent` and break\n\t\t\t\t\t\t\t\tif ( node.nodeType === 1 && ++diff && node === elem ) {\n\t\t\t\t\t\t\t\t\tuniqueCache[ type ] = [ dirruns, nodeIndex, diff ];\n\t\t\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t// Use previously-cached element index if available\n\t\t\t\t\t\t\tif ( useCache ) {\n\t\t\t\t\t\t\t\t// ...in a gzip-friendly way\n\t\t\t\t\t\t\t\tnode = elem;\n\t\t\t\t\t\t\t\touterCache = node[ expando ] || (node[ expando ] = {});\n\n\t\t\t\t\t\t\t\t// Support: IE <9 only\n\t\t\t\t\t\t\t\t// Defend against cloned attroperties (jQuery gh-1709)\n\t\t\t\t\t\t\t\tuniqueCache = outerCache[ node.uniqueID ] ||\n\t\t\t\t\t\t\t\t\t(outerCache[ node.uniqueID ] = {});\n\n\t\t\t\t\t\t\t\tcache = uniqueCache[ type ] || [];\n\t\t\t\t\t\t\t\tnodeIndex = cache[ 0 ] === dirruns && cache[ 1 ];\n\t\t\t\t\t\t\t\tdiff = nodeIndex;\n\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t// xml :nth-child(...)\n\t\t\t\t\t\t\t// or :nth-last-child(...) or :nth(-last)?-of-type(...)\n\t\t\t\t\t\t\tif ( diff === false ) {\n\t\t\t\t\t\t\t\t// Use the same loop as above to seek `elem` from the start\n\t\t\t\t\t\t\t\twhile ( (node = ++nodeIndex && node && node[ dir ] ||\n\t\t\t\t\t\t\t\t\t(diff = nodeIndex = 0) || start.pop()) ) {\n\n\t\t\t\t\t\t\t\t\tif ( ( ofType ?\n\t\t\t\t\t\t\t\t\t\tnode.nodeName.toLowerCase() === name :\n\t\t\t\t\t\t\t\t\t\tnode.nodeType === 1 ) &&\n\t\t\t\t\t\t\t\t\t\t++diff ) {\n\n\t\t\t\t\t\t\t\t\t\t// Cache the index of each encountered element\n\t\t\t\t\t\t\t\t\t\tif ( useCache ) {\n\t\t\t\t\t\t\t\t\t\t\touterCache = node[ expando ] || (node[ expando ] = {});\n\n\t\t\t\t\t\t\t\t\t\t\t// Support: IE <9 only\n\t\t\t\t\t\t\t\t\t\t\t// Defend against cloned attroperties (jQuery gh-1709)\n\t\t\t\t\t\t\t\t\t\t\tuniqueCache = outerCache[ node.uniqueID ] ||\n\t\t\t\t\t\t\t\t\t\t\t\t(outerCache[ node.uniqueID ] = {});\n\n\t\t\t\t\t\t\t\t\t\t\tuniqueCache[ type ] = [ dirruns, diff ];\n\t\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t\tif ( node === elem ) {\n\t\t\t\t\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// Incorporate the offset, then check against cycle size\n\t\t\t\t\t\tdiff -= last;\n\t\t\t\t\t\treturn diff === first || ( diff % first === 0 && diff / first >= 0 );\n\t\t\t\t\t}\n\t\t\t\t};\n\t\t},\n\n\t\t\"PSEUDO\": function( pseudo, argument ) {\n\t\t\t// pseudo-class names are case-insensitive\n\t\t\t// http://www.w3.org/TR/selectors/#pseudo-classes\n\t\t\t// Prioritize by case sensitivity in case custom pseudos are added with uppercase letters\n\t\t\t// Remember that setFilters inherits from pseudos\n\t\t\tvar args,\n\t\t\t\tfn = Expr.pseudos[ pseudo ] || Expr.setFilters[ pseudo.toLowerCase() ] ||\n\t\t\t\t\tSizzle.error( \"unsupported pseudo: \" + pseudo );\n\n\t\t\t// The user may use createPseudo to indicate that\n\t\t\t// arguments are needed to create the filter function\n\t\t\t// just as Sizzle does\n\t\t\tif ( fn[ expando ] ) {\n\t\t\t\treturn fn( argument );\n\t\t\t}\n\n\t\t\t// But maintain support for old signatures\n\t\t\tif ( fn.length > 1 ) {\n\t\t\t\targs = [ pseudo, pseudo, \"\", argument ];\n\t\t\t\treturn Expr.setFilters.hasOwnProperty( pseudo.toLowerCase() ) ?\n\t\t\t\t\tmarkFunction(function( seed, matches ) {\n\t\t\t\t\t\tvar idx,\n\t\t\t\t\t\t\tmatched = fn( seed, argument ),\n\t\t\t\t\t\t\ti = matched.length;\n\t\t\t\t\t\twhile ( i-- ) {\n\t\t\t\t\t\t\tidx = indexOf( seed, matched[i] );\n\t\t\t\t\t\t\tseed[ idx ] = !( matches[ idx ] = matched[i] );\n\t\t\t\t\t\t}\n\t\t\t\t\t}) :\n\t\t\t\t\tfunction( elem ) {\n\t\t\t\t\t\treturn fn( elem, 0, args );\n\t\t\t\t\t};\n\t\t\t}\n\n\t\t\treturn fn;\n\t\t}\n\t},\n\n\tpseudos: {\n\t\t// Potentially complex pseudos\n\t\t\"not\": markFunction(function( selector ) {\n\t\t\t// Trim the selector passed to compile\n\t\t\t// to avoid treating leading and trailing\n\t\t\t// spaces as combinators\n\t\t\tvar input = [],\n\t\t\t\tresults = [],\n\t\t\t\tmatcher = compile( selector.replace( rtrim, \"$1\" ) );\n\n\t\t\treturn matcher[ expando ] ?\n\t\t\t\tmarkFunction(function( seed, matches, context, xml ) {\n\t\t\t\t\tvar elem,\n\t\t\t\t\t\tunmatched = matcher( seed, null, xml, [] ),\n\t\t\t\t\t\ti = seed.length;\n\n\t\t\t\t\t// Match elements unmatched by `matcher`\n\t\t\t\t\twhile ( i-- ) {\n\t\t\t\t\t\tif ( (elem = unmatched[i]) ) {\n\t\t\t\t\t\t\tseed[i] = !(matches[i] = elem);\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}) :\n\t\t\t\tfunction( elem, context, xml ) {\n\t\t\t\t\tinput[0] = elem;\n\t\t\t\t\tmatcher( input, null, xml, results );\n\t\t\t\t\t// Don't keep the element (issue #299)\n\t\t\t\t\tinput[0] = null;\n\t\t\t\t\treturn !results.pop();\n\t\t\t\t};\n\t\t}),\n\n\t\t\"has\": markFunction(function( selector ) {\n\t\t\treturn function( elem ) {\n\t\t\t\treturn Sizzle( selector, elem ).length > 0;\n\t\t\t};\n\t\t}),\n\n\t\t\"contains\": markFunction(function( text ) {\n\t\t\ttext = text.replace( runescape, funescape );\n\t\t\treturn function( elem ) {\n\t\t\t\treturn ( elem.textContent || elem.innerText || getText( elem ) ).indexOf( text ) > -1;\n\t\t\t};\n\t\t}),\n\n\t\t// \"Whether an element is represented by a :lang() selector\n\t\t// is based solely on the element's language value\n\t\t// being equal to the identifier C,\n\t\t// or beginning with the identifier C immediately followed by \"-\".\n\t\t// The matching of C against the element's language value is performed case-insensitively.\n\t\t// The identifier C does not have to be a valid language name.\"\n\t\t// http://www.w3.org/TR/selectors/#lang-pseudo\n\t\t\"lang\": markFunction( function( lang ) {\n\t\t\t// lang value must be a valid identifier\n\t\t\tif ( !ridentifier.test(lang || \"\") ) {\n\t\t\t\tSizzle.error( \"unsupported lang: \" + lang );\n\t\t\t}\n\t\t\tlang = lang.replace( runescape, funescape ).toLowerCase();\n\t\t\treturn function( elem ) {\n\t\t\t\tvar elemLang;\n\t\t\t\tdo {\n\t\t\t\t\tif ( (elemLang = documentIsHTML ?\n\t\t\t\t\t\telem.lang :\n\t\t\t\t\t\telem.getAttribute(\"xml:lang\") || elem.getAttribute(\"lang\")) ) {\n\n\t\t\t\t\t\telemLang = elemLang.toLowerCase();\n\t\t\t\t\t\treturn elemLang === lang || elemLang.indexOf( lang + \"-\" ) === 0;\n\t\t\t\t\t}\n\t\t\t\t} while ( (elem = elem.parentNode) && elem.nodeType === 1 );\n\t\t\t\treturn false;\n\t\t\t};\n\t\t}),\n\n\t\t// Miscellaneous\n\t\t\"target\": function( elem ) {\n\t\t\tvar hash = window.location && window.location.hash;\n\t\t\treturn hash && hash.slice( 1 ) === elem.id;\n\t\t},\n\n\t\t\"root\": function( elem ) {\n\t\t\treturn elem === docElem;\n\t\t},\n\n\t\t\"focus\": function( elem ) {\n\t\t\treturn elem === document.activeElement && (!document.hasFocus || document.hasFocus()) && !!(elem.type || elem.href || ~elem.tabIndex);\n\t\t},\n\n\t\t// Boolean properties\n\t\t\"enabled\": createDisabledPseudo( false ),\n\t\t\"disabled\": createDisabledPseudo( true ),\n\n\t\t\"checked\": function( elem ) {\n\t\t\t// In CSS3, :checked should return both checked and selected elements\n\t\t\t// http://www.w3.org/TR/2011/REC-css3-selectors-20110929/#checked\n\t\t\tvar nodeName = elem.nodeName.toLowerCase();\n\t\t\treturn (nodeName === \"input\" && !!elem.checked) || (nodeName === \"option\" && !!elem.selected);\n\t\t},\n\n\t\t\"selected\": function( elem ) {\n\t\t\t// Accessing this property makes selected-by-default\n\t\t\t// options in Safari work properly\n\t\t\tif ( elem.parentNode ) {\n\t\t\t\telem.parentNode.selectedIndex;\n\t\t\t}\n\n\t\t\treturn elem.selected === true;\n\t\t},\n\n\t\t// Contents\n\t\t\"empty\": function( elem ) {\n\t\t\t// http://www.w3.org/TR/selectors/#empty-pseudo\n\t\t\t// :empty is negated by element (1) or content nodes (text: 3; cdata: 4; entity ref: 5),\n\t\t\t// but not by others (comment: 8; processing instruction: 7; etc.)\n\t\t\t// nodeType < 6 works because attributes (2) do not appear as children\n\t\t\tfor ( elem = elem.firstChild; elem; elem = elem.nextSibling ) {\n\t\t\t\tif ( elem.nodeType < 6 ) {\n\t\t\t\t\treturn false;\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn true;\n\t\t},\n\n\t\t\"parent\": function( elem ) {\n\t\t\treturn !Expr.pseudos[\"empty\"]( elem );\n\t\t},\n\n\t\t// Element/input types\n\t\t\"header\": function( elem ) {\n\t\t\treturn rheader.test( elem.nodeName );\n\t\t},\n\n\t\t\"input\": function( elem ) {\n\t\t\treturn rinputs.test( elem.nodeName );\n\t\t},\n\n\t\t\"button\": function( elem ) {\n\t\t\tvar name = elem.nodeName.toLowerCase();\n\t\t\treturn name === \"input\" && elem.type === \"button\" || name === \"button\";\n\t\t},\n\n\t\t\"text\": function( elem ) {\n\t\t\tvar attr;\n\t\t\treturn elem.nodeName.toLowerCase() === \"input\" &&\n\t\t\t\telem.type === \"text\" &&\n\n\t\t\t\t// Support: IE<8\n\t\t\t\t// New HTML5 attribute values (e.g., \"search\") appear with elem.type === \"text\"\n\t\t\t\t( (attr = elem.getAttribute(\"type\")) == null || attr.toLowerCase() === \"text\" );\n\t\t},\n\n\t\t// Position-in-collection\n\t\t\"first\": createPositionalPseudo(function() {\n\t\t\treturn [ 0 ];\n\t\t}),\n\n\t\t\"last\": createPositionalPseudo(function( matchIndexes, length ) {\n\t\t\treturn [ length - 1 ];\n\t\t}),\n\n\t\t\"eq\": createPositionalPseudo(function( matchIndexes, length, argument ) {\n\t\t\treturn [ argument < 0 ? argument + length : argument ];\n\t\t}),\n\n\t\t\"even\": createPositionalPseudo(function( matchIndexes, length ) {\n\t\t\tvar i = 0;\n\t\t\tfor ( ; i < length; i += 2 ) {\n\t\t\t\tmatchIndexes.push( i );\n\t\t\t}\n\t\t\treturn matchIndexes;\n\t\t}),\n\n\t\t\"odd\": createPositionalPseudo(function( matchIndexes, length ) {\n\t\t\tvar i = 1;\n\t\t\tfor ( ; i < length; i += 2 ) {\n\t\t\t\tmatchIndexes.push( i );\n\t\t\t}\n\t\t\treturn matchIndexes;\n\t\t}),\n\n\t\t\"lt\": createPositionalPseudo(function( matchIndexes, length, argument ) {\n\t\t\tvar i = argument < 0 ? argument + length : argument;\n\t\t\tfor ( ; --i >= 0; ) {\n\t\t\t\tmatchIndexes.push( i );\n\t\t\t}\n\t\t\treturn matchIndexes;\n\t\t}),\n\n\t\t\"gt\": createPositionalPseudo(function( matchIndexes, length, argument ) {\n\t\t\tvar i = argument < 0 ? argument + length : argument;\n\t\t\tfor ( ; ++i < length; ) {\n\t\t\t\tmatchIndexes.push( i );\n\t\t\t}\n\t\t\treturn matchIndexes;\n\t\t})\n\t}\n};\n\nExpr.pseudos[\"nth\"] = Expr.pseudos[\"eq\"];\n\n// Add button/input type pseudos\nfor ( i in { radio: true, checkbox: true, file: true, password: true, image: true } ) {\n\tExpr.pseudos[ i ] = createInputPseudo( i );\n}\nfor ( i in { submit: true, reset: true } ) {\n\tExpr.pseudos[ i ] = createButtonPseudo( i );\n}\n\n// Easy API for creating new setFilters\nfunction setFilters() {}\nsetFilters.prototype = Expr.filters = Expr.pseudos;\nExpr.setFilters = new setFilters();\n\ntokenize = Sizzle.tokenize = function( selector, parseOnly ) {\n\tvar matched, match, tokens, type,\n\t\tsoFar, groups, preFilters,\n\t\tcached = tokenCache[ selector + \" \" ];\n\n\tif ( cached ) {\n\t\treturn parseOnly ? 0 : cached.slice( 0 );\n\t}\n\n\tsoFar = selector;\n\tgroups = [];\n\tpreFilters = Expr.preFilter;\n\n\twhile ( soFar ) {\n\n\t\t// Comma and first run\n\t\tif ( !matched || (match = rcomma.exec( soFar )) ) {\n\t\t\tif ( match ) {\n\t\t\t\t// Don't consume trailing commas as valid\n\t\t\t\tsoFar = soFar.slice( match[0].length ) || soFar;\n\t\t\t}\n\t\t\tgroups.push( (tokens = []) );\n\t\t}\n\n\t\tmatched = false;\n\n\t\t// Combinators\n\t\tif ( (match = rcombinators.exec( soFar )) ) {\n\t\t\tmatched = match.shift();\n\t\t\ttokens.push({\n\t\t\t\tvalue: matched,\n\t\t\t\t// Cast descendant combinators to space\n\t\t\t\ttype: match[0].replace( rtrim, \" \" )\n\t\t\t});\n\t\t\tsoFar = soFar.slice( matched.length );\n\t\t}\n\n\t\t// Filters\n\t\tfor ( type in Expr.filter ) {\n\t\t\tif ( (match = matchExpr[ type ].exec( soFar )) && (!preFilters[ type ] ||\n\t\t\t\t(match = preFilters[ type ]( match ))) ) {\n\t\t\t\tmatched = match.shift();\n\t\t\t\ttokens.push({\n\t\t\t\t\tvalue: matched,\n\t\t\t\t\ttype: type,\n\t\t\t\t\tmatches: match\n\t\t\t\t});\n\t\t\t\tsoFar = soFar.slice( matched.length );\n\t\t\t}\n\t\t}\n\n\t\tif ( !matched ) {\n\t\t\tbreak;\n\t\t}\n\t}\n\n\t// Return the length of the invalid excess\n\t// if we're just parsing\n\t// Otherwise, throw an error or return tokens\n\treturn parseOnly ?\n\t\tsoFar.length :\n\t\tsoFar ?\n\t\t\tSizzle.error( selector ) :\n\t\t\t// Cache the tokens\n\t\t\ttokenCache( selector, groups ).slice( 0 );\n};\n\nfunction toSelector( tokens ) {\n\tvar i = 0,\n\t\tlen = tokens.length,\n\t\tselector = \"\";\n\tfor ( ; i < len; i++ ) {\n\t\tselector += tokens[i].value;\n\t}\n\treturn selector;\n}\n\nfunction addCombinator( matcher, combinator, base ) {\n\tvar dir = combinator.dir,\n\t\tskip = combinator.next,\n\t\tkey = skip || dir,\n\t\tcheckNonElements = base && key === \"parentNode\",\n\t\tdoneName = done++;\n\n\treturn combinator.first ?\n\t\t// Check against closest ancestor/preceding element\n\t\tfunction( elem, context, xml ) {\n\t\t\twhile ( (elem = elem[ dir ]) ) {\n\t\t\t\tif ( elem.nodeType === 1 || checkNonElements ) {\n\t\t\t\t\treturn matcher( elem, context, xml );\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn false;\n\t\t} :\n\n\t\t// Check against all ancestor/preceding elements\n\t\tfunction( elem, context, xml ) {\n\t\t\tvar oldCache, uniqueCache, outerCache,\n\t\t\t\tnewCache = [ dirruns, doneName ];\n\n\t\t\t// We can't set arbitrary data on XML nodes, so they don't benefit from combinator caching\n\t\t\tif ( xml ) {\n\t\t\t\twhile ( (elem = elem[ dir ]) ) {\n\t\t\t\t\tif ( elem.nodeType === 1 || checkNonElements ) {\n\t\t\t\t\t\tif ( matcher( elem, context, xml ) ) {\n\t\t\t\t\t\t\treturn true;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\twhile ( (elem = elem[ dir ]) ) {\n\t\t\t\t\tif ( elem.nodeType === 1 || checkNonElements ) {\n\t\t\t\t\t\touterCache = elem[ expando ] || (elem[ expando ] = {});\n\n\t\t\t\t\t\t// Support: IE <9 only\n\t\t\t\t\t\t// Defend against cloned attroperties (jQuery gh-1709)\n\t\t\t\t\t\tuniqueCache = outerCache[ elem.uniqueID ] || (outerCache[ elem.uniqueID ] = {});\n\n\t\t\t\t\t\tif ( skip && skip === elem.nodeName.toLowerCase() ) {\n\t\t\t\t\t\t\telem = elem[ dir ] || elem;\n\t\t\t\t\t\t} else if ( (oldCache = uniqueCache[ key ]) &&\n\t\t\t\t\t\t\toldCache[ 0 ] === dirruns && oldCache[ 1 ] === doneName ) {\n\n\t\t\t\t\t\t\t// Assign to newCache so results back-propagate to previous elements\n\t\t\t\t\t\t\treturn (newCache[ 2 ] = oldCache[ 2 ]);\n\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t// Reuse newcache so results back-propagate to previous elements\n\t\t\t\t\t\t\tuniqueCache[ key ] = newCache;\n\n\t\t\t\t\t\t\t// A match means we're done; a fail means we have to keep checking\n\t\t\t\t\t\t\tif ( (newCache[ 2 ] = matcher( elem, context, xml )) ) {\n\t\t\t\t\t\t\t\treturn true;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn false;\n\t\t};\n}\n\nfunction elementMatcher( matchers ) {\n\treturn matchers.length > 1 ?\n\t\tfunction( elem, context, xml ) {\n\t\t\tvar i = matchers.length;\n\t\t\twhile ( i-- ) {\n\t\t\t\tif ( !matchers[i]( elem, context, xml ) ) {\n\t\t\t\t\treturn false;\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn true;\n\t\t} :\n\t\tmatchers[0];\n}\n\nfunction multipleContexts( selector, contexts, results ) {\n\tvar i = 0,\n\t\tlen = contexts.length;\n\tfor ( ; i < len; i++ ) {\n\t\tSizzle( selector, contexts[i], results );\n\t}\n\treturn results;\n}\n\nfunction condense( unmatched, map, filter, context, xml ) {\n\tvar elem,\n\t\tnewUnmatched = [],\n\t\ti = 0,\n\t\tlen = unmatched.length,\n\t\tmapped = map != null;\n\n\tfor ( ; i < len; i++ ) {\n\t\tif ( (elem = unmatched[i]) ) {\n\t\t\tif ( !filter || filter( elem, context, xml ) ) {\n\t\t\t\tnewUnmatched.push( elem );\n\t\t\t\tif ( mapped ) {\n\t\t\t\t\tmap.push( i );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\treturn newUnmatched;\n}\n\nfunction setMatcher( preFilter, selector, matcher, postFilter, postFinder, postSelector ) {\n\tif ( postFilter && !postFilter[ expando ] ) {\n\t\tpostFilter = setMatcher( postFilter );\n\t}\n\tif ( postFinder && !postFinder[ expando ] ) {\n\t\tpostFinder = setMatcher( postFinder, postSelector );\n\t}\n\treturn markFunction(function( seed, results, context, xml ) {\n\t\tvar temp, i, elem,\n\t\t\tpreMap = [],\n\t\t\tpostMap = [],\n\t\t\tpreexisting = results.length,\n\n\t\t\t// Get initial elements from seed or context\n\t\t\telems = seed || multipleContexts( selector || \"*\", context.nodeType ? [ context ] : context, [] ),\n\n\t\t\t// Prefilter to get matcher input, preserving a map for seed-results synchronization\n\t\t\tmatcherIn = preFilter && ( seed || !selector ) ?\n\t\t\t\tcondense( elems, preMap, preFilter, context, xml ) :\n\t\t\t\telems,\n\n\t\t\tmatcherOut = matcher ?\n\t\t\t\t// If we have a postFinder, or filtered seed, or non-seed postFilter or preexisting results,\n\t\t\t\tpostFinder || ( seed ? preFilter : preexisting || postFilter ) ?\n\n\t\t\t\t\t// ...intermediate processing is necessary\n\t\t\t\t\t[] :\n\n\t\t\t\t\t// ...otherwise use results directly\n\t\t\t\t\tresults :\n\t\t\t\tmatcherIn;\n\n\t\t// Find primary matches\n\t\tif ( matcher ) {\n\t\t\tmatcher( matcherIn, matcherOut, context, xml );\n\t\t}\n\n\t\t// Apply postFilter\n\t\tif ( postFilter ) {\n\t\t\ttemp = condense( matcherOut, postMap );\n\t\t\tpostFilter( temp, [], context, xml );\n\n\t\t\t// Un-match failing elements by moving them back to matcherIn\n\t\t\ti = temp.length;\n\t\t\twhile ( i-- ) {\n\t\t\t\tif ( (elem = temp[i]) ) {\n\t\t\t\t\tmatcherOut[ postMap[i] ] = !(matcherIn[ postMap[i] ] = elem);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\tif ( seed ) {\n\t\t\tif ( postFinder || preFilter ) {\n\t\t\t\tif ( postFinder ) {\n\t\t\t\t\t// Get the final matcherOut by condensing this intermediate into postFinder contexts\n\t\t\t\t\ttemp = [];\n\t\t\t\t\ti = matcherOut.length;\n\t\t\t\t\twhile ( i-- ) {\n\t\t\t\t\t\tif ( (elem = matcherOut[i]) ) {\n\t\t\t\t\t\t\t// Restore matcherIn since elem is not yet a final match\n\t\t\t\t\t\t\ttemp.push( (matcherIn[i] = elem) );\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tpostFinder( null, (matcherOut = []), temp, xml );\n\t\t\t\t}\n\n\t\t\t\t// Move matched elements from seed to results to keep them synchronized\n\t\t\t\ti = matcherOut.length;\n\t\t\t\twhile ( i-- ) {\n\t\t\t\t\tif ( (elem = matcherOut[i]) &&\n\t\t\t\t\t\t(temp = postFinder ? indexOf( seed, elem ) : preMap[i]) > -1 ) {\n\n\t\t\t\t\t\tseed[temp] = !(results[temp] = elem);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t// Add elements to results, through postFinder if defined\n\t\t} else {\n\t\t\tmatcherOut = condense(\n\t\t\t\tmatcherOut === results ?\n\t\t\t\t\tmatcherOut.splice( preexisting, matcherOut.length ) :\n\t\t\t\t\tmatcherOut\n\t\t\t);\n\t\t\tif ( postFinder ) {\n\t\t\t\tpostFinder( null, results, matcherOut, xml );\n\t\t\t} else {\n\t\t\t\tpush.apply( results, matcherOut );\n\t\t\t}\n\t\t}\n\t});\n}\n\nfunction matcherFromTokens( tokens ) {\n\tvar checkContext, matcher, j,\n\t\tlen = tokens.length,\n\t\tleadingRelative = Expr.relative[ tokens[0].type ],\n\t\timplicitRelative = leadingRelative || Expr.relative[\" \"],\n\t\ti = leadingRelative ? 1 : 0,\n\n\t\t// The foundational matcher ensures that elements are reachable from top-level context(s)\n\t\tmatchContext = addCombinator( function( elem ) {\n\t\t\treturn elem === checkContext;\n\t\t}, implicitRelative, true ),\n\t\tmatchAnyContext = addCombinator( function( elem ) {\n\t\t\treturn indexOf( checkContext, elem ) > -1;\n\t\t}, implicitRelative, true ),\n\t\tmatchers = [ function( elem, context, xml ) {\n\t\t\tvar ret = ( !leadingRelative && ( xml || context !== outermostContext ) ) || (\n\t\t\t\t(checkContext = context).nodeType ?\n\t\t\t\t\tmatchContext( elem, context, xml ) :\n\t\t\t\t\tmatchAnyContext( elem, context, xml ) );\n\t\t\t// Avoid hanging onto element (issue #299)\n\t\t\tcheckContext = null;\n\t\t\treturn ret;\n\t\t} ];\n\n\tfor ( ; i < len; i++ ) {\n\t\tif ( (matcher = Expr.relative[ tokens[i].type ]) ) {\n\t\t\tmatchers = [ addCombinator(elementMatcher( matchers ), matcher) ];\n\t\t} else {\n\t\t\tmatcher = Expr.filter[ tokens[i].type ].apply( null, tokens[i].matches );\n\n\t\t\t// Return special upon seeing a positional matcher\n\t\t\tif ( matcher[ expando ] ) {\n\t\t\t\t// Find the next relative operator (if any) for proper handling\n\t\t\t\tj = ++i;\n\t\t\t\tfor ( ; j < len; j++ ) {\n\t\t\t\t\tif ( Expr.relative[ tokens[j].type ] ) {\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\treturn setMatcher(\n\t\t\t\t\ti > 1 && elementMatcher( matchers ),\n\t\t\t\t\ti > 1 && toSelector(\n\t\t\t\t\t\t// If the preceding token was a descendant combinator, insert an implicit any-element `*`\n\t\t\t\t\t\ttokens.slice( 0, i - 1 ).concat({ value: tokens[ i - 2 ].type === \" \" ? \"*\" : \"\" })\n\t\t\t\t\t).replace( rtrim, \"$1\" ),\n\t\t\t\t\tmatcher,\n\t\t\t\t\ti < j && matcherFromTokens( tokens.slice( i, j ) ),\n\t\t\t\t\tj < len && matcherFromTokens( (tokens = tokens.slice( j )) ),\n\t\t\t\t\tj < len && toSelector( tokens )\n\t\t\t\t);\n\t\t\t}\n\t\t\tmatchers.push( matcher );\n\t\t}\n\t}\n\n\treturn elementMatcher( matchers );\n}\n\nfunction matcherFromGroupMatchers( elementMatchers, setMatchers ) {\n\tvar bySet = setMatchers.length > 0,\n\t\tbyElement = elementMatchers.length > 0,\n\t\tsuperMatcher = function( seed, context, xml, results, outermost ) {\n\t\t\tvar elem, j, matcher,\n\t\t\t\tmatchedCount = 0,\n\t\t\t\ti = \"0\",\n\t\t\t\tunmatched = seed && [],\n\t\t\t\tsetMatched = [],\n\t\t\t\tcontextBackup = outermostContext,\n\t\t\t\t// We must always have either seed elements or outermost context\n\t\t\t\telems = seed || byElement && Expr.find[\"TAG\"]( \"*\", outermost ),\n\t\t\t\t// Use integer dirruns iff this is the outermost matcher\n\t\t\t\tdirrunsUnique = (dirruns += contextBackup == null ? 1 : Math.random() || 0.1),\n\t\t\t\tlen = elems.length;\n\n\t\t\tif ( outermost ) {\n\t\t\t\toutermostContext = context === document || context || outermost;\n\t\t\t}\n\n\t\t\t// Add elements passing elementMatchers directly to results\n\t\t\t// Support: IE<9, Safari\n\t\t\t// Tolerate NodeList properties (IE: \"length\"; Safari: ) matching elements by id\n\t\t\tfor ( ; i !== len && (elem = elems[i]) != null; i++ ) {\n\t\t\t\tif ( byElement && elem ) {\n\t\t\t\t\tj = 0;\n\t\t\t\t\tif ( !context && elem.ownerDocument !== document ) {\n\t\t\t\t\t\tsetDocument( elem );\n\t\t\t\t\t\txml = !documentIsHTML;\n\t\t\t\t\t}\n\t\t\t\t\twhile ( (matcher = elementMatchers[j++]) ) {\n\t\t\t\t\t\tif ( matcher( elem, context || document, xml) ) {\n\t\t\t\t\t\t\tresults.push( elem );\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif ( outermost ) {\n\t\t\t\t\t\tdirruns = dirrunsUnique;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Track unmatched elements for set filters\n\t\t\t\tif ( bySet ) {\n\t\t\t\t\t// They will have gone through all possible matchers\n\t\t\t\t\tif ( (elem = !matcher && elem) ) {\n\t\t\t\t\t\tmatchedCount--;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Lengthen the array for every element, matched or not\n\t\t\t\t\tif ( seed ) {\n\t\t\t\t\t\tunmatched.push( elem );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// `i` is now the count of elements visited above, and adding it to `matchedCount`\n\t\t\t// makes the latter nonnegative.\n\t\t\tmatchedCount += i;\n\n\t\t\t// Apply set filters to unmatched elements\n\t\t\t// NOTE: This can be skipped if there are no unmatched elements (i.e., `matchedCount`\n\t\t\t// equals `i`), unless we didn't visit _any_ elements in the above loop because we have\n\t\t\t// no element matchers and no seed.\n\t\t\t// Incrementing an initially-string \"0\" `i` allows `i` to remain a string only in that\n\t\t\t// case, which will result in a \"00\" `matchedCount` that differs from `i` but is also\n\t\t\t// numerically zero.\n\t\t\tif ( bySet && i !== matchedCount ) {\n\t\t\t\tj = 0;\n\t\t\t\twhile ( (matcher = setMatchers[j++]) ) {\n\t\t\t\t\tmatcher( unmatched, setMatched, context, xml );\n\t\t\t\t}\n\n\t\t\t\tif ( seed ) {\n\t\t\t\t\t// Reintegrate element matches to eliminate the need for sorting\n\t\t\t\t\tif ( matchedCount > 0 ) {\n\t\t\t\t\t\twhile ( i-- ) {\n\t\t\t\t\t\t\tif ( !(unmatched[i] || setMatched[i]) ) {\n\t\t\t\t\t\t\t\tsetMatched[i] = pop.call( results );\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\t// Discard index placeholder values to get only actual matches\n\t\t\t\t\tsetMatched = condense( setMatched );\n\t\t\t\t}\n\n\t\t\t\t// Add matches to results\n\t\t\t\tpush.apply( results, setMatched );\n\n\t\t\t\t// Seedless set matches succeeding multiple successful matchers stipulate sorting\n\t\t\t\tif ( outermost && !seed && setMatched.length > 0 &&\n\t\t\t\t\t( matchedCount + setMatchers.length ) > 1 ) {\n\n\t\t\t\t\tSizzle.uniqueSort( results );\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Override manipulation of globals by nested matchers\n\t\t\tif ( outermost ) {\n\t\t\t\tdirruns = dirrunsUnique;\n\t\t\t\toutermostContext = contextBackup;\n\t\t\t}\n\n\t\t\treturn unmatched;\n\t\t};\n\n\treturn bySet ?\n\t\tmarkFunction( superMatcher ) :\n\t\tsuperMatcher;\n}\n\ncompile = Sizzle.compile = function( selector, match /* Internal Use Only */ ) {\n\tvar i,\n\t\tsetMatchers = [],\n\t\telementMatchers = [],\n\t\tcached = compilerCache[ selector + \" \" ];\n\n\tif ( !cached ) {\n\t\t// Generate a function of recursive functions that can be used to check each element\n\t\tif ( !match ) {\n\t\t\tmatch = tokenize( selector );\n\t\t}\n\t\ti = match.length;\n\t\twhile ( i-- ) {\n\t\t\tcached = matcherFromTokens( match[i] );\n\t\t\tif ( cached[ expando ] ) {\n\t\t\t\tsetMatchers.push( cached );\n\t\t\t} else {\n\t\t\t\telementMatchers.push( cached );\n\t\t\t}\n\t\t}\n\n\t\t// Cache the compiled function\n\t\tcached = compilerCache( selector, matcherFromGroupMatchers( elementMatchers, setMatchers ) );\n\n\t\t// Save selector and tokenization\n\t\tcached.selector = selector;\n\t}\n\treturn cached;\n};\n\n/**\n * A low-level selection function that works with Sizzle's compiled\n * selector functions\n * @param {String|Function} selector A selector or a pre-compiled\n * selector function built with Sizzle.compile\n * @param {Element} context\n * @param {Array} [results]\n * @param {Array} [seed] A set of elements to match against\n */\nselect = Sizzle.select = function( selector, context, results, seed ) {\n\tvar i, tokens, token, type, find,\n\t\tcompiled = typeof selector === \"function\" && selector,\n\t\tmatch = !seed && tokenize( (selector = compiled.selector || selector) );\n\n\tresults = results || [];\n\n\t// Try to minimize operations if there is only one selector in the list and no seed\n\t// (the latter of which guarantees us context)\n\tif ( match.length === 1 ) {\n\n\t\t// Reduce context if the leading compound selector is an ID\n\t\ttokens = match[0] = match[0].slice( 0 );\n\t\tif ( tokens.length > 2 && (token = tokens[0]).type === \"ID\" &&\n\t\t\t\tcontext.nodeType === 9 && documentIsHTML && Expr.relative[ tokens[1].type ] ) {\n\n\t\t\tcontext = ( Expr.find[\"ID\"]( token.matches[0].replace(runescape, funescape), context ) || [] )[0];\n\t\t\tif ( !context ) {\n\t\t\t\treturn results;\n\n\t\t\t// Precompiled matchers will still verify ancestry, so step up a level\n\t\t\t} else if ( compiled ) {\n\t\t\t\tcontext = context.parentNode;\n\t\t\t}\n\n\t\t\tselector = selector.slice( tokens.shift().value.length );\n\t\t}\n\n\t\t// Fetch a seed set for right-to-left matching\n\t\ti = matchExpr[\"needsContext\"].test( selector ) ? 0 : tokens.length;\n\t\twhile ( i-- ) {\n\t\t\ttoken = tokens[i];\n\n\t\t\t// Abort if we hit a combinator\n\t\t\tif ( Expr.relative[ (type = token.type) ] ) {\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tif ( (find = Expr.find[ type ]) ) {\n\t\t\t\t// Search, expanding context for leading sibling combinators\n\t\t\t\tif ( (seed = find(\n\t\t\t\t\ttoken.matches[0].replace( runescape, funescape ),\n\t\t\t\t\trsibling.test( tokens[0].type ) && testContext( context.parentNode ) || context\n\t\t\t\t)) ) {\n\n\t\t\t\t\t// If seed is empty or no tokens remain, we can return early\n\t\t\t\t\ttokens.splice( i, 1 );\n\t\t\t\t\tselector = seed.length && toSelector( tokens );\n\t\t\t\t\tif ( !selector ) {\n\t\t\t\t\t\tpush.apply( results, seed );\n\t\t\t\t\t\treturn results;\n\t\t\t\t\t}\n\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Compile and execute a filtering function if one is not provided\n\t// Provide `match` to avoid retokenization if we modified the selector above\n\t( compiled || compile( selector, match ) )(\n\t\tseed,\n\t\tcontext,\n\t\t!documentIsHTML,\n\t\tresults,\n\t\t!context || rsibling.test( selector ) && testContext( context.parentNode ) || context\n\t);\n\treturn results;\n};\n\n// One-time assignments\n\n// Sort stability\nsupport.sortStable = expando.split(\"\").sort( sortOrder ).join(\"\") === expando;\n\n// Support: Chrome 14-35+\n// Always assume duplicates if they aren't passed to the comparison function\nsupport.detectDuplicates = !!hasDuplicate;\n\n// Initialize against the default document\nsetDocument();\n\n// Support: Webkit<537.32 - Safari 6.0.3/Chrome 25 (fixed in Chrome 27)\n// Detached nodes confoundingly follow *each other*\nsupport.sortDetached = assert(function( el ) {\n\t// Should return 1, but returns 4 (following)\n\treturn el.compareDocumentPosition( document.createElement(\"fieldset\") ) & 1;\n});\n\n// Support: IE<8\n// Prevent attribute/property \"interpolation\"\n// https://msdn.microsoft.com/en-us/library/ms536429%28VS.85%29.aspx\nif ( !assert(function( el ) {\n\tel.innerHTML = \"\";\n\treturn el.firstChild.getAttribute(\"href\") === \"#\" ;\n}) ) {\n\taddHandle( \"type|href|height|width\", function( elem, name, isXML ) {\n\t\tif ( !isXML ) {\n\t\t\treturn elem.getAttribute( name, name.toLowerCase() === \"type\" ? 1 : 2 );\n\t\t}\n\t});\n}\n\n// Support: IE<9\n// Use defaultValue in place of getAttribute(\"value\")\nif ( !support.attributes || !assert(function( el ) {\n\tel.innerHTML = \"\";\n\tel.firstChild.setAttribute( \"value\", \"\" );\n\treturn el.firstChild.getAttribute( \"value\" ) === \"\";\n}) ) {\n\taddHandle( \"value\", function( elem, name, isXML ) {\n\t\tif ( !isXML && elem.nodeName.toLowerCase() === \"input\" ) {\n\t\t\treturn elem.defaultValue;\n\t\t}\n\t});\n}\n\n// Support: IE<9\n// Use getAttributeNode to fetch booleans when getAttribute lies\nif ( !assert(function( el ) {\n\treturn el.getAttribute(\"disabled\") == null;\n}) ) {\n\taddHandle( booleans, function( elem, name, isXML ) {\n\t\tvar val;\n\t\tif ( !isXML ) {\n\t\t\treturn elem[ name ] === true ? name.toLowerCase() :\n\t\t\t\t\t(val = elem.getAttributeNode( name )) && val.specified ?\n\t\t\t\t\tval.value :\n\t\t\t\tnull;\n\t\t}\n\t});\n}\n\nreturn Sizzle;\n\n})( window );\n\n\n\njQuery.find = Sizzle;\njQuery.expr = Sizzle.selectors;\n\n// Deprecated\njQuery.expr[ \":\" ] = jQuery.expr.pseudos;\njQuery.uniqueSort = jQuery.unique = Sizzle.uniqueSort;\njQuery.text = Sizzle.getText;\njQuery.isXMLDoc = Sizzle.isXML;\njQuery.contains = Sizzle.contains;\njQuery.escapeSelector = Sizzle.escape;\n\n\n\n\nvar dir = function( elem, dir, until ) {\n\tvar matched = [],\n\t\ttruncate = until !== undefined;\n\n\twhile ( ( elem = elem[ dir ] ) && elem.nodeType !== 9 ) {\n\t\tif ( elem.nodeType === 1 ) {\n\t\t\tif ( truncate && jQuery( elem ).is( until ) ) {\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tmatched.push( elem );\n\t\t}\n\t}\n\treturn matched;\n};\n\n\nvar siblings = function( n, elem ) {\n\tvar matched = [];\n\n\tfor ( ; n; n = n.nextSibling ) {\n\t\tif ( n.nodeType === 1 && n !== elem ) {\n\t\t\tmatched.push( n );\n\t\t}\n\t}\n\n\treturn matched;\n};\n\n\nvar rneedsContext = jQuery.expr.match.needsContext;\n\n\n\nfunction nodeName( elem, name ) {\n\n return elem.nodeName && elem.nodeName.toLowerCase() === name.toLowerCase();\n\n};\nvar rsingleTag = ( /^<([a-z][^\\/\\0>:\\x20\\t\\r\\n\\f]*)[\\x20\\t\\r\\n\\f]*\\/?>(?:<\\/\\1>|)$/i );\n\n\n\nvar risSimple = /^.[^:#\\[\\.,]*$/;\n\n// Implement the identical functionality for filter and not\nfunction winnow( elements, qualifier, not ) {\n\tif ( jQuery.isFunction( qualifier ) ) {\n\t\treturn jQuery.grep( elements, function( elem, i ) {\n\t\t\treturn !!qualifier.call( elem, i, elem ) !== not;\n\t\t} );\n\t}\n\n\t// Single element\n\tif ( qualifier.nodeType ) {\n\t\treturn jQuery.grep( elements, function( elem ) {\n\t\t\treturn ( elem === qualifier ) !== not;\n\t\t} );\n\t}\n\n\t// Arraylike of elements (jQuery, arguments, Array)\n\tif ( typeof qualifier !== \"string\" ) {\n\t\treturn jQuery.grep( elements, function( elem ) {\n\t\t\treturn ( indexOf.call( qualifier, elem ) > -1 ) !== not;\n\t\t} );\n\t}\n\n\t// Simple selector that can be filtered directly, removing non-Elements\n\tif ( risSimple.test( qualifier ) ) {\n\t\treturn jQuery.filter( qualifier, elements, not );\n\t}\n\n\t// Complex selector, compare the two sets, removing non-Elements\n\tqualifier = jQuery.filter( qualifier, elements );\n\treturn jQuery.grep( elements, function( elem ) {\n\t\treturn ( indexOf.call( qualifier, elem ) > -1 ) !== not && elem.nodeType === 1;\n\t} );\n}\n\njQuery.filter = function( expr, elems, not ) {\n\tvar elem = elems[ 0 ];\n\n\tif ( not ) {\n\t\texpr = \":not(\" + expr + \")\";\n\t}\n\n\tif ( elems.length === 1 && elem.nodeType === 1 ) {\n\t\treturn jQuery.find.matchesSelector( elem, expr ) ? [ elem ] : [];\n\t}\n\n\treturn jQuery.find.matches( expr, jQuery.grep( elems, function( elem ) {\n\t\treturn elem.nodeType === 1;\n\t} ) );\n};\n\njQuery.fn.extend( {\n\tfind: function( selector ) {\n\t\tvar i, ret,\n\t\t\tlen = this.length,\n\t\t\tself = this;\n\n\t\tif ( typeof selector !== \"string\" ) {\n\t\t\treturn this.pushStack( jQuery( selector ).filter( function() {\n\t\t\t\tfor ( i = 0; i < len; i++ ) {\n\t\t\t\t\tif ( jQuery.contains( self[ i ], this ) ) {\n\t\t\t\t\t\treturn true;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} ) );\n\t\t}\n\n\t\tret = this.pushStack( [] );\n\n\t\tfor ( i = 0; i < len; i++ ) {\n\t\t\tjQuery.find( selector, self[ i ], ret );\n\t\t}\n\n\t\treturn len > 1 ? jQuery.uniqueSort( ret ) : ret;\n\t},\n\tfilter: function( selector ) {\n\t\treturn this.pushStack( winnow( this, selector || [], false ) );\n\t},\n\tnot: function( selector ) {\n\t\treturn this.pushStack( winnow( this, selector || [], true ) );\n\t},\n\tis: function( selector ) {\n\t\treturn !!winnow(\n\t\t\tthis,\n\n\t\t\t// If this is a positional/relative selector, check membership in the returned set\n\t\t\t// so $(\"p:first\").is(\"p:last\") won't return true for a doc with two \"p\".\n\t\t\ttypeof selector === \"string\" && rneedsContext.test( selector ) ?\n\t\t\t\tjQuery( selector ) :\n\t\t\t\tselector || [],\n\t\t\tfalse\n\t\t).length;\n\t}\n} );\n\n\n// Initialize a jQuery object\n\n\n// A central reference to the root jQuery(document)\nvar rootjQuery,\n\n\t// A simple way to check for HTML strings\n\t// Prioritize #id over to avoid XSS via location.hash (#9521)\n\t// Strict HTML recognition (#11290: must start with <)\n\t// Shortcut simple #id case for speed\n\trquickExpr = /^(?:\\s*(<[\\w\\W]+>)[^>]*|#([\\w-]+))$/,\n\n\tinit = jQuery.fn.init = function( selector, context, root ) {\n\t\tvar match, elem;\n\n\t\t// HANDLE: $(\"\"), $(null), $(undefined), $(false)\n\t\tif ( !selector ) {\n\t\t\treturn this;\n\t\t}\n\n\t\t// Method init() accepts an alternate rootjQuery\n\t\t// so migrate can support jQuery.sub (gh-2101)\n\t\troot = root || rootjQuery;\n\n\t\t// Handle HTML strings\n\t\tif ( typeof selector === \"string\" ) {\n\t\t\tif ( selector[ 0 ] === \"<\" &&\n\t\t\t\tselector[ selector.length - 1 ] === \">\" &&\n\t\t\t\tselector.length >= 3 ) {\n\n\t\t\t\t// Assume that strings that start and end with <> are HTML and skip the regex check\n\t\t\t\tmatch = [ null, selector, null ];\n\n\t\t\t} else {\n\t\t\t\tmatch = rquickExpr.exec( selector );\n\t\t\t}\n\n\t\t\t// Match html or make sure no context is specified for #id\n\t\t\tif ( match && ( match[ 1 ] || !context ) ) {\n\n\t\t\t\t// HANDLE: $(html) -> $(array)\n\t\t\t\tif ( match[ 1 ] ) {\n\t\t\t\t\tcontext = context instanceof jQuery ? context[ 0 ] : context;\n\n\t\t\t\t\t// Option to run scripts is true for back-compat\n\t\t\t\t\t// Intentionally let the error be thrown if parseHTML is not present\n\t\t\t\t\tjQuery.merge( this, jQuery.parseHTML(\n\t\t\t\t\t\tmatch[ 1 ],\n\t\t\t\t\t\tcontext && context.nodeType ? context.ownerDocument || context : document,\n\t\t\t\t\t\ttrue\n\t\t\t\t\t) );\n\n\t\t\t\t\t// HANDLE: $(html, props)\n\t\t\t\t\tif ( rsingleTag.test( match[ 1 ] ) && jQuery.isPlainObject( context ) ) {\n\t\t\t\t\t\tfor ( match in context ) {\n\n\t\t\t\t\t\t\t// Properties of context are called as methods if possible\n\t\t\t\t\t\t\tif ( jQuery.isFunction( this[ match ] ) ) {\n\t\t\t\t\t\t\t\tthis[ match ]( context[ match ] );\n\n\t\t\t\t\t\t\t// ...and otherwise set as attributes\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tthis.attr( match, context[ match ] );\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\treturn this;\n\n\t\t\t\t// HANDLE: $(#id)\n\t\t\t\t} else {\n\t\t\t\t\telem = document.getElementById( match[ 2 ] );\n\n\t\t\t\t\tif ( elem ) {\n\n\t\t\t\t\t\t// Inject the element directly into the jQuery object\n\t\t\t\t\t\tthis[ 0 ] = elem;\n\t\t\t\t\t\tthis.length = 1;\n\t\t\t\t\t}\n\t\t\t\t\treturn this;\n\t\t\t\t}\n\n\t\t\t// HANDLE: $(expr, $(...))\n\t\t\t} else if ( !context || context.jquery ) {\n\t\t\t\treturn ( context || root ).find( selector );\n\n\t\t\t// HANDLE: $(expr, context)\n\t\t\t// (which is just equivalent to: $(context).find(expr)\n\t\t\t} else {\n\t\t\t\treturn this.constructor( context ).find( selector );\n\t\t\t}\n\n\t\t// HANDLE: $(DOMElement)\n\t\t} else if ( selector.nodeType ) {\n\t\t\tthis[ 0 ] = selector;\n\t\t\tthis.length = 1;\n\t\t\treturn this;\n\n\t\t// HANDLE: $(function)\n\t\t// Shortcut for document ready\n\t\t} else if ( jQuery.isFunction( selector ) ) {\n\t\t\treturn root.ready !== undefined ?\n\t\t\t\troot.ready( selector ) :\n\n\t\t\t\t// Execute immediately if ready is not present\n\t\t\t\tselector( jQuery );\n\t\t}\n\n\t\treturn jQuery.makeArray( selector, this );\n\t};\n\n// Give the init function the jQuery prototype for later instantiation\ninit.prototype = jQuery.fn;\n\n// Initialize central reference\nrootjQuery = jQuery( document );\n\n\nvar rparentsprev = /^(?:parents|prev(?:Until|All))/,\n\n\t// Methods guaranteed to produce a unique set when starting from a unique set\n\tguaranteedUnique = {\n\t\tchildren: true,\n\t\tcontents: true,\n\t\tnext: true,\n\t\tprev: true\n\t};\n\njQuery.fn.extend( {\n\thas: function( target ) {\n\t\tvar targets = jQuery( target, this ),\n\t\t\tl = targets.length;\n\n\t\treturn this.filter( function() {\n\t\t\tvar i = 0;\n\t\t\tfor ( ; i < l; i++ ) {\n\t\t\t\tif ( jQuery.contains( this, targets[ i ] ) ) {\n\t\t\t\t\treturn true;\n\t\t\t\t}\n\t\t\t}\n\t\t} );\n\t},\n\n\tclosest: function( selectors, context ) {\n\t\tvar cur,\n\t\t\ti = 0,\n\t\t\tl = this.length,\n\t\t\tmatched = [],\n\t\t\ttargets = typeof selectors !== \"string\" && jQuery( selectors );\n\n\t\t// Positional selectors never match, since there's no _selection_ context\n\t\tif ( !rneedsContext.test( selectors ) ) {\n\t\t\tfor ( ; i < l; i++ ) {\n\t\t\t\tfor ( cur = this[ i ]; cur && cur !== context; cur = cur.parentNode ) {\n\n\t\t\t\t\t// Always skip document fragments\n\t\t\t\t\tif ( cur.nodeType < 11 && ( targets ?\n\t\t\t\t\t\ttargets.index( cur ) > -1 :\n\n\t\t\t\t\t\t// Don't pass non-elements to Sizzle\n\t\t\t\t\t\tcur.nodeType === 1 &&\n\t\t\t\t\t\t\tjQuery.find.matchesSelector( cur, selectors ) ) ) {\n\n\t\t\t\t\t\tmatched.push( cur );\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn this.pushStack( matched.length > 1 ? jQuery.uniqueSort( matched ) : matched );\n\t},\n\n\t// Determine the position of an element within the set\n\tindex: function( elem ) {\n\n\t\t// No argument, return index in parent\n\t\tif ( !elem ) {\n\t\t\treturn ( this[ 0 ] && this[ 0 ].parentNode ) ? this.first().prevAll().length : -1;\n\t\t}\n\n\t\t// Index in selector\n\t\tif ( typeof elem === \"string\" ) {\n\t\t\treturn indexOf.call( jQuery( elem ), this[ 0 ] );\n\t\t}\n\n\t\t// Locate the position of the desired element\n\t\treturn indexOf.call( this,\n\n\t\t\t// If it receives a jQuery object, the first element is used\n\t\t\telem.jquery ? elem[ 0 ] : elem\n\t\t);\n\t},\n\n\tadd: function( selector, context ) {\n\t\treturn this.pushStack(\n\t\t\tjQuery.uniqueSort(\n\t\t\t\tjQuery.merge( this.get(), jQuery( selector, context ) )\n\t\t\t)\n\t\t);\n\t},\n\n\taddBack: function( selector ) {\n\t\treturn this.add( selector == null ?\n\t\t\tthis.prevObject : this.prevObject.filter( selector )\n\t\t);\n\t}\n} );\n\nfunction sibling( cur, dir ) {\n\twhile ( ( cur = cur[ dir ] ) && cur.nodeType !== 1 ) {}\n\treturn cur;\n}\n\njQuery.each( {\n\tparent: function( elem ) {\n\t\tvar parent = elem.parentNode;\n\t\treturn parent && parent.nodeType !== 11 ? parent : null;\n\t},\n\tparents: function( elem ) {\n\t\treturn dir( elem, \"parentNode\" );\n\t},\n\tparentsUntil: function( elem, i, until ) {\n\t\treturn dir( elem, \"parentNode\", until );\n\t},\n\tnext: function( elem ) {\n\t\treturn sibling( elem, \"nextSibling\" );\n\t},\n\tprev: function( elem ) {\n\t\treturn sibling( elem, \"previousSibling\" );\n\t},\n\tnextAll: function( elem ) {\n\t\treturn dir( elem, \"nextSibling\" );\n\t},\n\tprevAll: function( elem ) {\n\t\treturn dir( elem, \"previousSibling\" );\n\t},\n\tnextUntil: function( elem, i, until ) {\n\t\treturn dir( elem, \"nextSibling\", until );\n\t},\n\tprevUntil: function( elem, i, until ) {\n\t\treturn dir( elem, \"previousSibling\", until );\n\t},\n\tsiblings: function( elem ) {\n\t\treturn siblings( ( elem.parentNode || {} ).firstChild, elem );\n\t},\n\tchildren: function( elem ) {\n\t\treturn siblings( elem.firstChild );\n\t},\n\tcontents: function( elem ) {\n if ( nodeName( elem, \"iframe\" ) ) {\n return elem.contentDocument;\n }\n\n // Support: IE 9 - 11 only, iOS 7 only, Android Browser <=4.3 only\n // Treat the template element as a regular one in browsers that\n // don't support it.\n if ( nodeName( elem, \"template\" ) ) {\n elem = elem.content || elem;\n }\n\n return jQuery.merge( [], elem.childNodes );\n\t}\n}, function( name, fn ) {\n\tjQuery.fn[ name ] = function( until, selector ) {\n\t\tvar matched = jQuery.map( this, fn, until );\n\n\t\tif ( name.slice( -5 ) !== \"Until\" ) {\n\t\t\tselector = until;\n\t\t}\n\n\t\tif ( selector && typeof selector === \"string\" ) {\n\t\t\tmatched = jQuery.filter( selector, matched );\n\t\t}\n\n\t\tif ( this.length > 1 ) {\n\n\t\t\t// Remove duplicates\n\t\t\tif ( !guaranteedUnique[ name ] ) {\n\t\t\t\tjQuery.uniqueSort( matched );\n\t\t\t}\n\n\t\t\t// Reverse order for parents* and prev-derivatives\n\t\t\tif ( rparentsprev.test( name ) ) {\n\t\t\t\tmatched.reverse();\n\t\t\t}\n\t\t}\n\n\t\treturn this.pushStack( matched );\n\t};\n} );\nvar rnothtmlwhite = ( /[^\\x20\\t\\r\\n\\f]+/g );\n\n\n\n// Convert String-formatted options into Object-formatted ones\nfunction createOptions( options ) {\n\tvar object = {};\n\tjQuery.each( options.match( rnothtmlwhite ) || [], function( _, flag ) {\n\t\tobject[ flag ] = true;\n\t} );\n\treturn object;\n}\n\n/*\n * Create a callback list using the following parameters:\n *\n *\toptions: an optional list of space-separated options that will change how\n *\t\t\tthe callback list behaves or a more traditional option object\n *\n * By default a callback list will act like an event callback list and can be\n * \"fired\" multiple times.\n *\n * Possible options:\n *\n *\tonce:\t\t\twill ensure the callback list can only be fired once (like a Deferred)\n *\n *\tmemory:\t\t\twill keep track of previous values and will call any callback added\n *\t\t\t\t\tafter the list has been fired right away with the latest \"memorized\"\n *\t\t\t\t\tvalues (like a Deferred)\n *\n *\tunique:\t\t\twill ensure a callback can only be added once (no duplicate in the list)\n *\n *\tstopOnFalse:\tinterrupt callings when a callback returns false\n *\n */\njQuery.Callbacks = function( options ) {\n\n\t// Convert options from String-formatted to Object-formatted if needed\n\t// (we check in cache first)\n\toptions = typeof options === \"string\" ?\n\t\tcreateOptions( options ) :\n\t\tjQuery.extend( {}, options );\n\n\tvar // Flag to know if list is currently firing\n\t\tfiring,\n\n\t\t// Last fire value for non-forgettable lists\n\t\tmemory,\n\n\t\t// Flag to know if list was already fired\n\t\tfired,\n\n\t\t// Flag to prevent firing\n\t\tlocked,\n\n\t\t// Actual callback list\n\t\tlist = [],\n\n\t\t// Queue of execution data for repeatable lists\n\t\tqueue = [],\n\n\t\t// Index of currently firing callback (modified by add/remove as needed)\n\t\tfiringIndex = -1,\n\n\t\t// Fire callbacks\n\t\tfire = function() {\n\n\t\t\t// Enforce single-firing\n\t\t\tlocked = locked || options.once;\n\n\t\t\t// Execute callbacks for all pending executions,\n\t\t\t// respecting firingIndex overrides and runtime changes\n\t\t\tfired = firing = true;\n\t\t\tfor ( ; queue.length; firingIndex = -1 ) {\n\t\t\t\tmemory = queue.shift();\n\t\t\t\twhile ( ++firingIndex < list.length ) {\n\n\t\t\t\t\t// Run callback and check for early termination\n\t\t\t\t\tif ( list[ firingIndex ].apply( memory[ 0 ], memory[ 1 ] ) === false &&\n\t\t\t\t\t\toptions.stopOnFalse ) {\n\n\t\t\t\t\t\t// Jump to end and forget the data so .add doesn't re-fire\n\t\t\t\t\t\tfiringIndex = list.length;\n\t\t\t\t\t\tmemory = false;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Forget the data if we're done with it\n\t\t\tif ( !options.memory ) {\n\t\t\t\tmemory = false;\n\t\t\t}\n\n\t\t\tfiring = false;\n\n\t\t\t// Clean up if we're done firing for good\n\t\t\tif ( locked ) {\n\n\t\t\t\t// Keep an empty list if we have data for future add calls\n\t\t\t\tif ( memory ) {\n\t\t\t\t\tlist = [];\n\n\t\t\t\t// Otherwise, this object is spent\n\t\t\t\t} else {\n\t\t\t\t\tlist = \"\";\n\t\t\t\t}\n\t\t\t}\n\t\t},\n\n\t\t// Actual Callbacks object\n\t\tself = {\n\n\t\t\t// Add a callback or a collection of callbacks to the list\n\t\t\tadd: function() {\n\t\t\t\tif ( list ) {\n\n\t\t\t\t\t// If we have memory from a past run, we should fire after adding\n\t\t\t\t\tif ( memory && !firing ) {\n\t\t\t\t\t\tfiringIndex = list.length - 1;\n\t\t\t\t\t\tqueue.push( memory );\n\t\t\t\t\t}\n\n\t\t\t\t\t( function add( args ) {\n\t\t\t\t\t\tjQuery.each( args, function( _, arg ) {\n\t\t\t\t\t\t\tif ( jQuery.isFunction( arg ) ) {\n\t\t\t\t\t\t\t\tif ( !options.unique || !self.has( arg ) ) {\n\t\t\t\t\t\t\t\t\tlist.push( arg );\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} else if ( arg && arg.length && jQuery.type( arg ) !== \"string\" ) {\n\n\t\t\t\t\t\t\t\t// Inspect recursively\n\t\t\t\t\t\t\t\tadd( arg );\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t} );\n\t\t\t\t\t} )( arguments );\n\n\t\t\t\t\tif ( memory && !firing ) {\n\t\t\t\t\t\tfire();\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\treturn this;\n\t\t\t},\n\n\t\t\t// Remove a callback from the list\n\t\t\tremove: function() {\n\t\t\t\tjQuery.each( arguments, function( _, arg ) {\n\t\t\t\t\tvar index;\n\t\t\t\t\twhile ( ( index = jQuery.inArray( arg, list, index ) ) > -1 ) {\n\t\t\t\t\t\tlist.splice( index, 1 );\n\n\t\t\t\t\t\t// Handle firing indexes\n\t\t\t\t\t\tif ( index <= firingIndex ) {\n\t\t\t\t\t\t\tfiringIndex--;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t} );\n\t\t\t\treturn this;\n\t\t\t},\n\n\t\t\t// Check if a given callback is in the list.\n\t\t\t// If no argument is given, return whether or not list has callbacks attached.\n\t\t\thas: function( fn ) {\n\t\t\t\treturn fn ?\n\t\t\t\t\tjQuery.inArray( fn, list ) > -1 :\n\t\t\t\t\tlist.length > 0;\n\t\t\t},\n\n\t\t\t// Remove all callbacks from the list\n\t\t\tempty: function() {\n\t\t\t\tif ( list ) {\n\t\t\t\t\tlist = [];\n\t\t\t\t}\n\t\t\t\treturn this;\n\t\t\t},\n\n\t\t\t// Disable .fire and .add\n\t\t\t// Abort any current/pending executions\n\t\t\t// Clear all callbacks and values\n\t\t\tdisable: function() {\n\t\t\t\tlocked = queue = [];\n\t\t\t\tlist = memory = \"\";\n\t\t\t\treturn this;\n\t\t\t},\n\t\t\tdisabled: function() {\n\t\t\t\treturn !list;\n\t\t\t},\n\n\t\t\t// Disable .fire\n\t\t\t// Also disable .add unless we have memory (since it would have no effect)\n\t\t\t// Abort any pending executions\n\t\t\tlock: function() {\n\t\t\t\tlocked = queue = [];\n\t\t\t\tif ( !memory && !firing ) {\n\t\t\t\t\tlist = memory = \"\";\n\t\t\t\t}\n\t\t\t\treturn this;\n\t\t\t},\n\t\t\tlocked: function() {\n\t\t\t\treturn !!locked;\n\t\t\t},\n\n\t\t\t// Call all callbacks with the given context and arguments\n\t\t\tfireWith: function( context, args ) {\n\t\t\t\tif ( !locked ) {\n\t\t\t\t\targs = args || [];\n\t\t\t\t\targs = [ context, args.slice ? args.slice() : args ];\n\t\t\t\t\tqueue.push( args );\n\t\t\t\t\tif ( !firing ) {\n\t\t\t\t\t\tfire();\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\treturn this;\n\t\t\t},\n\n\t\t\t// Call all the callbacks with the given arguments\n\t\t\tfire: function() {\n\t\t\t\tself.fireWith( this, arguments );\n\t\t\t\treturn this;\n\t\t\t},\n\n\t\t\t// To know if the callbacks have already been called at least once\n\t\t\tfired: function() {\n\t\t\t\treturn !!fired;\n\t\t\t}\n\t\t};\n\n\treturn self;\n};\n\n\nfunction Identity( v ) {\n\treturn v;\n}\nfunction Thrower( ex ) {\n\tthrow ex;\n}\n\nfunction adoptValue( value, resolve, reject, noValue ) {\n\tvar method;\n\n\ttry {\n\n\t\t// Check for promise aspect first to privilege synchronous behavior\n\t\tif ( value && jQuery.isFunction( ( method = value.promise ) ) ) {\n\t\t\tmethod.call( value ).done( resolve ).fail( reject );\n\n\t\t// Other thenables\n\t\t} else if ( value && jQuery.isFunction( ( method = value.then ) ) ) {\n\t\t\tmethod.call( value, resolve, reject );\n\n\t\t// Other non-thenables\n\t\t} else {\n\n\t\t\t// Control `resolve` arguments by letting Array#slice cast boolean `noValue` to integer:\n\t\t\t// * false: [ value ].slice( 0 ) => resolve( value )\n\t\t\t// * true: [ value ].slice( 1 ) => resolve()\n\t\t\tresolve.apply( undefined, [ value ].slice( noValue ) );\n\t\t}\n\n\t// For Promises/A+, convert exceptions into rejections\n\t// Since jQuery.when doesn't unwrap thenables, we can skip the extra checks appearing in\n\t// Deferred#then to conditionally suppress rejection.\n\t} catch ( value ) {\n\n\t\t// Support: Android 4.0 only\n\t\t// Strict mode functions invoked without .call/.apply get global-object context\n\t\treject.apply( undefined, [ value ] );\n\t}\n}\n\njQuery.extend( {\n\n\tDeferred: function( func ) {\n\t\tvar tuples = [\n\n\t\t\t\t// action, add listener, callbacks,\n\t\t\t\t// ... .then handlers, argument index, [final state]\n\t\t\t\t[ \"notify\", \"progress\", jQuery.Callbacks( \"memory\" ),\n\t\t\t\t\tjQuery.Callbacks( \"memory\" ), 2 ],\n\t\t\t\t[ \"resolve\", \"done\", jQuery.Callbacks( \"once memory\" ),\n\t\t\t\t\tjQuery.Callbacks( \"once memory\" ), 0, \"resolved\" ],\n\t\t\t\t[ \"reject\", \"fail\", jQuery.Callbacks( \"once memory\" ),\n\t\t\t\t\tjQuery.Callbacks( \"once memory\" ), 1, \"rejected\" ]\n\t\t\t],\n\t\t\tstate = \"pending\",\n\t\t\tpromise = {\n\t\t\t\tstate: function() {\n\t\t\t\t\treturn state;\n\t\t\t\t},\n\t\t\t\talways: function() {\n\t\t\t\t\tdeferred.done( arguments ).fail( arguments );\n\t\t\t\t\treturn this;\n\t\t\t\t},\n\t\t\t\t\"catch\": function( fn ) {\n\t\t\t\t\treturn promise.then( null, fn );\n\t\t\t\t},\n\n\t\t\t\t// Keep pipe for back-compat\n\t\t\t\tpipe: function( /* fnDone, fnFail, fnProgress */ ) {\n\t\t\t\t\tvar fns = arguments;\n\n\t\t\t\t\treturn jQuery.Deferred( function( newDefer ) {\n\t\t\t\t\t\tjQuery.each( tuples, function( i, tuple ) {\n\n\t\t\t\t\t\t\t// Map tuples (progress, done, fail) to arguments (done, fail, progress)\n\t\t\t\t\t\t\tvar fn = jQuery.isFunction( fns[ tuple[ 4 ] ] ) && fns[ tuple[ 4 ] ];\n\n\t\t\t\t\t\t\t// deferred.progress(function() { bind to newDefer or newDefer.notify })\n\t\t\t\t\t\t\t// deferred.done(function() { bind to newDefer or newDefer.resolve })\n\t\t\t\t\t\t\t// deferred.fail(function() { bind to newDefer or newDefer.reject })\n\t\t\t\t\t\t\tdeferred[ tuple[ 1 ] ]( function() {\n\t\t\t\t\t\t\t\tvar returned = fn && fn.apply( this, arguments );\n\t\t\t\t\t\t\t\tif ( returned && jQuery.isFunction( returned.promise ) ) {\n\t\t\t\t\t\t\t\t\treturned.promise()\n\t\t\t\t\t\t\t\t\t\t.progress( newDefer.notify )\n\t\t\t\t\t\t\t\t\t\t.done( newDefer.resolve )\n\t\t\t\t\t\t\t\t\t\t.fail( newDefer.reject );\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tnewDefer[ tuple[ 0 ] + \"With\" ](\n\t\t\t\t\t\t\t\t\t\tthis,\n\t\t\t\t\t\t\t\t\t\tfn ? [ returned ] : arguments\n\t\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} );\n\t\t\t\t\t\t} );\n\t\t\t\t\t\tfns = null;\n\t\t\t\t\t} ).promise();\n\t\t\t\t},\n\t\t\t\tthen: function( onFulfilled, onRejected, onProgress ) {\n\t\t\t\t\tvar maxDepth = 0;\n\t\t\t\t\tfunction resolve( depth, deferred, handler, special ) {\n\t\t\t\t\t\treturn function() {\n\t\t\t\t\t\t\tvar that = this,\n\t\t\t\t\t\t\t\targs = arguments,\n\t\t\t\t\t\t\t\tmightThrow = function() {\n\t\t\t\t\t\t\t\t\tvar returned, then;\n\n\t\t\t\t\t\t\t\t\t// Support: Promises/A+ section 2.3.3.3.3\n\t\t\t\t\t\t\t\t\t// https://promisesaplus.com/#point-59\n\t\t\t\t\t\t\t\t\t// Ignore double-resolution attempts\n\t\t\t\t\t\t\t\t\tif ( depth < maxDepth ) {\n\t\t\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\treturned = handler.apply( that, args );\n\n\t\t\t\t\t\t\t\t\t// Support: Promises/A+ section 2.3.1\n\t\t\t\t\t\t\t\t\t// https://promisesaplus.com/#point-48\n\t\t\t\t\t\t\t\t\tif ( returned === deferred.promise() ) {\n\t\t\t\t\t\t\t\t\t\tthrow new TypeError( \"Thenable self-resolution\" );\n\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t// Support: Promises/A+ sections 2.3.3.1, 3.5\n\t\t\t\t\t\t\t\t\t// https://promisesaplus.com/#point-54\n\t\t\t\t\t\t\t\t\t// https://promisesaplus.com/#point-75\n\t\t\t\t\t\t\t\t\t// Retrieve `then` only once\n\t\t\t\t\t\t\t\t\tthen = returned &&\n\n\t\t\t\t\t\t\t\t\t\t// Support: Promises/A+ section 2.3.4\n\t\t\t\t\t\t\t\t\t\t// https://promisesaplus.com/#point-64\n\t\t\t\t\t\t\t\t\t\t// Only check objects and functions for thenability\n\t\t\t\t\t\t\t\t\t\t( typeof returned === \"object\" ||\n\t\t\t\t\t\t\t\t\t\t\ttypeof returned === \"function\" ) &&\n\t\t\t\t\t\t\t\t\t\treturned.then;\n\n\t\t\t\t\t\t\t\t\t// Handle a returned thenable\n\t\t\t\t\t\t\t\t\tif ( jQuery.isFunction( then ) ) {\n\n\t\t\t\t\t\t\t\t\t\t// Special processors (notify) just wait for resolution\n\t\t\t\t\t\t\t\t\t\tif ( special ) {\n\t\t\t\t\t\t\t\t\t\t\tthen.call(\n\t\t\t\t\t\t\t\t\t\t\t\treturned,\n\t\t\t\t\t\t\t\t\t\t\t\tresolve( maxDepth, deferred, Identity, special ),\n\t\t\t\t\t\t\t\t\t\t\t\tresolve( maxDepth, deferred, Thrower, special )\n\t\t\t\t\t\t\t\t\t\t\t);\n\n\t\t\t\t\t\t\t\t\t\t// Normal processors (resolve) also hook into progress\n\t\t\t\t\t\t\t\t\t\t} else {\n\n\t\t\t\t\t\t\t\t\t\t\t// ...and disregard older resolution values\n\t\t\t\t\t\t\t\t\t\t\tmaxDepth++;\n\n\t\t\t\t\t\t\t\t\t\t\tthen.call(\n\t\t\t\t\t\t\t\t\t\t\t\treturned,\n\t\t\t\t\t\t\t\t\t\t\t\tresolve( maxDepth, deferred, Identity, special ),\n\t\t\t\t\t\t\t\t\t\t\t\tresolve( maxDepth, deferred, Thrower, special ),\n\t\t\t\t\t\t\t\t\t\t\t\tresolve( maxDepth, deferred, Identity,\n\t\t\t\t\t\t\t\t\t\t\t\t\tdeferred.notifyWith )\n\t\t\t\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t// Handle all other returned values\n\t\t\t\t\t\t\t\t\t} else {\n\n\t\t\t\t\t\t\t\t\t\t// Only substitute handlers pass on context\n\t\t\t\t\t\t\t\t\t\t// and multiple values (non-spec behavior)\n\t\t\t\t\t\t\t\t\t\tif ( handler !== Identity ) {\n\t\t\t\t\t\t\t\t\t\t\tthat = undefined;\n\t\t\t\t\t\t\t\t\t\t\targs = [ returned ];\n\t\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t\t// Process the value(s)\n\t\t\t\t\t\t\t\t\t\t// Default process is resolve\n\t\t\t\t\t\t\t\t\t\t( special || deferred.resolveWith )( that, args );\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t},\n\n\t\t\t\t\t\t\t\t// Only normal processors (resolve) catch and reject exceptions\n\t\t\t\t\t\t\t\tprocess = special ?\n\t\t\t\t\t\t\t\t\tmightThrow :\n\t\t\t\t\t\t\t\t\tfunction() {\n\t\t\t\t\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\t\t\t\t\tmightThrow();\n\t\t\t\t\t\t\t\t\t\t} catch ( e ) {\n\n\t\t\t\t\t\t\t\t\t\t\tif ( jQuery.Deferred.exceptionHook ) {\n\t\t\t\t\t\t\t\t\t\t\t\tjQuery.Deferred.exceptionHook( e,\n\t\t\t\t\t\t\t\t\t\t\t\t\tprocess.stackTrace );\n\t\t\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t\t\t// Support: Promises/A+ section 2.3.3.3.4.1\n\t\t\t\t\t\t\t\t\t\t\t// https://promisesaplus.com/#point-61\n\t\t\t\t\t\t\t\t\t\t\t// Ignore post-resolution exceptions\n\t\t\t\t\t\t\t\t\t\t\tif ( depth + 1 >= maxDepth ) {\n\n\t\t\t\t\t\t\t\t\t\t\t\t// Only substitute handlers pass on context\n\t\t\t\t\t\t\t\t\t\t\t\t// and multiple values (non-spec behavior)\n\t\t\t\t\t\t\t\t\t\t\t\tif ( handler !== Thrower ) {\n\t\t\t\t\t\t\t\t\t\t\t\t\tthat = undefined;\n\t\t\t\t\t\t\t\t\t\t\t\t\targs = [ e ];\n\t\t\t\t\t\t\t\t\t\t\t\t}\n\n\t\t\t\t\t\t\t\t\t\t\t\tdeferred.rejectWith( that, args );\n\t\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\t};\n\n\t\t\t\t\t\t\t// Support: Promises/A+ section 2.3.3.3.1\n\t\t\t\t\t\t\t// https://promisesaplus.com/#point-57\n\t\t\t\t\t\t\t// Re-resolve promises immediately to dodge false rejection from\n\t\t\t\t\t\t\t// subsequent errors\n\t\t\t\t\t\t\tif ( depth ) {\n\t\t\t\t\t\t\t\tprocess();\n\t\t\t\t\t\t\t} else {\n\n\t\t\t\t\t\t\t\t// Call an optional hook to record the stack, in case of exception\n\t\t\t\t\t\t\t\t// since it's otherwise lost when execution goes async\n\t\t\t\t\t\t\t\tif ( jQuery.Deferred.getStackHook ) {\n\t\t\t\t\t\t\t\t\tprocess.stackTrace = jQuery.Deferred.getStackHook();\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\twindow.setTimeout( process );\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t};\n\t\t\t\t\t}\n\n\t\t\t\t\treturn jQuery.Deferred( function( newDefer ) {\n\n\t\t\t\t\t\t// progress_handlers.add( ... )\n\t\t\t\t\t\ttuples[ 0 ][ 3 ].add(\n\t\t\t\t\t\t\tresolve(\n\t\t\t\t\t\t\t\t0,\n\t\t\t\t\t\t\t\tnewDefer,\n\t\t\t\t\t\t\t\tjQuery.isFunction( onProgress ) ?\n\t\t\t\t\t\t\t\t\tonProgress :\n\t\t\t\t\t\t\t\t\tIdentity,\n\t\t\t\t\t\t\t\tnewDefer.notifyWith\n\t\t\t\t\t\t\t)\n\t\t\t\t\t\t);\n\n\t\t\t\t\t\t// fulfilled_handlers.add( ... )\n\t\t\t\t\t\ttuples[ 1 ][ 3 ].add(\n\t\t\t\t\t\t\tresolve(\n\t\t\t\t\t\t\t\t0,\n\t\t\t\t\t\t\t\tnewDefer,\n\t\t\t\t\t\t\t\tjQuery.isFunction( onFulfilled ) ?\n\t\t\t\t\t\t\t\t\tonFulfilled :\n\t\t\t\t\t\t\t\t\tIdentity\n\t\t\t\t\t\t\t)\n\t\t\t\t\t\t);\n\n\t\t\t\t\t\t// rejected_handlers.add( ... )\n\t\t\t\t\t\ttuples[ 2 ][ 3 ].add(\n\t\t\t\t\t\t\tresolve(\n\t\t\t\t\t\t\t\t0,\n\t\t\t\t\t\t\t\tnewDefer,\n\t\t\t\t\t\t\t\tjQuery.isFunction( onRejected ) ?\n\t\t\t\t\t\t\t\t\tonRejected :\n\t\t\t\t\t\t\t\t\tThrower\n\t\t\t\t\t\t\t)\n\t\t\t\t\t\t);\n\t\t\t\t\t} ).promise();\n\t\t\t\t},\n\n\t\t\t\t// Get a promise for this deferred\n\t\t\t\t// If obj is provided, the promise aspect is added to the object\n\t\t\t\tpromise: function( obj ) {\n\t\t\t\t\treturn obj != null ? jQuery.extend( obj, promise ) : promise;\n\t\t\t\t}\n\t\t\t},\n\t\t\tdeferred = {};\n\n\t\t// Add list-specific methods\n\t\tjQuery.each( tuples, function( i, tuple ) {\n\t\t\tvar list = tuple[ 2 ],\n\t\t\t\tstateString = tuple[ 5 ];\n\n\t\t\t// promise.progress = list.add\n\t\t\t// promise.done = list.add\n\t\t\t// promise.fail = list.add\n\t\t\tpromise[ tuple[ 1 ] ] = list.add;\n\n\t\t\t// Handle state\n\t\t\tif ( stateString ) {\n\t\t\t\tlist.add(\n\t\t\t\t\tfunction() {\n\n\t\t\t\t\t\t// state = \"resolved\" (i.e., fulfilled)\n\t\t\t\t\t\t// state = \"rejected\"\n\t\t\t\t\t\tstate = stateString;\n\t\t\t\t\t},\n\n\t\t\t\t\t// rejected_callbacks.disable\n\t\t\t\t\t// fulfilled_callbacks.disable\n\t\t\t\t\ttuples[ 3 - i ][ 2 ].disable,\n\n\t\t\t\t\t// progress_callbacks.lock\n\t\t\t\t\ttuples[ 0 ][ 2 ].lock\n\t\t\t\t);\n\t\t\t}\n\n\t\t\t// progress_handlers.fire\n\t\t\t// fulfilled_handlers.fire\n\t\t\t// rejected_handlers.fire\n\t\t\tlist.add( tuple[ 3 ].fire );\n\n\t\t\t// deferred.notify = function() { deferred.notifyWith(...) }\n\t\t\t// deferred.resolve = function() { deferred.resolveWith(...) }\n\t\t\t// deferred.reject = function() { deferred.rejectWith(...) }\n\t\t\tdeferred[ tuple[ 0 ] ] = function() {\n\t\t\t\tdeferred[ tuple[ 0 ] + \"With\" ]( this === deferred ? undefined : this, arguments );\n\t\t\t\treturn this;\n\t\t\t};\n\n\t\t\t// deferred.notifyWith = list.fireWith\n\t\t\t// deferred.resolveWith = list.fireWith\n\t\t\t// deferred.rejectWith = list.fireWith\n\t\t\tdeferred[ tuple[ 0 ] + \"With\" ] = list.fireWith;\n\t\t} );\n\n\t\t// Make the deferred a promise\n\t\tpromise.promise( deferred );\n\n\t\t// Call given func if any\n\t\tif ( func ) {\n\t\t\tfunc.call( deferred, deferred );\n\t\t}\n\n\t\t// All done!\n\t\treturn deferred;\n\t},\n\n\t// Deferred helper\n\twhen: function( singleValue ) {\n\t\tvar\n\n\t\t\t// count of uncompleted subordinates\n\t\t\tremaining = arguments.length,\n\n\t\t\t// count of unprocessed arguments\n\t\t\ti = remaining,\n\n\t\t\t// subordinate fulfillment data\n\t\t\tresolveContexts = Array( i ),\n\t\t\tresolveValues = slice.call( arguments ),\n\n\t\t\t// the master Deferred\n\t\t\tmaster = jQuery.Deferred(),\n\n\t\t\t// subordinate callback factory\n\t\t\tupdateFunc = function( i ) {\n\t\t\t\treturn function( value ) {\n\t\t\t\t\tresolveContexts[ i ] = this;\n\t\t\t\t\tresolveValues[ i ] = arguments.length > 1 ? slice.call( arguments ) : value;\n\t\t\t\t\tif ( !( --remaining ) ) {\n\t\t\t\t\t\tmaster.resolveWith( resolveContexts, resolveValues );\n\t\t\t\t\t}\n\t\t\t\t};\n\t\t\t};\n\n\t\t// Single- and empty arguments are adopted like Promise.resolve\n\t\tif ( remaining <= 1 ) {\n\t\t\tadoptValue( singleValue, master.done( updateFunc( i ) ).resolve, master.reject,\n\t\t\t\t!remaining );\n\n\t\t\t// Use .then() to unwrap secondary thenables (cf. gh-3000)\n\t\t\tif ( master.state() === \"pending\" ||\n\t\t\t\tjQuery.isFunction( resolveValues[ i ] && resolveValues[ i ].then ) ) {\n\n\t\t\t\treturn master.then();\n\t\t\t}\n\t\t}\n\n\t\t// Multiple arguments are aggregated like Promise.all array elements\n\t\twhile ( i-- ) {\n\t\t\tadoptValue( resolveValues[ i ], updateFunc( i ), master.reject );\n\t\t}\n\n\t\treturn master.promise();\n\t}\n} );\n\n\n// These usually indicate a programmer mistake during development,\n// warn about them ASAP rather than swallowing them by default.\nvar rerrorNames = /^(Eval|Internal|Range|Reference|Syntax|Type|URI)Error$/;\n\njQuery.Deferred.exceptionHook = function( error, stack ) {\n\n\t// Support: IE 8 - 9 only\n\t// Console exists when dev tools are open, which can happen at any time\n\tif ( window.console && window.console.warn && error && rerrorNames.test( error.name ) ) {\n\t\twindow.console.warn( \"jQuery.Deferred exception: \" + error.message, error.stack, stack );\n\t}\n};\n\n\n\n\njQuery.readyException = function( error ) {\n\twindow.setTimeout( function() {\n\t\tthrow error;\n\t} );\n};\n\n\n\n\n// The deferred used on DOM ready\nvar readyList = jQuery.Deferred();\n\njQuery.fn.ready = function( fn ) {\n\n\treadyList\n\t\t.then( fn )\n\n\t\t// Wrap jQuery.readyException in a function so that the lookup\n\t\t// happens at the time of error handling instead of callback\n\t\t// registration.\n\t\t.catch( function( error ) {\n\t\t\tjQuery.readyException( error );\n\t\t} );\n\n\treturn this;\n};\n\njQuery.extend( {\n\n\t// Is the DOM ready to be used? Set to true once it occurs.\n\tisReady: false,\n\n\t// A counter to track how many items to wait for before\n\t// the ready event fires. See #6781\n\treadyWait: 1,\n\n\t// Handle when the DOM is ready\n\tready: function( wait ) {\n\n\t\t// Abort if there are pending holds or we're already ready\n\t\tif ( wait === true ? --jQuery.readyWait : jQuery.isReady ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// Remember that the DOM is ready\n\t\tjQuery.isReady = true;\n\n\t\t// If a normal DOM Ready event fired, decrement, and wait if need be\n\t\tif ( wait !== true && --jQuery.readyWait > 0 ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// If there are functions bound, to execute\n\t\treadyList.resolveWith( document, [ jQuery ] );\n\t}\n} );\n\njQuery.ready.then = readyList.then;\n\n// The ready event handler and self cleanup method\nfunction completed() {\n\tdocument.removeEventListener( \"DOMContentLoaded\", completed );\n\twindow.removeEventListener( \"load\", completed );\n\tjQuery.ready();\n}\n\n// Catch cases where $(document).ready() is called\n// after the browser event has already occurred.\n// Support: IE <=9 - 10 only\n// Older IE sometimes signals \"interactive\" too soon\nif ( document.readyState === \"complete\" ||\n\t( document.readyState !== \"loading\" && !document.documentElement.doScroll ) ) {\n\n\t// Handle it asynchronously to allow scripts the opportunity to delay ready\n\twindow.setTimeout( jQuery.ready );\n\n} else {\n\n\t// Use the handy event callback\n\tdocument.addEventListener( \"DOMContentLoaded\", completed );\n\n\t// A fallback to window.onload, that will always work\n\twindow.addEventListener( \"load\", completed );\n}\n\n\n\n\n// Multifunctional method to get and set values of a collection\n// The value/s can optionally be executed if it's a function\nvar access = function( elems, fn, key, value, chainable, emptyGet, raw ) {\n\tvar i = 0,\n\t\tlen = elems.length,\n\t\tbulk = key == null;\n\n\t// Sets many values\n\tif ( jQuery.type( key ) === \"object\" ) {\n\t\tchainable = true;\n\t\tfor ( i in key ) {\n\t\t\taccess( elems, fn, i, key[ i ], true, emptyGet, raw );\n\t\t}\n\n\t// Sets one value\n\t} else if ( value !== undefined ) {\n\t\tchainable = true;\n\n\t\tif ( !jQuery.isFunction( value ) ) {\n\t\t\traw = true;\n\t\t}\n\n\t\tif ( bulk ) {\n\n\t\t\t// Bulk operations run against the entire set\n\t\t\tif ( raw ) {\n\t\t\t\tfn.call( elems, value );\n\t\t\t\tfn = null;\n\n\t\t\t// ...except when executing function values\n\t\t\t} else {\n\t\t\t\tbulk = fn;\n\t\t\t\tfn = function( elem, key, value ) {\n\t\t\t\t\treturn bulk.call( jQuery( elem ), value );\n\t\t\t\t};\n\t\t\t}\n\t\t}\n\n\t\tif ( fn ) {\n\t\t\tfor ( ; i < len; i++ ) {\n\t\t\t\tfn(\n\t\t\t\t\telems[ i ], key, raw ?\n\t\t\t\t\tvalue :\n\t\t\t\t\tvalue.call( elems[ i ], i, fn( elems[ i ], key ) )\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\t}\n\n\tif ( chainable ) {\n\t\treturn elems;\n\t}\n\n\t// Gets\n\tif ( bulk ) {\n\t\treturn fn.call( elems );\n\t}\n\n\treturn len ? fn( elems[ 0 ], key ) : emptyGet;\n};\nvar acceptData = function( owner ) {\n\n\t// Accepts only:\n\t// - Node\n\t// - Node.ELEMENT_NODE\n\t// - Node.DOCUMENT_NODE\n\t// - Object\n\t// - Any\n\treturn owner.nodeType === 1 || owner.nodeType === 9 || !( +owner.nodeType );\n};\n\n\n\n\nfunction Data() {\n\tthis.expando = jQuery.expando + Data.uid++;\n}\n\nData.uid = 1;\n\nData.prototype = {\n\n\tcache: function( owner ) {\n\n\t\t// Check if the owner object already has a cache\n\t\tvar value = owner[ this.expando ];\n\n\t\t// If not, create one\n\t\tif ( !value ) {\n\t\t\tvalue = {};\n\n\t\t\t// We can accept data for non-element nodes in modern browsers,\n\t\t\t// but we should not, see #8335.\n\t\t\t// Always return an empty object.\n\t\t\tif ( acceptData( owner ) ) {\n\n\t\t\t\t// If it is a node unlikely to be stringify-ed or looped over\n\t\t\t\t// use plain assignment\n\t\t\t\tif ( owner.nodeType ) {\n\t\t\t\t\towner[ this.expando ] = value;\n\n\t\t\t\t// Otherwise secure it in a non-enumerable property\n\t\t\t\t// configurable must be true to allow the property to be\n\t\t\t\t// deleted when data is removed\n\t\t\t\t} else {\n\t\t\t\t\tObject.defineProperty( owner, this.expando, {\n\t\t\t\t\t\tvalue: value,\n\t\t\t\t\t\tconfigurable: true\n\t\t\t\t\t} );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn value;\n\t},\n\tset: function( owner, data, value ) {\n\t\tvar prop,\n\t\t\tcache = this.cache( owner );\n\n\t\t// Handle: [ owner, key, value ] args\n\t\t// Always use camelCase key (gh-2257)\n\t\tif ( typeof data === \"string\" ) {\n\t\t\tcache[ jQuery.camelCase( data ) ] = value;\n\n\t\t// Handle: [ owner, { properties } ] args\n\t\t} else {\n\n\t\t\t// Copy the properties one-by-one to the cache object\n\t\t\tfor ( prop in data ) {\n\t\t\t\tcache[ jQuery.camelCase( prop ) ] = data[ prop ];\n\t\t\t}\n\t\t}\n\t\treturn cache;\n\t},\n\tget: function( owner, key ) {\n\t\treturn key === undefined ?\n\t\t\tthis.cache( owner ) :\n\n\t\t\t// Always use camelCase key (gh-2257)\n\t\t\towner[ this.expando ] && owner[ this.expando ][ jQuery.camelCase( key ) ];\n\t},\n\taccess: function( owner, key, value ) {\n\n\t\t// In cases where either:\n\t\t//\n\t\t// 1. No key was specified\n\t\t// 2. A string key was specified, but no value provided\n\t\t//\n\t\t// Take the \"read\" path and allow the get method to determine\n\t\t// which value to return, respectively either:\n\t\t//\n\t\t// 1. The entire cache object\n\t\t// 2. The data stored at the key\n\t\t//\n\t\tif ( key === undefined ||\n\t\t\t\t( ( key && typeof key === \"string\" ) && value === undefined ) ) {\n\n\t\t\treturn this.get( owner, key );\n\t\t}\n\n\t\t// When the key is not a string, or both a key and value\n\t\t// are specified, set or extend (existing objects) with either:\n\t\t//\n\t\t// 1. An object of properties\n\t\t// 2. A key and value\n\t\t//\n\t\tthis.set( owner, key, value );\n\n\t\t// Since the \"set\" path can have two possible entry points\n\t\t// return the expected data based on which path was taken[*]\n\t\treturn value !== undefined ? value : key;\n\t},\n\tremove: function( owner, key ) {\n\t\tvar i,\n\t\t\tcache = owner[ this.expando ];\n\n\t\tif ( cache === undefined ) {\n\t\t\treturn;\n\t\t}\n\n\t\tif ( key !== undefined ) {\n\n\t\t\t// Support array or space separated string of keys\n\t\t\tif ( Array.isArray( key ) ) {\n\n\t\t\t\t// If key is an array of keys...\n\t\t\t\t// We always set camelCase keys, so remove that.\n\t\t\t\tkey = key.map( jQuery.camelCase );\n\t\t\t} else {\n\t\t\t\tkey = jQuery.camelCase( key );\n\n\t\t\t\t// If a key with the spaces exists, use it.\n\t\t\t\t// Otherwise, create an array by matching non-whitespace\n\t\t\t\tkey = key in cache ?\n\t\t\t\t\t[ key ] :\n\t\t\t\t\t( key.match( rnothtmlwhite ) || [] );\n\t\t\t}\n\n\t\t\ti = key.length;\n\n\t\t\twhile ( i-- ) {\n\t\t\t\tdelete cache[ key[ i ] ];\n\t\t\t}\n\t\t}\n\n\t\t// Remove the expando if there's no more data\n\t\tif ( key === undefined || jQuery.isEmptyObject( cache ) ) {\n\n\t\t\t// Support: Chrome <=35 - 45\n\t\t\t// Webkit & Blink performance suffers when deleting properties\n\t\t\t// from DOM nodes, so set to undefined instead\n\t\t\t// https://bugs.chromium.org/p/chromium/issues/detail?id=378607 (bug restricted)\n\t\t\tif ( owner.nodeType ) {\n\t\t\t\towner[ this.expando ] = undefined;\n\t\t\t} else {\n\t\t\t\tdelete owner[ this.expando ];\n\t\t\t}\n\t\t}\n\t},\n\thasData: function( owner ) {\n\t\tvar cache = owner[ this.expando ];\n\t\treturn cache !== undefined && !jQuery.isEmptyObject( cache );\n\t}\n};\nvar dataPriv = new Data();\n\nvar dataUser = new Data();\n\n\n\n//\tImplementation Summary\n//\n//\t1. Enforce API surface and semantic compatibility with 1.9.x branch\n//\t2. Improve the module's maintainability by reducing the storage\n//\t\tpaths to a single mechanism.\n//\t3. Use the same single mechanism to support \"private\" and \"user\" data.\n//\t4. _Never_ expose \"private\" data to user code (TODO: Drop _data, _removeData)\n//\t5. Avoid exposing implementation details on user objects (eg. expando properties)\n//\t6. Provide a clear path for implementation upgrade to WeakMap in 2014\n\nvar rbrace = /^(?:\\{[\\w\\W]*\\}|\\[[\\w\\W]*\\])$/,\n\trmultiDash = /[A-Z]/g;\n\nfunction getData( data ) {\n\tif ( data === \"true\" ) {\n\t\treturn true;\n\t}\n\n\tif ( data === \"false\" ) {\n\t\treturn false;\n\t}\n\n\tif ( data === \"null\" ) {\n\t\treturn null;\n\t}\n\n\t// Only convert to a number if it doesn't change the string\n\tif ( data === +data + \"\" ) {\n\t\treturn +data;\n\t}\n\n\tif ( rbrace.test( data ) ) {\n\t\treturn JSON.parse( data );\n\t}\n\n\treturn data;\n}\n\nfunction dataAttr( elem, key, data ) {\n\tvar name;\n\n\t// If nothing was found internally, try to fetch any\n\t// data from the HTML5 data-* attribute\n\tif ( data === undefined && elem.nodeType === 1 ) {\n\t\tname = \"data-\" + key.replace( rmultiDash, \"-$&\" ).toLowerCase();\n\t\tdata = elem.getAttribute( name );\n\n\t\tif ( typeof data === \"string\" ) {\n\t\t\ttry {\n\t\t\t\tdata = getData( data );\n\t\t\t} catch ( e ) {}\n\n\t\t\t// Make sure we set the data so it isn't changed later\n\t\t\tdataUser.set( elem, key, data );\n\t\t} else {\n\t\t\tdata = undefined;\n\t\t}\n\t}\n\treturn data;\n}\n\njQuery.extend( {\n\thasData: function( elem ) {\n\t\treturn dataUser.hasData( elem ) || dataPriv.hasData( elem );\n\t},\n\n\tdata: function( elem, name, data ) {\n\t\treturn dataUser.access( elem, name, data );\n\t},\n\n\tremoveData: function( elem, name ) {\n\t\tdataUser.remove( elem, name );\n\t},\n\n\t// TODO: Now that all calls to _data and _removeData have been replaced\n\t// with direct calls to dataPriv methods, these can be deprecated.\n\t_data: function( elem, name, data ) {\n\t\treturn dataPriv.access( elem, name, data );\n\t},\n\n\t_removeData: function( elem, name ) {\n\t\tdataPriv.remove( elem, name );\n\t}\n} );\n\njQuery.fn.extend( {\n\tdata: function( key, value ) {\n\t\tvar i, name, data,\n\t\t\telem = this[ 0 ],\n\t\t\tattrs = elem && elem.attributes;\n\n\t\t// Gets all values\n\t\tif ( key === undefined ) {\n\t\t\tif ( this.length ) {\n\t\t\t\tdata = dataUser.get( elem );\n\n\t\t\t\tif ( elem.nodeType === 1 && !dataPriv.get( elem, \"hasDataAttrs\" ) ) {\n\t\t\t\t\ti = attrs.length;\n\t\t\t\t\twhile ( i-- ) {\n\n\t\t\t\t\t\t// Support: IE 11 only\n\t\t\t\t\t\t// The attrs elements can be null (#14894)\n\t\t\t\t\t\tif ( attrs[ i ] ) {\n\t\t\t\t\t\t\tname = attrs[ i ].name;\n\t\t\t\t\t\t\tif ( name.indexOf( \"data-\" ) === 0 ) {\n\t\t\t\t\t\t\t\tname = jQuery.camelCase( name.slice( 5 ) );\n\t\t\t\t\t\t\t\tdataAttr( elem, name, data[ name ] );\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tdataPriv.set( elem, \"hasDataAttrs\", true );\n\t\t\t\t}\n\t\t\t}\n\n\t\t\treturn data;\n\t\t}\n\n\t\t// Sets multiple values\n\t\tif ( typeof key === \"object\" ) {\n\t\t\treturn this.each( function() {\n\t\t\t\tdataUser.set( this, key );\n\t\t\t} );\n\t\t}\n\n\t\treturn access( this, function( value ) {\n\t\t\tvar data;\n\n\t\t\t// The calling jQuery object (element matches) is not empty\n\t\t\t// (and therefore has an element appears at this[ 0 ]) and the\n\t\t\t// `value` parameter was not undefined. An empty jQuery object\n\t\t\t// will result in `undefined` for elem = this[ 0 ] which will\n\t\t\t// throw an exception if an attempt to read a data cache is made.\n\t\t\tif ( elem && value === undefined ) {\n\n\t\t\t\t// Attempt to get data from the cache\n\t\t\t\t// The key will always be camelCased in Data\n\t\t\t\tdata = dataUser.get( elem, key );\n\t\t\t\tif ( data !== undefined ) {\n\t\t\t\t\treturn data;\n\t\t\t\t}\n\n\t\t\t\t// Attempt to \"discover\" the data in\n\t\t\t\t// HTML5 custom data-* attrs\n\t\t\t\tdata = dataAttr( elem, key );\n\t\t\t\tif ( data !== undefined ) {\n\t\t\t\t\treturn data;\n\t\t\t\t}\n\n\t\t\t\t// We tried really hard, but the data doesn't exist.\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// Set the data...\n\t\t\tthis.each( function() {\n\n\t\t\t\t// We always store the camelCased key\n\t\t\t\tdataUser.set( this, key, value );\n\t\t\t} );\n\t\t}, null, value, arguments.length > 1, null, true );\n\t},\n\n\tremoveData: function( key ) {\n\t\treturn this.each( function() {\n\t\t\tdataUser.remove( this, key );\n\t\t} );\n\t}\n} );\n\n\njQuery.extend( {\n\tqueue: function( elem, type, data ) {\n\t\tvar queue;\n\n\t\tif ( elem ) {\n\t\t\ttype = ( type || \"fx\" ) + \"queue\";\n\t\t\tqueue = dataPriv.get( elem, type );\n\n\t\t\t// Speed up dequeue by getting out quickly if this is just a lookup\n\t\t\tif ( data ) {\n\t\t\t\tif ( !queue || Array.isArray( data ) ) {\n\t\t\t\t\tqueue = dataPriv.access( elem, type, jQuery.makeArray( data ) );\n\t\t\t\t} else {\n\t\t\t\t\tqueue.push( data );\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn queue || [];\n\t\t}\n\t},\n\n\tdequeue: function( elem, type ) {\n\t\ttype = type || \"fx\";\n\n\t\tvar queue = jQuery.queue( elem, type ),\n\t\t\tstartLength = queue.length,\n\t\t\tfn = queue.shift(),\n\t\t\thooks = jQuery._queueHooks( elem, type ),\n\t\t\tnext = function() {\n\t\t\t\tjQuery.dequeue( elem, type );\n\t\t\t};\n\n\t\t// If the fx queue is dequeued, always remove the progress sentinel\n\t\tif ( fn === \"inprogress\" ) {\n\t\t\tfn = queue.shift();\n\t\t\tstartLength--;\n\t\t}\n\n\t\tif ( fn ) {\n\n\t\t\t// Add a progress sentinel to prevent the fx queue from being\n\t\t\t// automatically dequeued\n\t\t\tif ( type === \"fx\" ) {\n\t\t\t\tqueue.unshift( \"inprogress\" );\n\t\t\t}\n\n\t\t\t// Clear up the last queue stop function\n\t\t\tdelete hooks.stop;\n\t\t\tfn.call( elem, next, hooks );\n\t\t}\n\n\t\tif ( !startLength && hooks ) {\n\t\t\thooks.empty.fire();\n\t\t}\n\t},\n\n\t// Not public - generate a queueHooks object, or return the current one\n\t_queueHooks: function( elem, type ) {\n\t\tvar key = type + \"queueHooks\";\n\t\treturn dataPriv.get( elem, key ) || dataPriv.access( elem, key, {\n\t\t\tempty: jQuery.Callbacks( \"once memory\" ).add( function() {\n\t\t\t\tdataPriv.remove( elem, [ type + \"queue\", key ] );\n\t\t\t} )\n\t\t} );\n\t}\n} );\n\njQuery.fn.extend( {\n\tqueue: function( type, data ) {\n\t\tvar setter = 2;\n\n\t\tif ( typeof type !== \"string\" ) {\n\t\t\tdata = type;\n\t\t\ttype = \"fx\";\n\t\t\tsetter--;\n\t\t}\n\n\t\tif ( arguments.length < setter ) {\n\t\t\treturn jQuery.queue( this[ 0 ], type );\n\t\t}\n\n\t\treturn data === undefined ?\n\t\t\tthis :\n\t\t\tthis.each( function() {\n\t\t\t\tvar queue = jQuery.queue( this, type, data );\n\n\t\t\t\t// Ensure a hooks for this queue\n\t\t\t\tjQuery._queueHooks( this, type );\n\n\t\t\t\tif ( type === \"fx\" && queue[ 0 ] !== \"inprogress\" ) {\n\t\t\t\t\tjQuery.dequeue( this, type );\n\t\t\t\t}\n\t\t\t} );\n\t},\n\tdequeue: function( type ) {\n\t\treturn this.each( function() {\n\t\t\tjQuery.dequeue( this, type );\n\t\t} );\n\t},\n\tclearQueue: function( type ) {\n\t\treturn this.queue( type || \"fx\", [] );\n\t},\n\n\t// Get a promise resolved when queues of a certain type\n\t// are emptied (fx is the type by default)\n\tpromise: function( type, obj ) {\n\t\tvar tmp,\n\t\t\tcount = 1,\n\t\t\tdefer = jQuery.Deferred(),\n\t\t\telements = this,\n\t\t\ti = this.length,\n\t\t\tresolve = function() {\n\t\t\t\tif ( !( --count ) ) {\n\t\t\t\t\tdefer.resolveWith( elements, [ elements ] );\n\t\t\t\t}\n\t\t\t};\n\n\t\tif ( typeof type !== \"string\" ) {\n\t\t\tobj = type;\n\t\t\ttype = undefined;\n\t\t}\n\t\ttype = type || \"fx\";\n\n\t\twhile ( i-- ) {\n\t\t\ttmp = dataPriv.get( elements[ i ], type + \"queueHooks\" );\n\t\t\tif ( tmp && tmp.empty ) {\n\t\t\t\tcount++;\n\t\t\t\ttmp.empty.add( resolve );\n\t\t\t}\n\t\t}\n\t\tresolve();\n\t\treturn defer.promise( obj );\n\t}\n} );\nvar pnum = ( /[+-]?(?:\\d*\\.|)\\d+(?:[eE][+-]?\\d+|)/ ).source;\n\nvar rcssNum = new RegExp( \"^(?:([+-])=|)(\" + pnum + \")([a-z%]*)$\", \"i\" );\n\n\nvar cssExpand = [ \"Top\", \"Right\", \"Bottom\", \"Left\" ];\n\nvar isHiddenWithinTree = function( elem, el ) {\n\n\t\t// isHiddenWithinTree might be called from jQuery#filter function;\n\t\t// in that case, element will be second argument\n\t\telem = el || elem;\n\n\t\t// Inline style trumps all\n\t\treturn elem.style.display === \"none\" ||\n\t\t\telem.style.display === \"\" &&\n\n\t\t\t// Otherwise, check computed style\n\t\t\t// Support: Firefox <=43 - 45\n\t\t\t// Disconnected elements can have computed display: none, so first confirm that elem is\n\t\t\t// in the document.\n\t\t\tjQuery.contains( elem.ownerDocument, elem ) &&\n\n\t\t\tjQuery.css( elem, \"display\" ) === \"none\";\n\t};\n\nvar swap = function( elem, options, callback, args ) {\n\tvar ret, name,\n\t\told = {};\n\n\t// Remember the old values, and insert the new ones\n\tfor ( name in options ) {\n\t\told[ name ] = elem.style[ name ];\n\t\telem.style[ name ] = options[ name ];\n\t}\n\n\tret = callback.apply( elem, args || [] );\n\n\t// Revert the old values\n\tfor ( name in options ) {\n\t\telem.style[ name ] = old[ name ];\n\t}\n\n\treturn ret;\n};\n\n\n\n\nfunction adjustCSS( elem, prop, valueParts, tween ) {\n\tvar adjusted,\n\t\tscale = 1,\n\t\tmaxIterations = 20,\n\t\tcurrentValue = tween ?\n\t\t\tfunction() {\n\t\t\t\treturn tween.cur();\n\t\t\t} :\n\t\t\tfunction() {\n\t\t\t\treturn jQuery.css( elem, prop, \"\" );\n\t\t\t},\n\t\tinitial = currentValue(),\n\t\tunit = valueParts && valueParts[ 3 ] || ( jQuery.cssNumber[ prop ] ? \"\" : \"px\" ),\n\n\t\t// Starting value computation is required for potential unit mismatches\n\t\tinitialInUnit = ( jQuery.cssNumber[ prop ] || unit !== \"px\" && +initial ) &&\n\t\t\trcssNum.exec( jQuery.css( elem, prop ) );\n\n\tif ( initialInUnit && initialInUnit[ 3 ] !== unit ) {\n\n\t\t// Trust units reported by jQuery.css\n\t\tunit = unit || initialInUnit[ 3 ];\n\n\t\t// Make sure we update the tween properties later on\n\t\tvalueParts = valueParts || [];\n\n\t\t// Iteratively approximate from a nonzero starting point\n\t\tinitialInUnit = +initial || 1;\n\n\t\tdo {\n\n\t\t\t// If previous iteration zeroed out, double until we get *something*.\n\t\t\t// Use string for doubling so we don't accidentally see scale as unchanged below\n\t\t\tscale = scale || \".5\";\n\n\t\t\t// Adjust and apply\n\t\t\tinitialInUnit = initialInUnit / scale;\n\t\t\tjQuery.style( elem, prop, initialInUnit + unit );\n\n\t\t// Update scale, tolerating zero or NaN from tween.cur()\n\t\t// Break the loop if scale is unchanged or perfect, or if we've just had enough.\n\t\t} while (\n\t\t\tscale !== ( scale = currentValue() / initial ) && scale !== 1 && --maxIterations\n\t\t);\n\t}\n\n\tif ( valueParts ) {\n\t\tinitialInUnit = +initialInUnit || +initial || 0;\n\n\t\t// Apply relative offset (+=/-=) if specified\n\t\tadjusted = valueParts[ 1 ] ?\n\t\t\tinitialInUnit + ( valueParts[ 1 ] + 1 ) * valueParts[ 2 ] :\n\t\t\t+valueParts[ 2 ];\n\t\tif ( tween ) {\n\t\t\ttween.unit = unit;\n\t\t\ttween.start = initialInUnit;\n\t\t\ttween.end = adjusted;\n\t\t}\n\t}\n\treturn adjusted;\n}\n\n\nvar defaultDisplayMap = {};\n\nfunction getDefaultDisplay( elem ) {\n\tvar temp,\n\t\tdoc = elem.ownerDocument,\n\t\tnodeName = elem.nodeName,\n\t\tdisplay = defaultDisplayMap[ nodeName ];\n\n\tif ( display ) {\n\t\treturn display;\n\t}\n\n\ttemp = doc.body.appendChild( doc.createElement( nodeName ) );\n\tdisplay = jQuery.css( temp, \"display\" );\n\n\ttemp.parentNode.removeChild( temp );\n\n\tif ( display === \"none\" ) {\n\t\tdisplay = \"block\";\n\t}\n\tdefaultDisplayMap[ nodeName ] = display;\n\n\treturn display;\n}\n\nfunction showHide( elements, show ) {\n\tvar display, elem,\n\t\tvalues = [],\n\t\tindex = 0,\n\t\tlength = elements.length;\n\n\t// Determine new display value for elements that need to change\n\tfor ( ; index < length; index++ ) {\n\t\telem = elements[ index ];\n\t\tif ( !elem.style ) {\n\t\t\tcontinue;\n\t\t}\n\n\t\tdisplay = elem.style.display;\n\t\tif ( show ) {\n\n\t\t\t// Since we force visibility upon cascade-hidden elements, an immediate (and slow)\n\t\t\t// check is required in this first loop unless we have a nonempty display value (either\n\t\t\t// inline or about-to-be-restored)\n\t\t\tif ( display === \"none\" ) {\n\t\t\t\tvalues[ index ] = dataPriv.get( elem, \"display\" ) || null;\n\t\t\t\tif ( !values[ index ] ) {\n\t\t\t\t\telem.style.display = \"\";\n\t\t\t\t}\n\t\t\t}\n\t\t\tif ( elem.style.display === \"\" && isHiddenWithinTree( elem ) ) {\n\t\t\t\tvalues[ index ] = getDefaultDisplay( elem );\n\t\t\t}\n\t\t} else {\n\t\t\tif ( display !== \"none\" ) {\n\t\t\t\tvalues[ index ] = \"none\";\n\n\t\t\t\t// Remember what we're overwriting\n\t\t\t\tdataPriv.set( elem, \"display\", display );\n\t\t\t}\n\t\t}\n\t}\n\n\t// Set the display of the elements in a second loop to avoid constant reflow\n\tfor ( index = 0; index < length; index++ ) {\n\t\tif ( values[ index ] != null ) {\n\t\t\telements[ index ].style.display = values[ index ];\n\t\t}\n\t}\n\n\treturn elements;\n}\n\njQuery.fn.extend( {\n\tshow: function() {\n\t\treturn showHide( this, true );\n\t},\n\thide: function() {\n\t\treturn showHide( this );\n\t},\n\ttoggle: function( state ) {\n\t\tif ( typeof state === \"boolean\" ) {\n\t\t\treturn state ? this.show() : this.hide();\n\t\t}\n\n\t\treturn this.each( function() {\n\t\t\tif ( isHiddenWithinTree( this ) ) {\n\t\t\t\tjQuery( this ).show();\n\t\t\t} else {\n\t\t\t\tjQuery( this ).hide();\n\t\t\t}\n\t\t} );\n\t}\n} );\nvar rcheckableType = ( /^(?:checkbox|radio)$/i );\n\nvar rtagName = ( /<([a-z][^\\/\\0>\\x20\\t\\r\\n\\f]+)/i );\n\nvar rscriptType = ( /^$|\\/(?:java|ecma)script/i );\n\n\n\n// We have to close these tags to support XHTML (#13200)\nvar wrapMap = {\n\n\t// Support: IE <=9 only\n\toption: [ 1, \"\" ],\n\n\t// XHTML parsers do not magically insert elements in the\n\t// same way that tag soup parsers do. So we cannot shorten\n\t// this by omitting or other required elements.\n\tthead: [ 1, \"\", \"
\" ],\n\tcol: [ 2, \"\", \"
\" ],\n\ttr: [ 2, \"\", \"
\" ],\n\ttd: [ 3, \"\", \"
\" ],\n\n\t_default: [ 0, \"\", \"\" ]\n};\n\n// Support: IE <=9 only\nwrapMap.optgroup = wrapMap.option;\n\nwrapMap.tbody = wrapMap.tfoot = wrapMap.colgroup = wrapMap.caption = wrapMap.thead;\nwrapMap.th = wrapMap.td;\n\n\nfunction getAll( context, tag ) {\n\n\t// Support: IE <=9 - 11 only\n\t// Use typeof to avoid zero-argument method invocation on host objects (#15151)\n\tvar ret;\n\n\tif ( typeof context.getElementsByTagName !== \"undefined\" ) {\n\t\tret = context.getElementsByTagName( tag || \"*\" );\n\n\t} else if ( typeof context.querySelectorAll !== \"undefined\" ) {\n\t\tret = context.querySelectorAll( tag || \"*\" );\n\n\t} else {\n\t\tret = [];\n\t}\n\n\tif ( tag === undefined || tag && nodeName( context, tag ) ) {\n\t\treturn jQuery.merge( [ context ], ret );\n\t}\n\n\treturn ret;\n}\n\n\n// Mark scripts as having already been evaluated\nfunction setGlobalEval( elems, refElements ) {\n\tvar i = 0,\n\t\tl = elems.length;\n\n\tfor ( ; i < l; i++ ) {\n\t\tdataPriv.set(\n\t\t\telems[ i ],\n\t\t\t\"globalEval\",\n\t\t\t!refElements || dataPriv.get( refElements[ i ], \"globalEval\" )\n\t\t);\n\t}\n}\n\n\nvar rhtml = /<|&#?\\w+;/;\n\nfunction buildFragment( elems, context, scripts, selection, ignored ) {\n\tvar elem, tmp, tag, wrap, contains, j,\n\t\tfragment = context.createDocumentFragment(),\n\t\tnodes = [],\n\t\ti = 0,\n\t\tl = elems.length;\n\n\tfor ( ; i < l; i++ ) {\n\t\telem = elems[ i ];\n\n\t\tif ( elem || elem === 0 ) {\n\n\t\t\t// Add nodes directly\n\t\t\tif ( jQuery.type( elem ) === \"object\" ) {\n\n\t\t\t\t// Support: Android <=4.0 only, PhantomJS 1 only\n\t\t\t\t// push.apply(_, arraylike) throws on ancient WebKit\n\t\t\t\tjQuery.merge( nodes, elem.nodeType ? [ elem ] : elem );\n\n\t\t\t// Convert non-html into a text node\n\t\t\t} else if ( !rhtml.test( elem ) ) {\n\t\t\t\tnodes.push( context.createTextNode( elem ) );\n\n\t\t\t// Convert html into DOM nodes\n\t\t\t} else {\n\t\t\t\ttmp = tmp || fragment.appendChild( context.createElement( \"div\" ) );\n\n\t\t\t\t// Deserialize a standard representation\n\t\t\t\ttag = ( rtagName.exec( elem ) || [ \"\", \"\" ] )[ 1 ].toLowerCase();\n\t\t\t\twrap = wrapMap[ tag ] || wrapMap._default;\n\t\t\t\ttmp.innerHTML = wrap[ 1 ] + jQuery.htmlPrefilter( elem ) + wrap[ 2 ];\n\n\t\t\t\t// Descend through wrappers to the right content\n\t\t\t\tj = wrap[ 0 ];\n\t\t\t\twhile ( j-- ) {\n\t\t\t\t\ttmp = tmp.lastChild;\n\t\t\t\t}\n\n\t\t\t\t// Support: Android <=4.0 only, PhantomJS 1 only\n\t\t\t\t// push.apply(_, arraylike) throws on ancient WebKit\n\t\t\t\tjQuery.merge( nodes, tmp.childNodes );\n\n\t\t\t\t// Remember the top-level container\n\t\t\t\ttmp = fragment.firstChild;\n\n\t\t\t\t// Ensure the created nodes are orphaned (#12392)\n\t\t\t\ttmp.textContent = \"\";\n\t\t\t}\n\t\t}\n\t}\n\n\t// Remove wrapper from fragment\n\tfragment.textContent = \"\";\n\n\ti = 0;\n\twhile ( ( elem = nodes[ i++ ] ) ) {\n\n\t\t// Skip elements already in the context collection (trac-4087)\n\t\tif ( selection && jQuery.inArray( elem, selection ) > -1 ) {\n\t\t\tif ( ignored ) {\n\t\t\t\tignored.push( elem );\n\t\t\t}\n\t\t\tcontinue;\n\t\t}\n\n\t\tcontains = jQuery.contains( elem.ownerDocument, elem );\n\n\t\t// Append to fragment\n\t\ttmp = getAll( fragment.appendChild( elem ), \"script\" );\n\n\t\t// Preserve script evaluation history\n\t\tif ( contains ) {\n\t\t\tsetGlobalEval( tmp );\n\t\t}\n\n\t\t// Capture executables\n\t\tif ( scripts ) {\n\t\t\tj = 0;\n\t\t\twhile ( ( elem = tmp[ j++ ] ) ) {\n\t\t\t\tif ( rscriptType.test( elem.type || \"\" ) ) {\n\t\t\t\t\tscripts.push( elem );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\treturn fragment;\n}\n\n\n( function() {\n\tvar fragment = document.createDocumentFragment(),\n\t\tdiv = fragment.appendChild( document.createElement( \"div\" ) ),\n\t\tinput = document.createElement( \"input\" );\n\n\t// Support: Android 4.0 - 4.3 only\n\t// Check state lost if the name is set (#11217)\n\t// Support: Windows Web Apps (WWA)\n\t// `name` and `type` must use .setAttribute for WWA (#14901)\n\tinput.setAttribute( \"type\", \"radio\" );\n\tinput.setAttribute( \"checked\", \"checked\" );\n\tinput.setAttribute( \"name\", \"t\" );\n\n\tdiv.appendChild( input );\n\n\t// Support: Android <=4.1 only\n\t// Older WebKit doesn't clone checked state correctly in fragments\n\tsupport.checkClone = div.cloneNode( true ).cloneNode( true ).lastChild.checked;\n\n\t// Support: IE <=11 only\n\t// Make sure textarea (and checkbox) defaultValue is properly cloned\n\tdiv.innerHTML = \"\";\n\tsupport.noCloneChecked = !!div.cloneNode( true ).lastChild.defaultValue;\n} )();\nvar documentElement = document.documentElement;\n\n\n\nvar\n\trkeyEvent = /^key/,\n\trmouseEvent = /^(?:mouse|pointer|contextmenu|drag|drop)|click/,\n\trtypenamespace = /^([^.]*)(?:\\.(.+)|)/;\n\nfunction returnTrue() {\n\treturn true;\n}\n\nfunction returnFalse() {\n\treturn false;\n}\n\n// Support: IE <=9 only\n// See #13393 for more info\nfunction safeActiveElement() {\n\ttry {\n\t\treturn document.activeElement;\n\t} catch ( err ) { }\n}\n\nfunction on( elem, types, selector, data, fn, one ) {\n\tvar origFn, type;\n\n\t// Types can be a map of types/handlers\n\tif ( typeof types === \"object\" ) {\n\n\t\t// ( types-Object, selector, data )\n\t\tif ( typeof selector !== \"string\" ) {\n\n\t\t\t// ( types-Object, data )\n\t\t\tdata = data || selector;\n\t\t\tselector = undefined;\n\t\t}\n\t\tfor ( type in types ) {\n\t\t\ton( elem, type, selector, data, types[ type ], one );\n\t\t}\n\t\treturn elem;\n\t}\n\n\tif ( data == null && fn == null ) {\n\n\t\t// ( types, fn )\n\t\tfn = selector;\n\t\tdata = selector = undefined;\n\t} else if ( fn == null ) {\n\t\tif ( typeof selector === \"string\" ) {\n\n\t\t\t// ( types, selector, fn )\n\t\t\tfn = data;\n\t\t\tdata = undefined;\n\t\t} else {\n\n\t\t\t// ( types, data, fn )\n\t\t\tfn = data;\n\t\t\tdata = selector;\n\t\t\tselector = undefined;\n\t\t}\n\t}\n\tif ( fn === false ) {\n\t\tfn = returnFalse;\n\t} else if ( !fn ) {\n\t\treturn elem;\n\t}\n\n\tif ( one === 1 ) {\n\t\torigFn = fn;\n\t\tfn = function( event ) {\n\n\t\t\t// Can use an empty set, since event contains the info\n\t\t\tjQuery().off( event );\n\t\t\treturn origFn.apply( this, arguments );\n\t\t};\n\n\t\t// Use same guid so caller can remove using origFn\n\t\tfn.guid = origFn.guid || ( origFn.guid = jQuery.guid++ );\n\t}\n\treturn elem.each( function() {\n\t\tjQuery.event.add( this, types, fn, data, selector );\n\t} );\n}\n\n/*\n * Helper functions for managing events -- not part of the public interface.\n * Props to Dean Edwards' addEvent library for many of the ideas.\n */\njQuery.event = {\n\n\tglobal: {},\n\n\tadd: function( elem, types, handler, data, selector ) {\n\n\t\tvar handleObjIn, eventHandle, tmp,\n\t\t\tevents, t, handleObj,\n\t\t\tspecial, handlers, type, namespaces, origType,\n\t\t\telemData = dataPriv.get( elem );\n\n\t\t// Don't attach events to noData or text/comment nodes (but allow plain objects)\n\t\tif ( !elemData ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// Caller can pass in an object of custom data in lieu of the handler\n\t\tif ( handler.handler ) {\n\t\t\thandleObjIn = handler;\n\t\t\thandler = handleObjIn.handler;\n\t\t\tselector = handleObjIn.selector;\n\t\t}\n\n\t\t// Ensure that invalid selectors throw exceptions at attach time\n\t\t// Evaluate against documentElement in case elem is a non-element node (e.g., document)\n\t\tif ( selector ) {\n\t\t\tjQuery.find.matchesSelector( documentElement, selector );\n\t\t}\n\n\t\t// Make sure that the handler has a unique ID, used to find/remove it later\n\t\tif ( !handler.guid ) {\n\t\t\thandler.guid = jQuery.guid++;\n\t\t}\n\n\t\t// Init the element's event structure and main handler, if this is the first\n\t\tif ( !( events = elemData.events ) ) {\n\t\t\tevents = elemData.events = {};\n\t\t}\n\t\tif ( !( eventHandle = elemData.handle ) ) {\n\t\t\teventHandle = elemData.handle = function( e ) {\n\n\t\t\t\t// Discard the second event of a jQuery.event.trigger() and\n\t\t\t\t// when an event is called after a page has unloaded\n\t\t\t\treturn typeof jQuery !== \"undefined\" && jQuery.event.triggered !== e.type ?\n\t\t\t\t\tjQuery.event.dispatch.apply( elem, arguments ) : undefined;\n\t\t\t};\n\t\t}\n\n\t\t// Handle multiple events separated by a space\n\t\ttypes = ( types || \"\" ).match( rnothtmlwhite ) || [ \"\" ];\n\t\tt = types.length;\n\t\twhile ( t-- ) {\n\t\t\ttmp = rtypenamespace.exec( types[ t ] ) || [];\n\t\t\ttype = origType = tmp[ 1 ];\n\t\t\tnamespaces = ( tmp[ 2 ] || \"\" ).split( \".\" ).sort();\n\n\t\t\t// There *must* be a type, no attaching namespace-only handlers\n\t\t\tif ( !type ) {\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\t// If event changes its type, use the special event handlers for the changed type\n\t\t\tspecial = jQuery.event.special[ type ] || {};\n\n\t\t\t// If selector defined, determine special event api type, otherwise given type\n\t\t\ttype = ( selector ? special.delegateType : special.bindType ) || type;\n\n\t\t\t// Update special based on newly reset type\n\t\t\tspecial = jQuery.event.special[ type ] || {};\n\n\t\t\t// handleObj is passed to all event handlers\n\t\t\thandleObj = jQuery.extend( {\n\t\t\t\ttype: type,\n\t\t\t\torigType: origType,\n\t\t\t\tdata: data,\n\t\t\t\thandler: handler,\n\t\t\t\tguid: handler.guid,\n\t\t\t\tselector: selector,\n\t\t\t\tneedsContext: selector && jQuery.expr.match.needsContext.test( selector ),\n\t\t\t\tnamespace: namespaces.join( \".\" )\n\t\t\t}, handleObjIn );\n\n\t\t\t// Init the event handler queue if we're the first\n\t\t\tif ( !( handlers = events[ type ] ) ) {\n\t\t\t\thandlers = events[ type ] = [];\n\t\t\t\thandlers.delegateCount = 0;\n\n\t\t\t\t// Only use addEventListener if the special events handler returns false\n\t\t\t\tif ( !special.setup ||\n\t\t\t\t\tspecial.setup.call( elem, data, namespaces, eventHandle ) === false ) {\n\n\t\t\t\t\tif ( elem.addEventListener ) {\n\t\t\t\t\t\telem.addEventListener( type, eventHandle );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif ( special.add ) {\n\t\t\t\tspecial.add.call( elem, handleObj );\n\n\t\t\t\tif ( !handleObj.handler.guid ) {\n\t\t\t\t\thandleObj.handler.guid = handler.guid;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Add to the element's handler list, delegates in front\n\t\t\tif ( selector ) {\n\t\t\t\thandlers.splice( handlers.delegateCount++, 0, handleObj );\n\t\t\t} else {\n\t\t\t\thandlers.push( handleObj );\n\t\t\t}\n\n\t\t\t// Keep track of which events have ever been used, for event optimization\n\t\t\tjQuery.event.global[ type ] = true;\n\t\t}\n\n\t},\n\n\t// Detach an event or set of events from an element\n\tremove: function( elem, types, handler, selector, mappedTypes ) {\n\n\t\tvar j, origCount, tmp,\n\t\t\tevents, t, handleObj,\n\t\t\tspecial, handlers, type, namespaces, origType,\n\t\t\telemData = dataPriv.hasData( elem ) && dataPriv.get( elem );\n\n\t\tif ( !elemData || !( events = elemData.events ) ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// Once for each type.namespace in types; type may be omitted\n\t\ttypes = ( types || \"\" ).match( rnothtmlwhite ) || [ \"\" ];\n\t\tt = types.length;\n\t\twhile ( t-- ) {\n\t\t\ttmp = rtypenamespace.exec( types[ t ] ) || [];\n\t\t\ttype = origType = tmp[ 1 ];\n\t\t\tnamespaces = ( tmp[ 2 ] || \"\" ).split( \".\" ).sort();\n\n\t\t\t// Unbind all events (on this namespace, if provided) for the element\n\t\t\tif ( !type ) {\n\t\t\t\tfor ( type in events ) {\n\t\t\t\t\tjQuery.event.remove( elem, type + types[ t ], handler, selector, true );\n\t\t\t\t}\n\t\t\t\tcontinue;\n\t\t\t}\n\n\t\t\tspecial = jQuery.event.special[ type ] || {};\n\t\t\ttype = ( selector ? special.delegateType : special.bindType ) || type;\n\t\t\thandlers = events[ type ] || [];\n\t\t\ttmp = tmp[ 2 ] &&\n\t\t\t\tnew RegExp( \"(^|\\\\.)\" + namespaces.join( \"\\\\.(?:.*\\\\.|)\" ) + \"(\\\\.|$)\" );\n\n\t\t\t// Remove matching events\n\t\t\torigCount = j = handlers.length;\n\t\t\twhile ( j-- ) {\n\t\t\t\thandleObj = handlers[ j ];\n\n\t\t\t\tif ( ( mappedTypes || origType === handleObj.origType ) &&\n\t\t\t\t\t( !handler || handler.guid === handleObj.guid ) &&\n\t\t\t\t\t( !tmp || tmp.test( handleObj.namespace ) ) &&\n\t\t\t\t\t( !selector || selector === handleObj.selector ||\n\t\t\t\t\t\tselector === \"**\" && handleObj.selector ) ) {\n\t\t\t\t\thandlers.splice( j, 1 );\n\n\t\t\t\t\tif ( handleObj.selector ) {\n\t\t\t\t\t\thandlers.delegateCount--;\n\t\t\t\t\t}\n\t\t\t\t\tif ( special.remove ) {\n\t\t\t\t\t\tspecial.remove.call( elem, handleObj );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Remove generic event handler if we removed something and no more handlers exist\n\t\t\t// (avoids potential for endless recursion during removal of special event handlers)\n\t\t\tif ( origCount && !handlers.length ) {\n\t\t\t\tif ( !special.teardown ||\n\t\t\t\t\tspecial.teardown.call( elem, namespaces, elemData.handle ) === false ) {\n\n\t\t\t\t\tjQuery.removeEvent( elem, type, elemData.handle );\n\t\t\t\t}\n\n\t\t\t\tdelete events[ type ];\n\t\t\t}\n\t\t}\n\n\t\t// Remove data and the expando if it's no longer used\n\t\tif ( jQuery.isEmptyObject( events ) ) {\n\t\t\tdataPriv.remove( elem, \"handle events\" );\n\t\t}\n\t},\n\n\tdispatch: function( nativeEvent ) {\n\n\t\t// Make a writable jQuery.Event from the native event object\n\t\tvar event = jQuery.event.fix( nativeEvent );\n\n\t\tvar i, j, ret, matched, handleObj, handlerQueue,\n\t\t\targs = new Array( arguments.length ),\n\t\t\thandlers = ( dataPriv.get( this, \"events\" ) || {} )[ event.type ] || [],\n\t\t\tspecial = jQuery.event.special[ event.type ] || {};\n\n\t\t// Use the fix-ed jQuery.Event rather than the (read-only) native event\n\t\targs[ 0 ] = event;\n\n\t\tfor ( i = 1; i < arguments.length; i++ ) {\n\t\t\targs[ i ] = arguments[ i ];\n\t\t}\n\n\t\tevent.delegateTarget = this;\n\n\t\t// Call the preDispatch hook for the mapped type, and let it bail if desired\n\t\tif ( special.preDispatch && special.preDispatch.call( this, event ) === false ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// Determine handlers\n\t\thandlerQueue = jQuery.event.handlers.call( this, event, handlers );\n\n\t\t// Run delegates first; they may want to stop propagation beneath us\n\t\ti = 0;\n\t\twhile ( ( matched = handlerQueue[ i++ ] ) && !event.isPropagationStopped() ) {\n\t\t\tevent.currentTarget = matched.elem;\n\n\t\t\tj = 0;\n\t\t\twhile ( ( handleObj = matched.handlers[ j++ ] ) &&\n\t\t\t\t!event.isImmediatePropagationStopped() ) {\n\n\t\t\t\t// Triggered event must either 1) have no namespace, or 2) have namespace(s)\n\t\t\t\t// a subset or equal to those in the bound event (both can have no namespace).\n\t\t\t\tif ( !event.rnamespace || event.rnamespace.test( handleObj.namespace ) ) {\n\n\t\t\t\t\tevent.handleObj = handleObj;\n\t\t\t\t\tevent.data = handleObj.data;\n\n\t\t\t\t\tret = ( ( jQuery.event.special[ handleObj.origType ] || {} ).handle ||\n\t\t\t\t\t\thandleObj.handler ).apply( matched.elem, args );\n\n\t\t\t\t\tif ( ret !== undefined ) {\n\t\t\t\t\t\tif ( ( event.result = ret ) === false ) {\n\t\t\t\t\t\t\tevent.preventDefault();\n\t\t\t\t\t\t\tevent.stopPropagation();\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Call the postDispatch hook for the mapped type\n\t\tif ( special.postDispatch ) {\n\t\t\tspecial.postDispatch.call( this, event );\n\t\t}\n\n\t\treturn event.result;\n\t},\n\n\thandlers: function( event, handlers ) {\n\t\tvar i, handleObj, sel, matchedHandlers, matchedSelectors,\n\t\t\thandlerQueue = [],\n\t\t\tdelegateCount = handlers.delegateCount,\n\t\t\tcur = event.target;\n\n\t\t// Find delegate handlers\n\t\tif ( delegateCount &&\n\n\t\t\t// Support: IE <=9\n\t\t\t// Black-hole SVG instance trees (trac-13180)\n\t\t\tcur.nodeType &&\n\n\t\t\t// Support: Firefox <=42\n\t\t\t// Suppress spec-violating clicks indicating a non-primary pointer button (trac-3861)\n\t\t\t// https://www.w3.org/TR/DOM-Level-3-Events/#event-type-click\n\t\t\t// Support: IE 11 only\n\t\t\t// ...but not arrow key \"clicks\" of radio inputs, which can have `button` -1 (gh-2343)\n\t\t\t!( event.type === \"click\" && event.button >= 1 ) ) {\n\n\t\t\tfor ( ; cur !== this; cur = cur.parentNode || this ) {\n\n\t\t\t\t// Don't check non-elements (#13208)\n\t\t\t\t// Don't process clicks on disabled elements (#6911, #8165, #11382, #11764)\n\t\t\t\tif ( cur.nodeType === 1 && !( event.type === \"click\" && cur.disabled === true ) ) {\n\t\t\t\t\tmatchedHandlers = [];\n\t\t\t\t\tmatchedSelectors = {};\n\t\t\t\t\tfor ( i = 0; i < delegateCount; i++ ) {\n\t\t\t\t\t\thandleObj = handlers[ i ];\n\n\t\t\t\t\t\t// Don't conflict with Object.prototype properties (#13203)\n\t\t\t\t\t\tsel = handleObj.selector + \" \";\n\n\t\t\t\t\t\tif ( matchedSelectors[ sel ] === undefined ) {\n\t\t\t\t\t\t\tmatchedSelectors[ sel ] = handleObj.needsContext ?\n\t\t\t\t\t\t\t\tjQuery( sel, this ).index( cur ) > -1 :\n\t\t\t\t\t\t\t\tjQuery.find( sel, this, null, [ cur ] ).length;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif ( matchedSelectors[ sel ] ) {\n\t\t\t\t\t\t\tmatchedHandlers.push( handleObj );\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif ( matchedHandlers.length ) {\n\t\t\t\t\t\thandlerQueue.push( { elem: cur, handlers: matchedHandlers } );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Add the remaining (directly-bound) handlers\n\t\tcur = this;\n\t\tif ( delegateCount < handlers.length ) {\n\t\t\thandlerQueue.push( { elem: cur, handlers: handlers.slice( delegateCount ) } );\n\t\t}\n\n\t\treturn handlerQueue;\n\t},\n\n\taddProp: function( name, hook ) {\n\t\tObject.defineProperty( jQuery.Event.prototype, name, {\n\t\t\tenumerable: true,\n\t\t\tconfigurable: true,\n\n\t\t\tget: jQuery.isFunction( hook ) ?\n\t\t\t\tfunction() {\n\t\t\t\t\tif ( this.originalEvent ) {\n\t\t\t\t\t\t\treturn hook( this.originalEvent );\n\t\t\t\t\t}\n\t\t\t\t} :\n\t\t\t\tfunction() {\n\t\t\t\t\tif ( this.originalEvent ) {\n\t\t\t\t\t\t\treturn this.originalEvent[ name ];\n\t\t\t\t\t}\n\t\t\t\t},\n\n\t\t\tset: function( value ) {\n\t\t\t\tObject.defineProperty( this, name, {\n\t\t\t\t\tenumerable: true,\n\t\t\t\t\tconfigurable: true,\n\t\t\t\t\twritable: true,\n\t\t\t\t\tvalue: value\n\t\t\t\t} );\n\t\t\t}\n\t\t} );\n\t},\n\n\tfix: function( originalEvent ) {\n\t\treturn originalEvent[ jQuery.expando ] ?\n\t\t\toriginalEvent :\n\t\t\tnew jQuery.Event( originalEvent );\n\t},\n\n\tspecial: {\n\t\tload: {\n\n\t\t\t// Prevent triggered image.load events from bubbling to window.load\n\t\t\tnoBubble: true\n\t\t},\n\t\tfocus: {\n\n\t\t\t// Fire native event if possible so blur/focus sequence is correct\n\t\t\ttrigger: function() {\n\t\t\t\tif ( this !== safeActiveElement() && this.focus ) {\n\t\t\t\t\tthis.focus();\n\t\t\t\t\treturn false;\n\t\t\t\t}\n\t\t\t},\n\t\t\tdelegateType: \"focusin\"\n\t\t},\n\t\tblur: {\n\t\t\ttrigger: function() {\n\t\t\t\tif ( this === safeActiveElement() && this.blur ) {\n\t\t\t\t\tthis.blur();\n\t\t\t\t\treturn false;\n\t\t\t\t}\n\t\t\t},\n\t\t\tdelegateType: \"focusout\"\n\t\t},\n\t\tclick: {\n\n\t\t\t// For checkbox, fire native event so checked state will be right\n\t\t\ttrigger: function() {\n\t\t\t\tif ( this.type === \"checkbox\" && this.click && nodeName( this, \"input\" ) ) {\n\t\t\t\t\tthis.click();\n\t\t\t\t\treturn false;\n\t\t\t\t}\n\t\t\t},\n\n\t\t\t// For cross-browser consistency, don't fire native .click() on links\n\t\t\t_default: function( event ) {\n\t\t\t\treturn nodeName( event.target, \"a\" );\n\t\t\t}\n\t\t},\n\n\t\tbeforeunload: {\n\t\t\tpostDispatch: function( event ) {\n\n\t\t\t\t// Support: Firefox 20+\n\t\t\t\t// Firefox doesn't alert if the returnValue field is not set.\n\t\t\t\tif ( event.result !== undefined && event.originalEvent ) {\n\t\t\t\t\tevent.originalEvent.returnValue = event.result;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n};\n\njQuery.removeEvent = function( elem, type, handle ) {\n\n\t// This \"if\" is needed for plain objects\n\tif ( elem.removeEventListener ) {\n\t\telem.removeEventListener( type, handle );\n\t}\n};\n\njQuery.Event = function( src, props ) {\n\n\t// Allow instantiation without the 'new' keyword\n\tif ( !( this instanceof jQuery.Event ) ) {\n\t\treturn new jQuery.Event( src, props );\n\t}\n\n\t// Event object\n\tif ( src && src.type ) {\n\t\tthis.originalEvent = src;\n\t\tthis.type = src.type;\n\n\t\t// Events bubbling up the document may have been marked as prevented\n\t\t// by a handler lower down the tree; reflect the correct value.\n\t\tthis.isDefaultPrevented = src.defaultPrevented ||\n\t\t\t\tsrc.defaultPrevented === undefined &&\n\n\t\t\t\t// Support: Android <=2.3 only\n\t\t\t\tsrc.returnValue === false ?\n\t\t\treturnTrue :\n\t\t\treturnFalse;\n\n\t\t// Create target properties\n\t\t// Support: Safari <=6 - 7 only\n\t\t// Target should not be a text node (#504, #13143)\n\t\tthis.target = ( src.target && src.target.nodeType === 3 ) ?\n\t\t\tsrc.target.parentNode :\n\t\t\tsrc.target;\n\n\t\tthis.currentTarget = src.currentTarget;\n\t\tthis.relatedTarget = src.relatedTarget;\n\n\t// Event type\n\t} else {\n\t\tthis.type = src;\n\t}\n\n\t// Put explicitly provided properties onto the event object\n\tif ( props ) {\n\t\tjQuery.extend( this, props );\n\t}\n\n\t// Create a timestamp if incoming event doesn't have one\n\tthis.timeStamp = src && src.timeStamp || jQuery.now();\n\n\t// Mark it as fixed\n\tthis[ jQuery.expando ] = true;\n};\n\n// jQuery.Event is based on DOM3 Events as specified by the ECMAScript Language Binding\n// https://www.w3.org/TR/2003/WD-DOM-Level-3-Events-20030331/ecma-script-binding.html\njQuery.Event.prototype = {\n\tconstructor: jQuery.Event,\n\tisDefaultPrevented: returnFalse,\n\tisPropagationStopped: returnFalse,\n\tisImmediatePropagationStopped: returnFalse,\n\tisSimulated: false,\n\n\tpreventDefault: function() {\n\t\tvar e = this.originalEvent;\n\n\t\tthis.isDefaultPrevented = returnTrue;\n\n\t\tif ( e && !this.isSimulated ) {\n\t\t\te.preventDefault();\n\t\t}\n\t},\n\tstopPropagation: function() {\n\t\tvar e = this.originalEvent;\n\n\t\tthis.isPropagationStopped = returnTrue;\n\n\t\tif ( e && !this.isSimulated ) {\n\t\t\te.stopPropagation();\n\t\t}\n\t},\n\tstopImmediatePropagation: function() {\n\t\tvar e = this.originalEvent;\n\n\t\tthis.isImmediatePropagationStopped = returnTrue;\n\n\t\tif ( e && !this.isSimulated ) {\n\t\t\te.stopImmediatePropagation();\n\t\t}\n\n\t\tthis.stopPropagation();\n\t}\n};\n\n// Includes all common event props including KeyEvent and MouseEvent specific props\njQuery.each( {\n\taltKey: true,\n\tbubbles: true,\n\tcancelable: true,\n\tchangedTouches: true,\n\tctrlKey: true,\n\tdetail: true,\n\teventPhase: true,\n\tmetaKey: true,\n\tpageX: true,\n\tpageY: true,\n\tshiftKey: true,\n\tview: true,\n\t\"char\": true,\n\tcharCode: true,\n\tkey: true,\n\tkeyCode: true,\n\tbutton: true,\n\tbuttons: true,\n\tclientX: true,\n\tclientY: true,\n\toffsetX: true,\n\toffsetY: true,\n\tpointerId: true,\n\tpointerType: true,\n\tscreenX: true,\n\tscreenY: true,\n\ttargetTouches: true,\n\ttoElement: true,\n\ttouches: true,\n\n\twhich: function( event ) {\n\t\tvar button = event.button;\n\n\t\t// Add which for key events\n\t\tif ( event.which == null && rkeyEvent.test( event.type ) ) {\n\t\t\treturn event.charCode != null ? event.charCode : event.keyCode;\n\t\t}\n\n\t\t// Add which for click: 1 === left; 2 === middle; 3 === right\n\t\tif ( !event.which && button !== undefined && rmouseEvent.test( event.type ) ) {\n\t\t\tif ( button & 1 ) {\n\t\t\t\treturn 1;\n\t\t\t}\n\n\t\t\tif ( button & 2 ) {\n\t\t\t\treturn 3;\n\t\t\t}\n\n\t\t\tif ( button & 4 ) {\n\t\t\t\treturn 2;\n\t\t\t}\n\n\t\t\treturn 0;\n\t\t}\n\n\t\treturn event.which;\n\t}\n}, jQuery.event.addProp );\n\n// Create mouseenter/leave events using mouseover/out and event-time checks\n// so that event delegation works in jQuery.\n// Do the same for pointerenter/pointerleave and pointerover/pointerout\n//\n// Support: Safari 7 only\n// Safari sends mouseenter too often; see:\n// https://bugs.chromium.org/p/chromium/issues/detail?id=470258\n// for the description of the bug (it existed in older Chrome versions as well).\njQuery.each( {\n\tmouseenter: \"mouseover\",\n\tmouseleave: \"mouseout\",\n\tpointerenter: \"pointerover\",\n\tpointerleave: \"pointerout\"\n}, function( orig, fix ) {\n\tjQuery.event.special[ orig ] = {\n\t\tdelegateType: fix,\n\t\tbindType: fix,\n\n\t\thandle: function( event ) {\n\t\t\tvar ret,\n\t\t\t\ttarget = this,\n\t\t\t\trelated = event.relatedTarget,\n\t\t\t\thandleObj = event.handleObj;\n\n\t\t\t// For mouseenter/leave call the handler if related is outside the target.\n\t\t\t// NB: No relatedTarget if the mouse left/entered the browser window\n\t\t\tif ( !related || ( related !== target && !jQuery.contains( target, related ) ) ) {\n\t\t\t\tevent.type = handleObj.origType;\n\t\t\t\tret = handleObj.handler.apply( this, arguments );\n\t\t\t\tevent.type = fix;\n\t\t\t}\n\t\t\treturn ret;\n\t\t}\n\t};\n} );\n\njQuery.fn.extend( {\n\n\ton: function( types, selector, data, fn ) {\n\t\treturn on( this, types, selector, data, fn );\n\t},\n\tone: function( types, selector, data, fn ) {\n\t\treturn on( this, types, selector, data, fn, 1 );\n\t},\n\toff: function( types, selector, fn ) {\n\t\tvar handleObj, type;\n\t\tif ( types && types.preventDefault && types.handleObj ) {\n\n\t\t\t// ( event ) dispatched jQuery.Event\n\t\t\thandleObj = types.handleObj;\n\t\t\tjQuery( types.delegateTarget ).off(\n\t\t\t\thandleObj.namespace ?\n\t\t\t\t\thandleObj.origType + \".\" + handleObj.namespace :\n\t\t\t\t\thandleObj.origType,\n\t\t\t\thandleObj.selector,\n\t\t\t\thandleObj.handler\n\t\t\t);\n\t\t\treturn this;\n\t\t}\n\t\tif ( typeof types === \"object\" ) {\n\n\t\t\t// ( types-object [, selector] )\n\t\t\tfor ( type in types ) {\n\t\t\t\tthis.off( type, selector, types[ type ] );\n\t\t\t}\n\t\t\treturn this;\n\t\t}\n\t\tif ( selector === false || typeof selector === \"function\" ) {\n\n\t\t\t// ( types [, fn] )\n\t\t\tfn = selector;\n\t\t\tselector = undefined;\n\t\t}\n\t\tif ( fn === false ) {\n\t\t\tfn = returnFalse;\n\t\t}\n\t\treturn this.each( function() {\n\t\t\tjQuery.event.remove( this, types, fn, selector );\n\t\t} );\n\t}\n} );\n\n\nvar\n\n\t/* eslint-disable max-len */\n\n\t// See https://github.com/eslint/eslint/issues/3229\n\trxhtmlTag = /<(?!area|br|col|embed|hr|img|input|link|meta|param)(([a-z][^\\/\\0>\\x20\\t\\r\\n\\f]*)[^>]*)\\/>/gi,\n\n\t/* eslint-enable */\n\n\t// Support: IE <=10 - 11, Edge 12 - 13\n\t// In IE/Edge using regex groups here causes severe slowdowns.\n\t// See https://connect.microsoft.com/IE/feedback/details/1736512/\n\trnoInnerhtml = /\\s*$/g;\n\n// Prefer a tbody over its parent table for containing new rows\nfunction manipulationTarget( elem, content ) {\n\tif ( nodeName( elem, \"table\" ) &&\n\t\tnodeName( content.nodeType !== 11 ? content : content.firstChild, \"tr\" ) ) {\n\n\t\treturn jQuery( \">tbody\", elem )[ 0 ] || elem;\n\t}\n\n\treturn elem;\n}\n\n// Replace/restore the type attribute of script elements for safe DOM manipulation\nfunction disableScript( elem ) {\n\telem.type = ( elem.getAttribute( \"type\" ) !== null ) + \"/\" + elem.type;\n\treturn elem;\n}\nfunction restoreScript( elem ) {\n\tvar match = rscriptTypeMasked.exec( elem.type );\n\n\tif ( match ) {\n\t\telem.type = match[ 1 ];\n\t} else {\n\t\telem.removeAttribute( \"type\" );\n\t}\n\n\treturn elem;\n}\n\nfunction cloneCopyEvent( src, dest ) {\n\tvar i, l, type, pdataOld, pdataCur, udataOld, udataCur, events;\n\n\tif ( dest.nodeType !== 1 ) {\n\t\treturn;\n\t}\n\n\t// 1. Copy private data: events, handlers, etc.\n\tif ( dataPriv.hasData( src ) ) {\n\t\tpdataOld = dataPriv.access( src );\n\t\tpdataCur = dataPriv.set( dest, pdataOld );\n\t\tevents = pdataOld.events;\n\n\t\tif ( events ) {\n\t\t\tdelete pdataCur.handle;\n\t\t\tpdataCur.events = {};\n\n\t\t\tfor ( type in events ) {\n\t\t\t\tfor ( i = 0, l = events[ type ].length; i < l; i++ ) {\n\t\t\t\t\tjQuery.event.add( dest, type, events[ type ][ i ] );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// 2. Copy user data\n\tif ( dataUser.hasData( src ) ) {\n\t\tudataOld = dataUser.access( src );\n\t\tudataCur = jQuery.extend( {}, udataOld );\n\n\t\tdataUser.set( dest, udataCur );\n\t}\n}\n\n// Fix IE bugs, see support tests\nfunction fixInput( src, dest ) {\n\tvar nodeName = dest.nodeName.toLowerCase();\n\n\t// Fails to persist the checked state of a cloned checkbox or radio button.\n\tif ( nodeName === \"input\" && rcheckableType.test( src.type ) ) {\n\t\tdest.checked = src.checked;\n\n\t// Fails to return the selected option to the default selected state when cloning options\n\t} else if ( nodeName === \"input\" || nodeName === \"textarea\" ) {\n\t\tdest.defaultValue = src.defaultValue;\n\t}\n}\n\nfunction domManip( collection, args, callback, ignored ) {\n\n\t// Flatten any nested arrays\n\targs = concat.apply( [], args );\n\n\tvar fragment, first, scripts, hasScripts, node, doc,\n\t\ti = 0,\n\t\tl = collection.length,\n\t\tiNoClone = l - 1,\n\t\tvalue = args[ 0 ],\n\t\tisFunction = jQuery.isFunction( value );\n\n\t// We can't cloneNode fragments that contain checked, in WebKit\n\tif ( isFunction ||\n\t\t\t( l > 1 && typeof value === \"string\" &&\n\t\t\t\t!support.checkClone && rchecked.test( value ) ) ) {\n\t\treturn collection.each( function( index ) {\n\t\t\tvar self = collection.eq( index );\n\t\t\tif ( isFunction ) {\n\t\t\t\targs[ 0 ] = value.call( this, index, self.html() );\n\t\t\t}\n\t\t\tdomManip( self, args, callback, ignored );\n\t\t} );\n\t}\n\n\tif ( l ) {\n\t\tfragment = buildFragment( args, collection[ 0 ].ownerDocument, false, collection, ignored );\n\t\tfirst = fragment.firstChild;\n\n\t\tif ( fragment.childNodes.length === 1 ) {\n\t\t\tfragment = first;\n\t\t}\n\n\t\t// Require either new content or an interest in ignored elements to invoke the callback\n\t\tif ( first || ignored ) {\n\t\t\tscripts = jQuery.map( getAll( fragment, \"script\" ), disableScript );\n\t\t\thasScripts = scripts.length;\n\n\t\t\t// Use the original fragment for the last item\n\t\t\t// instead of the first because it can end up\n\t\t\t// being emptied incorrectly in certain situations (#8070).\n\t\t\tfor ( ; i < l; i++ ) {\n\t\t\t\tnode = fragment;\n\n\t\t\t\tif ( i !== iNoClone ) {\n\t\t\t\t\tnode = jQuery.clone( node, true, true );\n\n\t\t\t\t\t// Keep references to cloned scripts for later restoration\n\t\t\t\t\tif ( hasScripts ) {\n\n\t\t\t\t\t\t// Support: Android <=4.0 only, PhantomJS 1 only\n\t\t\t\t\t\t// push.apply(_, arraylike) throws on ancient WebKit\n\t\t\t\t\t\tjQuery.merge( scripts, getAll( node, \"script\" ) );\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\tcallback.call( collection[ i ], node, i );\n\t\t\t}\n\n\t\t\tif ( hasScripts ) {\n\t\t\t\tdoc = scripts[ scripts.length - 1 ].ownerDocument;\n\n\t\t\t\t// Reenable scripts\n\t\t\t\tjQuery.map( scripts, restoreScript );\n\n\t\t\t\t// Evaluate executable scripts on first document insertion\n\t\t\t\tfor ( i = 0; i < hasScripts; i++ ) {\n\t\t\t\t\tnode = scripts[ i ];\n\t\t\t\t\tif ( rscriptType.test( node.type || \"\" ) &&\n\t\t\t\t\t\t!dataPriv.access( node, \"globalEval\" ) &&\n\t\t\t\t\t\tjQuery.contains( doc, node ) ) {\n\n\t\t\t\t\t\tif ( node.src ) {\n\n\t\t\t\t\t\t\t// Optional AJAX dependency, but won't run scripts if not present\n\t\t\t\t\t\t\tif ( jQuery._evalUrl ) {\n\t\t\t\t\t\t\t\tjQuery._evalUrl( node.src );\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\tDOMEval( node.textContent.replace( rcleanScript, \"\" ), doc );\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\treturn collection;\n}\n\nfunction remove( elem, selector, keepData ) {\n\tvar node,\n\t\tnodes = selector ? jQuery.filter( selector, elem ) : elem,\n\t\ti = 0;\n\n\tfor ( ; ( node = nodes[ i ] ) != null; i++ ) {\n\t\tif ( !keepData && node.nodeType === 1 ) {\n\t\t\tjQuery.cleanData( getAll( node ) );\n\t\t}\n\n\t\tif ( node.parentNode ) {\n\t\t\tif ( keepData && jQuery.contains( node.ownerDocument, node ) ) {\n\t\t\t\tsetGlobalEval( getAll( node, \"script\" ) );\n\t\t\t}\n\t\t\tnode.parentNode.removeChild( node );\n\t\t}\n\t}\n\n\treturn elem;\n}\n\njQuery.extend( {\n\thtmlPrefilter: function( html ) {\n\t\treturn html.replace( rxhtmlTag, \"<$1>\" );\n\t},\n\n\tclone: function( elem, dataAndEvents, deepDataAndEvents ) {\n\t\tvar i, l, srcElements, destElements,\n\t\t\tclone = elem.cloneNode( true ),\n\t\t\tinPage = jQuery.contains( elem.ownerDocument, elem );\n\n\t\t// Fix IE cloning issues\n\t\tif ( !support.noCloneChecked && ( elem.nodeType === 1 || elem.nodeType === 11 ) &&\n\t\t\t\t!jQuery.isXMLDoc( elem ) ) {\n\n\t\t\t// We eschew Sizzle here for performance reasons: https://jsperf.com/getall-vs-sizzle/2\n\t\t\tdestElements = getAll( clone );\n\t\t\tsrcElements = getAll( elem );\n\n\t\t\tfor ( i = 0, l = srcElements.length; i < l; i++ ) {\n\t\t\t\tfixInput( srcElements[ i ], destElements[ i ] );\n\t\t\t}\n\t\t}\n\n\t\t// Copy the events from the original to the clone\n\t\tif ( dataAndEvents ) {\n\t\t\tif ( deepDataAndEvents ) {\n\t\t\t\tsrcElements = srcElements || getAll( elem );\n\t\t\t\tdestElements = destElements || getAll( clone );\n\n\t\t\t\tfor ( i = 0, l = srcElements.length; i < l; i++ ) {\n\t\t\t\t\tcloneCopyEvent( srcElements[ i ], destElements[ i ] );\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tcloneCopyEvent( elem, clone );\n\t\t\t}\n\t\t}\n\n\t\t// Preserve script evaluation history\n\t\tdestElements = getAll( clone, \"script\" );\n\t\tif ( destElements.length > 0 ) {\n\t\t\tsetGlobalEval( destElements, !inPage && getAll( elem, \"script\" ) );\n\t\t}\n\n\t\t// Return the cloned set\n\t\treturn clone;\n\t},\n\n\tcleanData: function( elems ) {\n\t\tvar data, elem, type,\n\t\t\tspecial = jQuery.event.special,\n\t\t\ti = 0;\n\n\t\tfor ( ; ( elem = elems[ i ] ) !== undefined; i++ ) {\n\t\t\tif ( acceptData( elem ) ) {\n\t\t\t\tif ( ( data = elem[ dataPriv.expando ] ) ) {\n\t\t\t\t\tif ( data.events ) {\n\t\t\t\t\t\tfor ( type in data.events ) {\n\t\t\t\t\t\t\tif ( special[ type ] ) {\n\t\t\t\t\t\t\t\tjQuery.event.remove( elem, type );\n\n\t\t\t\t\t\t\t// This is a shortcut to avoid jQuery.event.remove's overhead\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tjQuery.removeEvent( elem, type, data.handle );\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\t// Support: Chrome <=35 - 45+\n\t\t\t\t\t// Assign undefined instead of using delete, see Data#remove\n\t\t\t\t\telem[ dataPriv.expando ] = undefined;\n\t\t\t\t}\n\t\t\t\tif ( elem[ dataUser.expando ] ) {\n\n\t\t\t\t\t// Support: Chrome <=35 - 45+\n\t\t\t\t\t// Assign undefined instead of using delete, see Data#remove\n\t\t\t\t\telem[ dataUser.expando ] = undefined;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n} );\n\njQuery.fn.extend( {\n\tdetach: function( selector ) {\n\t\treturn remove( this, selector, true );\n\t},\n\n\tremove: function( selector ) {\n\t\treturn remove( this, selector );\n\t},\n\n\ttext: function( value ) {\n\t\treturn access( this, function( value ) {\n\t\t\treturn value === undefined ?\n\t\t\t\tjQuery.text( this ) :\n\t\t\t\tthis.empty().each( function() {\n\t\t\t\t\tif ( this.nodeType === 1 || this.nodeType === 11 || this.nodeType === 9 ) {\n\t\t\t\t\t\tthis.textContent = value;\n\t\t\t\t\t}\n\t\t\t\t} );\n\t\t}, null, value, arguments.length );\n\t},\n\n\tappend: function() {\n\t\treturn domManip( this, arguments, function( elem ) {\n\t\t\tif ( this.nodeType === 1 || this.nodeType === 11 || this.nodeType === 9 ) {\n\t\t\t\tvar target = manipulationTarget( this, elem );\n\t\t\t\ttarget.appendChild( elem );\n\t\t\t}\n\t\t} );\n\t},\n\n\tprepend: function() {\n\t\treturn domManip( this, arguments, function( elem ) {\n\t\t\tif ( this.nodeType === 1 || this.nodeType === 11 || this.nodeType === 9 ) {\n\t\t\t\tvar target = manipulationTarget( this, elem );\n\t\t\t\ttarget.insertBefore( elem, target.firstChild );\n\t\t\t}\n\t\t} );\n\t},\n\n\tbefore: function() {\n\t\treturn domManip( this, arguments, function( elem ) {\n\t\t\tif ( this.parentNode ) {\n\t\t\t\tthis.parentNode.insertBefore( elem, this );\n\t\t\t}\n\t\t} );\n\t},\n\n\tafter: function() {\n\t\treturn domManip( this, arguments, function( elem ) {\n\t\t\tif ( this.parentNode ) {\n\t\t\t\tthis.parentNode.insertBefore( elem, this.nextSibling );\n\t\t\t}\n\t\t} );\n\t},\n\n\tempty: function() {\n\t\tvar elem,\n\t\t\ti = 0;\n\n\t\tfor ( ; ( elem = this[ i ] ) != null; i++ ) {\n\t\t\tif ( elem.nodeType === 1 ) {\n\n\t\t\t\t// Prevent memory leaks\n\t\t\t\tjQuery.cleanData( getAll( elem, false ) );\n\n\t\t\t\t// Remove any remaining nodes\n\t\t\t\telem.textContent = \"\";\n\t\t\t}\n\t\t}\n\n\t\treturn this;\n\t},\n\n\tclone: function( dataAndEvents, deepDataAndEvents ) {\n\t\tdataAndEvents = dataAndEvents == null ? false : dataAndEvents;\n\t\tdeepDataAndEvents = deepDataAndEvents == null ? dataAndEvents : deepDataAndEvents;\n\n\t\treturn this.map( function() {\n\t\t\treturn jQuery.clone( this, dataAndEvents, deepDataAndEvents );\n\t\t} );\n\t},\n\n\thtml: function( value ) {\n\t\treturn access( this, function( value ) {\n\t\t\tvar elem = this[ 0 ] || {},\n\t\t\t\ti = 0,\n\t\t\t\tl = this.length;\n\n\t\t\tif ( value === undefined && elem.nodeType === 1 ) {\n\t\t\t\treturn elem.innerHTML;\n\t\t\t}\n\n\t\t\t// See if we can take a shortcut and just use innerHTML\n\t\t\tif ( typeof value === \"string\" && !rnoInnerhtml.test( value ) &&\n\t\t\t\t!wrapMap[ ( rtagName.exec( value ) || [ \"\", \"\" ] )[ 1 ].toLowerCase() ] ) {\n\n\t\t\t\tvalue = jQuery.htmlPrefilter( value );\n\n\t\t\t\ttry {\n\t\t\t\t\tfor ( ; i < l; i++ ) {\n\t\t\t\t\t\telem = this[ i ] || {};\n\n\t\t\t\t\t\t// Remove element nodes and prevent memory leaks\n\t\t\t\t\t\tif ( elem.nodeType === 1 ) {\n\t\t\t\t\t\t\tjQuery.cleanData( getAll( elem, false ) );\n\t\t\t\t\t\t\telem.innerHTML = value;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\telem = 0;\n\n\t\t\t\t// If using innerHTML throws an exception, use the fallback method\n\t\t\t\t} catch ( e ) {}\n\t\t\t}\n\n\t\t\tif ( elem ) {\n\t\t\t\tthis.empty().append( value );\n\t\t\t}\n\t\t}, null, value, arguments.length );\n\t},\n\n\treplaceWith: function() {\n\t\tvar ignored = [];\n\n\t\t// Make the changes, replacing each non-ignored context element with the new content\n\t\treturn domManip( this, arguments, function( elem ) {\n\t\t\tvar parent = this.parentNode;\n\n\t\t\tif ( jQuery.inArray( this, ignored ) < 0 ) {\n\t\t\t\tjQuery.cleanData( getAll( this ) );\n\t\t\t\tif ( parent ) {\n\t\t\t\t\tparent.replaceChild( elem, this );\n\t\t\t\t}\n\t\t\t}\n\n\t\t// Force callback invocation\n\t\t}, ignored );\n\t}\n} );\n\njQuery.each( {\n\tappendTo: \"append\",\n\tprependTo: \"prepend\",\n\tinsertBefore: \"before\",\n\tinsertAfter: \"after\",\n\treplaceAll: \"replaceWith\"\n}, function( name, original ) {\n\tjQuery.fn[ name ] = function( selector ) {\n\t\tvar elems,\n\t\t\tret = [],\n\t\t\tinsert = jQuery( selector ),\n\t\t\tlast = insert.length - 1,\n\t\t\ti = 0;\n\n\t\tfor ( ; i <= last; i++ ) {\n\t\t\telems = i === last ? this : this.clone( true );\n\t\t\tjQuery( insert[ i ] )[ original ]( elems );\n\n\t\t\t// Support: Android <=4.0 only, PhantomJS 1 only\n\t\t\t// .get() because push.apply(_, arraylike) throws on ancient WebKit\n\t\t\tpush.apply( ret, elems.get() );\n\t\t}\n\n\t\treturn this.pushStack( ret );\n\t};\n} );\nvar rmargin = ( /^margin/ );\n\nvar rnumnonpx = new RegExp( \"^(\" + pnum + \")(?!px)[a-z%]+$\", \"i\" );\n\nvar getStyles = function( elem ) {\n\n\t\t// Support: IE <=11 only, Firefox <=30 (#15098, #14150)\n\t\t// IE throws on elements created in popups\n\t\t// FF meanwhile throws on frame elements through \"defaultView.getComputedStyle\"\n\t\tvar view = elem.ownerDocument.defaultView;\n\n\t\tif ( !view || !view.opener ) {\n\t\t\tview = window;\n\t\t}\n\n\t\treturn view.getComputedStyle( elem );\n\t};\n\n\n\n( function() {\n\n\t// Executing both pixelPosition & boxSizingReliable tests require only one layout\n\t// so they're executed at the same time to save the second computation.\n\tfunction computeStyleTests() {\n\n\t\t// This is a singleton, we need to execute it only once\n\t\tif ( !div ) {\n\t\t\treturn;\n\t\t}\n\n\t\tdiv.style.cssText =\n\t\t\t\"box-sizing:border-box;\" +\n\t\t\t\"position:relative;display:block;\" +\n\t\t\t\"margin:auto;border:1px;padding:1px;\" +\n\t\t\t\"top:1%;width:50%\";\n\t\tdiv.innerHTML = \"\";\n\t\tdocumentElement.appendChild( container );\n\n\t\tvar divStyle = window.getComputedStyle( div );\n\t\tpixelPositionVal = divStyle.top !== \"1%\";\n\n\t\t// Support: Android 4.0 - 4.3 only, Firefox <=3 - 44\n\t\treliableMarginLeftVal = divStyle.marginLeft === \"2px\";\n\t\tboxSizingReliableVal = divStyle.width === \"4px\";\n\n\t\t// Support: Android 4.0 - 4.3 only\n\t\t// Some styles come back with percentage values, even though they shouldn't\n\t\tdiv.style.marginRight = \"50%\";\n\t\tpixelMarginRightVal = divStyle.marginRight === \"4px\";\n\n\t\tdocumentElement.removeChild( container );\n\n\t\t// Nullify the div so it wouldn't be stored in the memory and\n\t\t// it will also be a sign that checks already performed\n\t\tdiv = null;\n\t}\n\n\tvar pixelPositionVal, boxSizingReliableVal, pixelMarginRightVal, reliableMarginLeftVal,\n\t\tcontainer = document.createElement( \"div\" ),\n\t\tdiv = document.createElement( \"div\" );\n\n\t// Finish early in limited (non-browser) environments\n\tif ( !div.style ) {\n\t\treturn;\n\t}\n\n\t// Support: IE <=9 - 11 only\n\t// Style of cloned element affects source element cloned (#8908)\n\tdiv.style.backgroundClip = \"content-box\";\n\tdiv.cloneNode( true ).style.backgroundClip = \"\";\n\tsupport.clearCloneStyle = div.style.backgroundClip === \"content-box\";\n\n\tcontainer.style.cssText = \"border:0;width:8px;height:0;top:0;left:-9999px;\" +\n\t\t\"padding:0;margin-top:1px;position:absolute\";\n\tcontainer.appendChild( div );\n\n\tjQuery.extend( support, {\n\t\tpixelPosition: function() {\n\t\t\tcomputeStyleTests();\n\t\t\treturn pixelPositionVal;\n\t\t},\n\t\tboxSizingReliable: function() {\n\t\t\tcomputeStyleTests();\n\t\t\treturn boxSizingReliableVal;\n\t\t},\n\t\tpixelMarginRight: function() {\n\t\t\tcomputeStyleTests();\n\t\t\treturn pixelMarginRightVal;\n\t\t},\n\t\treliableMarginLeft: function() {\n\t\t\tcomputeStyleTests();\n\t\t\treturn reliableMarginLeftVal;\n\t\t}\n\t} );\n} )();\n\n\nfunction curCSS( elem, name, computed ) {\n\tvar width, minWidth, maxWidth, ret,\n\n\t\t// Support: Firefox 51+\n\t\t// Retrieving style before computed somehow\n\t\t// fixes an issue with getting wrong values\n\t\t// on detached elements\n\t\tstyle = elem.style;\n\n\tcomputed = computed || getStyles( elem );\n\n\t// getPropertyValue is needed for:\n\t// .css('filter') (IE 9 only, #12537)\n\t// .css('--customProperty) (#3144)\n\tif ( computed ) {\n\t\tret = computed.getPropertyValue( name ) || computed[ name ];\n\n\t\tif ( ret === \"\" && !jQuery.contains( elem.ownerDocument, elem ) ) {\n\t\t\tret = jQuery.style( elem, name );\n\t\t}\n\n\t\t// A tribute to the \"awesome hack by Dean Edwards\"\n\t\t// Android Browser returns percentage for some values,\n\t\t// but width seems to be reliably pixels.\n\t\t// This is against the CSSOM draft spec:\n\t\t// https://drafts.csswg.org/cssom/#resolved-values\n\t\tif ( !support.pixelMarginRight() && rnumnonpx.test( ret ) && rmargin.test( name ) ) {\n\n\t\t\t// Remember the original values\n\t\t\twidth = style.width;\n\t\t\tminWidth = style.minWidth;\n\t\t\tmaxWidth = style.maxWidth;\n\n\t\t\t// Put in the new values to get a computed value out\n\t\t\tstyle.minWidth = style.maxWidth = style.width = ret;\n\t\t\tret = computed.width;\n\n\t\t\t// Revert the changed values\n\t\t\tstyle.width = width;\n\t\t\tstyle.minWidth = minWidth;\n\t\t\tstyle.maxWidth = maxWidth;\n\t\t}\n\t}\n\n\treturn ret !== undefined ?\n\n\t\t// Support: IE <=9 - 11 only\n\t\t// IE returns zIndex value as an integer.\n\t\tret + \"\" :\n\t\tret;\n}\n\n\nfunction addGetHookIf( conditionFn, hookFn ) {\n\n\t// Define the hook, we'll check on the first run if it's really needed.\n\treturn {\n\t\tget: function() {\n\t\t\tif ( conditionFn() ) {\n\n\t\t\t\t// Hook not needed (or it's not possible to use it due\n\t\t\t\t// to missing dependency), remove it.\n\t\t\t\tdelete this.get;\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// Hook needed; redefine it so that the support test is not executed again.\n\t\t\treturn ( this.get = hookFn ).apply( this, arguments );\n\t\t}\n\t};\n}\n\n\nvar\n\n\t// Swappable if display is none or starts with table\n\t// except \"table\", \"table-cell\", or \"table-caption\"\n\t// See here for display values: https://developer.mozilla.org/en-US/docs/CSS/display\n\trdisplayswap = /^(none|table(?!-c[ea]).+)/,\n\trcustomProp = /^--/,\n\tcssShow = { position: \"absolute\", visibility: \"hidden\", display: \"block\" },\n\tcssNormalTransform = {\n\t\tletterSpacing: \"0\",\n\t\tfontWeight: \"400\"\n\t},\n\n\tcssPrefixes = [ \"Webkit\", \"Moz\", \"ms\" ],\n\temptyStyle = document.createElement( \"div\" ).style;\n\n// Return a css property mapped to a potentially vendor prefixed property\nfunction vendorPropName( name ) {\n\n\t// Shortcut for names that are not vendor prefixed\n\tif ( name in emptyStyle ) {\n\t\treturn name;\n\t}\n\n\t// Check for vendor prefixed names\n\tvar capName = name[ 0 ].toUpperCase() + name.slice( 1 ),\n\t\ti = cssPrefixes.length;\n\n\twhile ( i-- ) {\n\t\tname = cssPrefixes[ i ] + capName;\n\t\tif ( name in emptyStyle ) {\n\t\t\treturn name;\n\t\t}\n\t}\n}\n\n// Return a property mapped along what jQuery.cssProps suggests or to\n// a vendor prefixed property.\nfunction finalPropName( name ) {\n\tvar ret = jQuery.cssProps[ name ];\n\tif ( !ret ) {\n\t\tret = jQuery.cssProps[ name ] = vendorPropName( name ) || name;\n\t}\n\treturn ret;\n}\n\nfunction setPositiveNumber( elem, value, subtract ) {\n\n\t// Any relative (+/-) values have already been\n\t// normalized at this point\n\tvar matches = rcssNum.exec( value );\n\treturn matches ?\n\n\t\t// Guard against undefined \"subtract\", e.g., when used as in cssHooks\n\t\tMath.max( 0, matches[ 2 ] - ( subtract || 0 ) ) + ( matches[ 3 ] || \"px\" ) :\n\t\tvalue;\n}\n\nfunction augmentWidthOrHeight( elem, name, extra, isBorderBox, styles ) {\n\tvar i,\n\t\tval = 0;\n\n\t// If we already have the right measurement, avoid augmentation\n\tif ( extra === ( isBorderBox ? \"border\" : \"content\" ) ) {\n\t\ti = 4;\n\n\t// Otherwise initialize for horizontal or vertical properties\n\t} else {\n\t\ti = name === \"width\" ? 1 : 0;\n\t}\n\n\tfor ( ; i < 4; i += 2 ) {\n\n\t\t// Both box models exclude margin, so add it if we want it\n\t\tif ( extra === \"margin\" ) {\n\t\t\tval += jQuery.css( elem, extra + cssExpand[ i ], true, styles );\n\t\t}\n\n\t\tif ( isBorderBox ) {\n\n\t\t\t// border-box includes padding, so remove it if we want content\n\t\t\tif ( extra === \"content\" ) {\n\t\t\t\tval -= jQuery.css( elem, \"padding\" + cssExpand[ i ], true, styles );\n\t\t\t}\n\n\t\t\t// At this point, extra isn't border nor margin, so remove border\n\t\t\tif ( extra !== \"margin\" ) {\n\t\t\t\tval -= jQuery.css( elem, \"border\" + cssExpand[ i ] + \"Width\", true, styles );\n\t\t\t}\n\t\t} else {\n\n\t\t\t// At this point, extra isn't content, so add padding\n\t\t\tval += jQuery.css( elem, \"padding\" + cssExpand[ i ], true, styles );\n\n\t\t\t// At this point, extra isn't content nor padding, so add border\n\t\t\tif ( extra !== \"padding\" ) {\n\t\t\t\tval += jQuery.css( elem, \"border\" + cssExpand[ i ] + \"Width\", true, styles );\n\t\t\t}\n\t\t}\n\t}\n\n\treturn val;\n}\n\nfunction getWidthOrHeight( elem, name, extra ) {\n\n\t// Start with computed style\n\tvar valueIsBorderBox,\n\t\tstyles = getStyles( elem ),\n\t\tval = curCSS( elem, name, styles ),\n\t\tisBorderBox = jQuery.css( elem, \"boxSizing\", false, styles ) === \"border-box\";\n\n\t// Computed unit is not pixels. Stop here and return.\n\tif ( rnumnonpx.test( val ) ) {\n\t\treturn val;\n\t}\n\n\t// Check for style in case a browser which returns unreliable values\n\t// for getComputedStyle silently falls back to the reliable elem.style\n\tvalueIsBorderBox = isBorderBox &&\n\t\t( support.boxSizingReliable() || val === elem.style[ name ] );\n\n\t// Fall back to offsetWidth/Height when value is \"auto\"\n\t// This happens for inline elements with no explicit setting (gh-3571)\n\tif ( val === \"auto\" ) {\n\t\tval = elem[ \"offset\" + name[ 0 ].toUpperCase() + name.slice( 1 ) ];\n\t}\n\n\t// Normalize \"\", auto, and prepare for extra\n\tval = parseFloat( val ) || 0;\n\n\t// Use the active box-sizing model to add/subtract irrelevant styles\n\treturn ( val +\n\t\taugmentWidthOrHeight(\n\t\t\telem,\n\t\t\tname,\n\t\t\textra || ( isBorderBox ? \"border\" : \"content\" ),\n\t\t\tvalueIsBorderBox,\n\t\t\tstyles\n\t\t)\n\t) + \"px\";\n}\n\njQuery.extend( {\n\n\t// Add in style property hooks for overriding the default\n\t// behavior of getting and setting a style property\n\tcssHooks: {\n\t\topacity: {\n\t\t\tget: function( elem, computed ) {\n\t\t\t\tif ( computed ) {\n\n\t\t\t\t\t// We should always get a number back from opacity\n\t\t\t\t\tvar ret = curCSS( elem, \"opacity\" );\n\t\t\t\t\treturn ret === \"\" ? \"1\" : ret;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t},\n\n\t// Don't automatically add \"px\" to these possibly-unitless properties\n\tcssNumber: {\n\t\t\"animationIterationCount\": true,\n\t\t\"columnCount\": true,\n\t\t\"fillOpacity\": true,\n\t\t\"flexGrow\": true,\n\t\t\"flexShrink\": true,\n\t\t\"fontWeight\": true,\n\t\t\"lineHeight\": true,\n\t\t\"opacity\": true,\n\t\t\"order\": true,\n\t\t\"orphans\": true,\n\t\t\"widows\": true,\n\t\t\"zIndex\": true,\n\t\t\"zoom\": true\n\t},\n\n\t// Add in properties whose names you wish to fix before\n\t// setting or getting the value\n\tcssProps: {\n\t\t\"float\": \"cssFloat\"\n\t},\n\n\t// Get and set the style property on a DOM Node\n\tstyle: function( elem, name, value, extra ) {\n\n\t\t// Don't set styles on text and comment nodes\n\t\tif ( !elem || elem.nodeType === 3 || elem.nodeType === 8 || !elem.style ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// Make sure that we're working with the right name\n\t\tvar ret, type, hooks,\n\t\t\torigName = jQuery.camelCase( name ),\n\t\t\tisCustomProp = rcustomProp.test( name ),\n\t\t\tstyle = elem.style;\n\n\t\t// Make sure that we're working with the right name. We don't\n\t\t// want to query the value if it is a CSS custom property\n\t\t// since they are user-defined.\n\t\tif ( !isCustomProp ) {\n\t\t\tname = finalPropName( origName );\n\t\t}\n\n\t\t// Gets hook for the prefixed version, then unprefixed version\n\t\thooks = jQuery.cssHooks[ name ] || jQuery.cssHooks[ origName ];\n\n\t\t// Check if we're setting a value\n\t\tif ( value !== undefined ) {\n\t\t\ttype = typeof value;\n\n\t\t\t// Convert \"+=\" or \"-=\" to relative numbers (#7345)\n\t\t\tif ( type === \"string\" && ( ret = rcssNum.exec( value ) ) && ret[ 1 ] ) {\n\t\t\t\tvalue = adjustCSS( elem, name, ret );\n\n\t\t\t\t// Fixes bug #9237\n\t\t\t\ttype = \"number\";\n\t\t\t}\n\n\t\t\t// Make sure that null and NaN values aren't set (#7116)\n\t\t\tif ( value == null || value !== value ) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\t// If a number was passed in, add the unit (except for certain CSS properties)\n\t\t\tif ( type === \"number\" ) {\n\t\t\t\tvalue += ret && ret[ 3 ] || ( jQuery.cssNumber[ origName ] ? \"\" : \"px\" );\n\t\t\t}\n\n\t\t\t// background-* props affect original clone's values\n\t\t\tif ( !support.clearCloneStyle && value === \"\" && name.indexOf( \"background\" ) === 0 ) {\n\t\t\t\tstyle[ name ] = \"inherit\";\n\t\t\t}\n\n\t\t\t// If a hook was provided, use that value, otherwise just set the specified value\n\t\t\tif ( !hooks || !( \"set\" in hooks ) ||\n\t\t\t\t( value = hooks.set( elem, value, extra ) ) !== undefined ) {\n\n\t\t\t\tif ( isCustomProp ) {\n\t\t\t\t\tstyle.setProperty( name, value );\n\t\t\t\t} else {\n\t\t\t\t\tstyle[ name ] = value;\n\t\t\t\t}\n\t\t\t}\n\n\t\t} else {\n\n\t\t\t// If a hook was provided get the non-computed value from there\n\t\t\tif ( hooks && \"get\" in hooks &&\n\t\t\t\t( ret = hooks.get( elem, false, extra ) ) !== undefined ) {\n\n\t\t\t\treturn ret;\n\t\t\t}\n\n\t\t\t// Otherwise just get the value from the style object\n\t\t\treturn style[ name ];\n\t\t}\n\t},\n\n\tcss: function( elem, name, extra, styles ) {\n\t\tvar val, num, hooks,\n\t\t\torigName = jQuery.camelCase( name ),\n\t\t\tisCustomProp = rcustomProp.test( name );\n\n\t\t// Make sure that we're working with the right name. We don't\n\t\t// want to modify the value if it is a CSS custom property\n\t\t// since they are user-defined.\n\t\tif ( !isCustomProp ) {\n\t\t\tname = finalPropName( origName );\n\t\t}\n\n\t\t// Try prefixed name followed by the unprefixed name\n\t\thooks = jQuery.cssHooks[ name ] || jQuery.cssHooks[ origName ];\n\n\t\t// If a hook was provided get the computed value from there\n\t\tif ( hooks && \"get\" in hooks ) {\n\t\t\tval = hooks.get( elem, true, extra );\n\t\t}\n\n\t\t// Otherwise, if a way to get the computed value exists, use that\n\t\tif ( val === undefined ) {\n\t\t\tval = curCSS( elem, name, styles );\n\t\t}\n\n\t\t// Convert \"normal\" to computed value\n\t\tif ( val === \"normal\" && name in cssNormalTransform ) {\n\t\t\tval = cssNormalTransform[ name ];\n\t\t}\n\n\t\t// Make numeric if forced or a qualifier was provided and val looks numeric\n\t\tif ( extra === \"\" || extra ) {\n\t\t\tnum = parseFloat( val );\n\t\t\treturn extra === true || isFinite( num ) ? num || 0 : val;\n\t\t}\n\n\t\treturn val;\n\t}\n} );\n\njQuery.each( [ \"height\", \"width\" ], function( i, name ) {\n\tjQuery.cssHooks[ name ] = {\n\t\tget: function( elem, computed, extra ) {\n\t\t\tif ( computed ) {\n\n\t\t\t\t// Certain elements can have dimension info if we invisibly show them\n\t\t\t\t// but it must have a current display style that would benefit\n\t\t\t\treturn rdisplayswap.test( jQuery.css( elem, \"display\" ) ) &&\n\n\t\t\t\t\t// Support: Safari 8+\n\t\t\t\t\t// Table columns in Safari have non-zero offsetWidth & zero\n\t\t\t\t\t// getBoundingClientRect().width unless display is changed.\n\t\t\t\t\t// Support: IE <=11 only\n\t\t\t\t\t// Running getBoundingClientRect on a disconnected node\n\t\t\t\t\t// in IE throws an error.\n\t\t\t\t\t( !elem.getClientRects().length || !elem.getBoundingClientRect().width ) ?\n\t\t\t\t\t\tswap( elem, cssShow, function() {\n\t\t\t\t\t\t\treturn getWidthOrHeight( elem, name, extra );\n\t\t\t\t\t\t} ) :\n\t\t\t\t\t\tgetWidthOrHeight( elem, name, extra );\n\t\t\t}\n\t\t},\n\n\t\tset: function( elem, value, extra ) {\n\t\t\tvar matches,\n\t\t\t\tstyles = extra && getStyles( elem ),\n\t\t\t\tsubtract = extra && augmentWidthOrHeight(\n\t\t\t\t\telem,\n\t\t\t\t\tname,\n\t\t\t\t\textra,\n\t\t\t\t\tjQuery.css( elem, \"boxSizing\", false, styles ) === \"border-box\",\n\t\t\t\t\tstyles\n\t\t\t\t);\n\n\t\t\t// Convert to pixels if value adjustment is needed\n\t\t\tif ( subtract && ( matches = rcssNum.exec( value ) ) &&\n\t\t\t\t( matches[ 3 ] || \"px\" ) !== \"px\" ) {\n\n\t\t\t\telem.style[ name ] = value;\n\t\t\t\tvalue = jQuery.css( elem, name );\n\t\t\t}\n\n\t\t\treturn setPositiveNumber( elem, value, subtract );\n\t\t}\n\t};\n} );\n\njQuery.cssHooks.marginLeft = addGetHookIf( support.reliableMarginLeft,\n\tfunction( elem, computed ) {\n\t\tif ( computed ) {\n\t\t\treturn ( parseFloat( curCSS( elem, \"marginLeft\" ) ) ||\n\t\t\t\telem.getBoundingClientRect().left -\n\t\t\t\t\tswap( elem, { marginLeft: 0 }, function() {\n\t\t\t\t\t\treturn elem.getBoundingClientRect().left;\n\t\t\t\t\t} )\n\t\t\t\t) + \"px\";\n\t\t}\n\t}\n);\n\n// These hooks are used by animate to expand properties\njQuery.each( {\n\tmargin: \"\",\n\tpadding: \"\",\n\tborder: \"Width\"\n}, function( prefix, suffix ) {\n\tjQuery.cssHooks[ prefix + suffix ] = {\n\t\texpand: function( value ) {\n\t\t\tvar i = 0,\n\t\t\t\texpanded = {},\n\n\t\t\t\t// Assumes a single number if not a string\n\t\t\t\tparts = typeof value === \"string\" ? value.split( \" \" ) : [ value ];\n\n\t\t\tfor ( ; i < 4; i++ ) {\n\t\t\t\texpanded[ prefix + cssExpand[ i ] + suffix ] =\n\t\t\t\t\tparts[ i ] || parts[ i - 2 ] || parts[ 0 ];\n\t\t\t}\n\n\t\t\treturn expanded;\n\t\t}\n\t};\n\n\tif ( !rmargin.test( prefix ) ) {\n\t\tjQuery.cssHooks[ prefix + suffix ].set = setPositiveNumber;\n\t}\n} );\n\njQuery.fn.extend( {\n\tcss: function( name, value ) {\n\t\treturn access( this, function( elem, name, value ) {\n\t\t\tvar styles, len,\n\t\t\t\tmap = {},\n\t\t\t\ti = 0;\n\n\t\t\tif ( Array.isArray( name ) ) {\n\t\t\t\tstyles = getStyles( elem );\n\t\t\t\tlen = name.length;\n\n\t\t\t\tfor ( ; i < len; i++ ) {\n\t\t\t\t\tmap[ name[ i ] ] = jQuery.css( elem, name[ i ], false, styles );\n\t\t\t\t}\n\n\t\t\t\treturn map;\n\t\t\t}\n\n\t\t\treturn value !== undefined ?\n\t\t\t\tjQuery.style( elem, name, value ) :\n\t\t\t\tjQuery.css( elem, name );\n\t\t}, name, value, arguments.length > 1 );\n\t}\n} );\n\n\nfunction Tween( elem, options, prop, end, easing ) {\n\treturn new Tween.prototype.init( elem, options, prop, end, easing );\n}\njQuery.Tween = Tween;\n\nTween.prototype = {\n\tconstructor: Tween,\n\tinit: function( elem, options, prop, end, easing, unit ) {\n\t\tthis.elem = elem;\n\t\tthis.prop = prop;\n\t\tthis.easing = easing || jQuery.easing._default;\n\t\tthis.options = options;\n\t\tthis.start = this.now = this.cur();\n\t\tthis.end = end;\n\t\tthis.unit = unit || ( jQuery.cssNumber[ prop ] ? \"\" : \"px\" );\n\t},\n\tcur: function() {\n\t\tvar hooks = Tween.propHooks[ this.prop ];\n\n\t\treturn hooks && hooks.get ?\n\t\t\thooks.get( this ) :\n\t\t\tTween.propHooks._default.get( this );\n\t},\n\trun: function( percent ) {\n\t\tvar eased,\n\t\t\thooks = Tween.propHooks[ this.prop ];\n\n\t\tif ( this.options.duration ) {\n\t\t\tthis.pos = eased = jQuery.easing[ this.easing ](\n\t\t\t\tpercent, this.options.duration * percent, 0, 1, this.options.duration\n\t\t\t);\n\t\t} else {\n\t\t\tthis.pos = eased = percent;\n\t\t}\n\t\tthis.now = ( this.end - this.start ) * eased + this.start;\n\n\t\tif ( this.options.step ) {\n\t\t\tthis.options.step.call( this.elem, this.now, this );\n\t\t}\n\n\t\tif ( hooks && hooks.set ) {\n\t\t\thooks.set( this );\n\t\t} else {\n\t\t\tTween.propHooks._default.set( this );\n\t\t}\n\t\treturn this;\n\t}\n};\n\nTween.prototype.init.prototype = Tween.prototype;\n\nTween.propHooks = {\n\t_default: {\n\t\tget: function( tween ) {\n\t\t\tvar result;\n\n\t\t\t// Use a property on the element directly when it is not a DOM element,\n\t\t\t// or when there is no matching style property that exists.\n\t\t\tif ( tween.elem.nodeType !== 1 ||\n\t\t\t\ttween.elem[ tween.prop ] != null && tween.elem.style[ tween.prop ] == null ) {\n\t\t\t\treturn tween.elem[ tween.prop ];\n\t\t\t}\n\n\t\t\t// Passing an empty string as a 3rd parameter to .css will automatically\n\t\t\t// attempt a parseFloat and fallback to a string if the parse fails.\n\t\t\t// Simple values such as \"10px\" are parsed to Float;\n\t\t\t// complex values such as \"rotate(1rad)\" are returned as-is.\n\t\t\tresult = jQuery.css( tween.elem, tween.prop, \"\" );\n\n\t\t\t// Empty strings, null, undefined and \"auto\" are converted to 0.\n\t\t\treturn !result || result === \"auto\" ? 0 : result;\n\t\t},\n\t\tset: function( tween ) {\n\n\t\t\t// Use step hook for back compat.\n\t\t\t// Use cssHook if its there.\n\t\t\t// Use .style if available and use plain properties where available.\n\t\t\tif ( jQuery.fx.step[ tween.prop ] ) {\n\t\t\t\tjQuery.fx.step[ tween.prop ]( tween );\n\t\t\t} else if ( tween.elem.nodeType === 1 &&\n\t\t\t\t( tween.elem.style[ jQuery.cssProps[ tween.prop ] ] != null ||\n\t\t\t\t\tjQuery.cssHooks[ tween.prop ] ) ) {\n\t\t\t\tjQuery.style( tween.elem, tween.prop, tween.now + tween.unit );\n\t\t\t} else {\n\t\t\t\ttween.elem[ tween.prop ] = tween.now;\n\t\t\t}\n\t\t}\n\t}\n};\n\n// Support: IE <=9 only\n// Panic based approach to setting things on disconnected nodes\nTween.propHooks.scrollTop = Tween.propHooks.scrollLeft = {\n\tset: function( tween ) {\n\t\tif ( tween.elem.nodeType && tween.elem.parentNode ) {\n\t\t\ttween.elem[ tween.prop ] = tween.now;\n\t\t}\n\t}\n};\n\njQuery.easing = {\n\tlinear: function( p ) {\n\t\treturn p;\n\t},\n\tswing: function( p ) {\n\t\treturn 0.5 - Math.cos( p * Math.PI ) / 2;\n\t},\n\t_default: \"swing\"\n};\n\njQuery.fx = Tween.prototype.init;\n\n// Back compat <1.8 extension point\njQuery.fx.step = {};\n\n\n\n\nvar\n\tfxNow, inProgress,\n\trfxtypes = /^(?:toggle|show|hide)$/,\n\trrun = /queueHooks$/;\n\nfunction schedule() {\n\tif ( inProgress ) {\n\t\tif ( document.hidden === false && window.requestAnimationFrame ) {\n\t\t\twindow.requestAnimationFrame( schedule );\n\t\t} else {\n\t\t\twindow.setTimeout( schedule, jQuery.fx.interval );\n\t\t}\n\n\t\tjQuery.fx.tick();\n\t}\n}\n\n// Animations created synchronously will run synchronously\nfunction createFxNow() {\n\twindow.setTimeout( function() {\n\t\tfxNow = undefined;\n\t} );\n\treturn ( fxNow = jQuery.now() );\n}\n\n// Generate parameters to create a standard animation\nfunction genFx( type, includeWidth ) {\n\tvar which,\n\t\ti = 0,\n\t\tattrs = { height: type };\n\n\t// If we include width, step value is 1 to do all cssExpand values,\n\t// otherwise step value is 2 to skip over Left and Right\n\tincludeWidth = includeWidth ? 1 : 0;\n\tfor ( ; i < 4; i += 2 - includeWidth ) {\n\t\twhich = cssExpand[ i ];\n\t\tattrs[ \"margin\" + which ] = attrs[ \"padding\" + which ] = type;\n\t}\n\n\tif ( includeWidth ) {\n\t\tattrs.opacity = attrs.width = type;\n\t}\n\n\treturn attrs;\n}\n\nfunction createTween( value, prop, animation ) {\n\tvar tween,\n\t\tcollection = ( Animation.tweeners[ prop ] || [] ).concat( Animation.tweeners[ \"*\" ] ),\n\t\tindex = 0,\n\t\tlength = collection.length;\n\tfor ( ; index < length; index++ ) {\n\t\tif ( ( tween = collection[ index ].call( animation, prop, value ) ) ) {\n\n\t\t\t// We're done with this property\n\t\t\treturn tween;\n\t\t}\n\t}\n}\n\nfunction defaultPrefilter( elem, props, opts ) {\n\tvar prop, value, toggle, hooks, oldfire, propTween, restoreDisplay, display,\n\t\tisBox = \"width\" in props || \"height\" in props,\n\t\tanim = this,\n\t\torig = {},\n\t\tstyle = elem.style,\n\t\thidden = elem.nodeType && isHiddenWithinTree( elem ),\n\t\tdataShow = dataPriv.get( elem, \"fxshow\" );\n\n\t// Queue-skipping animations hijack the fx hooks\n\tif ( !opts.queue ) {\n\t\thooks = jQuery._queueHooks( elem, \"fx\" );\n\t\tif ( hooks.unqueued == null ) {\n\t\t\thooks.unqueued = 0;\n\t\t\toldfire = hooks.empty.fire;\n\t\t\thooks.empty.fire = function() {\n\t\t\t\tif ( !hooks.unqueued ) {\n\t\t\t\t\toldfire();\n\t\t\t\t}\n\t\t\t};\n\t\t}\n\t\thooks.unqueued++;\n\n\t\tanim.always( function() {\n\n\t\t\t// Ensure the complete handler is called before this completes\n\t\t\tanim.always( function() {\n\t\t\t\thooks.unqueued--;\n\t\t\t\tif ( !jQuery.queue( elem, \"fx\" ).length ) {\n\t\t\t\t\thooks.empty.fire();\n\t\t\t\t}\n\t\t\t} );\n\t\t} );\n\t}\n\n\t// Detect show/hide animations\n\tfor ( prop in props ) {\n\t\tvalue = props[ prop ];\n\t\tif ( rfxtypes.test( value ) ) {\n\t\t\tdelete props[ prop ];\n\t\t\ttoggle = toggle || value === \"toggle\";\n\t\t\tif ( value === ( hidden ? \"hide\" : \"show\" ) ) {\n\n\t\t\t\t// Pretend to be hidden if this is a \"show\" and\n\t\t\t\t// there is still data from a stopped show/hide\n\t\t\t\tif ( value === \"show\" && dataShow && dataShow[ prop ] !== undefined ) {\n\t\t\t\t\thidden = true;\n\n\t\t\t\t// Ignore all other no-op show/hide data\n\t\t\t\t} else {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t}\n\t\t\torig[ prop ] = dataShow && dataShow[ prop ] || jQuery.style( elem, prop );\n\t\t}\n\t}\n\n\t// Bail out if this is a no-op like .hide().hide()\n\tpropTween = !jQuery.isEmptyObject( props );\n\tif ( !propTween && jQuery.isEmptyObject( orig ) ) {\n\t\treturn;\n\t}\n\n\t// Restrict \"overflow\" and \"display\" styles during box animations\n\tif ( isBox && elem.nodeType === 1 ) {\n\n\t\t// Support: IE <=9 - 11, Edge 12 - 13\n\t\t// Record all 3 overflow attributes because IE does not infer the shorthand\n\t\t// from identically-valued overflowX and overflowY\n\t\topts.overflow = [ style.overflow, style.overflowX, style.overflowY ];\n\n\t\t// Identify a display type, preferring old show/hide data over the CSS cascade\n\t\trestoreDisplay = dataShow && dataShow.display;\n\t\tif ( restoreDisplay == null ) {\n\t\t\trestoreDisplay = dataPriv.get( elem, \"display\" );\n\t\t}\n\t\tdisplay = jQuery.css( elem, \"display\" );\n\t\tif ( display === \"none\" ) {\n\t\t\tif ( restoreDisplay ) {\n\t\t\t\tdisplay = restoreDisplay;\n\t\t\t} else {\n\n\t\t\t\t// Get nonempty value(s) by temporarily forcing visibility\n\t\t\t\tshowHide( [ elem ], true );\n\t\t\t\trestoreDisplay = elem.style.display || restoreDisplay;\n\t\t\t\tdisplay = jQuery.css( elem, \"display\" );\n\t\t\t\tshowHide( [ elem ] );\n\t\t\t}\n\t\t}\n\n\t\t// Animate inline elements as inline-block\n\t\tif ( display === \"inline\" || display === \"inline-block\" && restoreDisplay != null ) {\n\t\t\tif ( jQuery.css( elem, \"float\" ) === \"none\" ) {\n\n\t\t\t\t// Restore the original display value at the end of pure show/hide animations\n\t\t\t\tif ( !propTween ) {\n\t\t\t\t\tanim.done( function() {\n\t\t\t\t\t\tstyle.display = restoreDisplay;\n\t\t\t\t\t} );\n\t\t\t\t\tif ( restoreDisplay == null ) {\n\t\t\t\t\t\tdisplay = style.display;\n\t\t\t\t\t\trestoreDisplay = display === \"none\" ? \"\" : display;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tstyle.display = \"inline-block\";\n\t\t\t}\n\t\t}\n\t}\n\n\tif ( opts.overflow ) {\n\t\tstyle.overflow = \"hidden\";\n\t\tanim.always( function() {\n\t\t\tstyle.overflow = opts.overflow[ 0 ];\n\t\t\tstyle.overflowX = opts.overflow[ 1 ];\n\t\t\tstyle.overflowY = opts.overflow[ 2 ];\n\t\t} );\n\t}\n\n\t// Implement show/hide animations\n\tpropTween = false;\n\tfor ( prop in orig ) {\n\n\t\t// General show/hide setup for this element animation\n\t\tif ( !propTween ) {\n\t\t\tif ( dataShow ) {\n\t\t\t\tif ( \"hidden\" in dataShow ) {\n\t\t\t\t\thidden = dataShow.hidden;\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tdataShow = dataPriv.access( elem, \"fxshow\", { display: restoreDisplay } );\n\t\t\t}\n\n\t\t\t// Store hidden/visible for toggle so `.stop().toggle()` \"reverses\"\n\t\t\tif ( toggle ) {\n\t\t\t\tdataShow.hidden = !hidden;\n\t\t\t}\n\n\t\t\t// Show elements before animating them\n\t\t\tif ( hidden ) {\n\t\t\t\tshowHide( [ elem ], true );\n\t\t\t}\n\n\t\t\t/* eslint-disable no-loop-func */\n\n\t\t\tanim.done( function() {\n\n\t\t\t/* eslint-enable no-loop-func */\n\n\t\t\t\t// The final step of a \"hide\" animation is actually hiding the element\n\t\t\t\tif ( !hidden ) {\n\t\t\t\t\tshowHide( [ elem ] );\n\t\t\t\t}\n\t\t\t\tdataPriv.remove( elem, \"fxshow\" );\n\t\t\t\tfor ( prop in orig ) {\n\t\t\t\t\tjQuery.style( elem, prop, orig[ prop ] );\n\t\t\t\t}\n\t\t\t} );\n\t\t}\n\n\t\t// Per-property setup\n\t\tpropTween = createTween( hidden ? dataShow[ prop ] : 0, prop, anim );\n\t\tif ( !( prop in dataShow ) ) {\n\t\t\tdataShow[ prop ] = propTween.start;\n\t\t\tif ( hidden ) {\n\t\t\t\tpropTween.end = propTween.start;\n\t\t\t\tpropTween.start = 0;\n\t\t\t}\n\t\t}\n\t}\n}\n\nfunction propFilter( props, specialEasing ) {\n\tvar index, name, easing, value, hooks;\n\n\t// camelCase, specialEasing and expand cssHook pass\n\tfor ( index in props ) {\n\t\tname = jQuery.camelCase( index );\n\t\teasing = specialEasing[ name ];\n\t\tvalue = props[ index ];\n\t\tif ( Array.isArray( value ) ) {\n\t\t\teasing = value[ 1 ];\n\t\t\tvalue = props[ index ] = value[ 0 ];\n\t\t}\n\n\t\tif ( index !== name ) {\n\t\t\tprops[ name ] = value;\n\t\t\tdelete props[ index ];\n\t\t}\n\n\t\thooks = jQuery.cssHooks[ name ];\n\t\tif ( hooks && \"expand\" in hooks ) {\n\t\t\tvalue = hooks.expand( value );\n\t\t\tdelete props[ name ];\n\n\t\t\t// Not quite $.extend, this won't overwrite existing keys.\n\t\t\t// Reusing 'index' because we have the correct \"name\"\n\t\t\tfor ( index in value ) {\n\t\t\t\tif ( !( index in props ) ) {\n\t\t\t\t\tprops[ index ] = value[ index ];\n\t\t\t\t\tspecialEasing[ index ] = easing;\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tspecialEasing[ name ] = easing;\n\t\t}\n\t}\n}\n\nfunction Animation( elem, properties, options ) {\n\tvar result,\n\t\tstopped,\n\t\tindex = 0,\n\t\tlength = Animation.prefilters.length,\n\t\tdeferred = jQuery.Deferred().always( function() {\n\n\t\t\t// Don't match elem in the :animated selector\n\t\t\tdelete tick.elem;\n\t\t} ),\n\t\ttick = function() {\n\t\t\tif ( stopped ) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tvar currentTime = fxNow || createFxNow(),\n\t\t\t\tremaining = Math.max( 0, animation.startTime + animation.duration - currentTime ),\n\n\t\t\t\t// Support: Android 2.3 only\n\t\t\t\t// Archaic crash bug won't allow us to use `1 - ( 0.5 || 0 )` (#12497)\n\t\t\t\ttemp = remaining / animation.duration || 0,\n\t\t\t\tpercent = 1 - temp,\n\t\t\t\tindex = 0,\n\t\t\t\tlength = animation.tweens.length;\n\n\t\t\tfor ( ; index < length; index++ ) {\n\t\t\t\tanimation.tweens[ index ].run( percent );\n\t\t\t}\n\n\t\t\tdeferred.notifyWith( elem, [ animation, percent, remaining ] );\n\n\t\t\t// If there's more to do, yield\n\t\t\tif ( percent < 1 && length ) {\n\t\t\t\treturn remaining;\n\t\t\t}\n\n\t\t\t// If this was an empty animation, synthesize a final progress notification\n\t\t\tif ( !length ) {\n\t\t\t\tdeferred.notifyWith( elem, [ animation, 1, 0 ] );\n\t\t\t}\n\n\t\t\t// Resolve the animation and report its conclusion\n\t\t\tdeferred.resolveWith( elem, [ animation ] );\n\t\t\treturn false;\n\t\t},\n\t\tanimation = deferred.promise( {\n\t\t\telem: elem,\n\t\t\tprops: jQuery.extend( {}, properties ),\n\t\t\topts: jQuery.extend( true, {\n\t\t\t\tspecialEasing: {},\n\t\t\t\teasing: jQuery.easing._default\n\t\t\t}, options ),\n\t\t\toriginalProperties: properties,\n\t\t\toriginalOptions: options,\n\t\t\tstartTime: fxNow || createFxNow(),\n\t\t\tduration: options.duration,\n\t\t\ttweens: [],\n\t\t\tcreateTween: function( prop, end ) {\n\t\t\t\tvar tween = jQuery.Tween( elem, animation.opts, prop, end,\n\t\t\t\t\t\tanimation.opts.specialEasing[ prop ] || animation.opts.easing );\n\t\t\t\tanimation.tweens.push( tween );\n\t\t\t\treturn tween;\n\t\t\t},\n\t\t\tstop: function( gotoEnd ) {\n\t\t\t\tvar index = 0,\n\n\t\t\t\t\t// If we are going to the end, we want to run all the tweens\n\t\t\t\t\t// otherwise we skip this part\n\t\t\t\t\tlength = gotoEnd ? animation.tweens.length : 0;\n\t\t\t\tif ( stopped ) {\n\t\t\t\t\treturn this;\n\t\t\t\t}\n\t\t\t\tstopped = true;\n\t\t\t\tfor ( ; index < length; index++ ) {\n\t\t\t\t\tanimation.tweens[ index ].run( 1 );\n\t\t\t\t}\n\n\t\t\t\t// Resolve when we played the last frame; otherwise, reject\n\t\t\t\tif ( gotoEnd ) {\n\t\t\t\t\tdeferred.notifyWith( elem, [ animation, 1, 0 ] );\n\t\t\t\t\tdeferred.resolveWith( elem, [ animation, gotoEnd ] );\n\t\t\t\t} else {\n\t\t\t\t\tdeferred.rejectWith( elem, [ animation, gotoEnd ] );\n\t\t\t\t}\n\t\t\t\treturn this;\n\t\t\t}\n\t\t} ),\n\t\tprops = animation.props;\n\n\tpropFilter( props, animation.opts.specialEasing );\n\n\tfor ( ; index < length; index++ ) {\n\t\tresult = Animation.prefilters[ index ].call( animation, elem, props, animation.opts );\n\t\tif ( result ) {\n\t\t\tif ( jQuery.isFunction( result.stop ) ) {\n\t\t\t\tjQuery._queueHooks( animation.elem, animation.opts.queue ).stop =\n\t\t\t\t\tjQuery.proxy( result.stop, result );\n\t\t\t}\n\t\t\treturn result;\n\t\t}\n\t}\n\n\tjQuery.map( props, createTween, animation );\n\n\tif ( jQuery.isFunction( animation.opts.start ) ) {\n\t\tanimation.opts.start.call( elem, animation );\n\t}\n\n\t// Attach callbacks from options\n\tanimation\n\t\t.progress( animation.opts.progress )\n\t\t.done( animation.opts.done, animation.opts.complete )\n\t\t.fail( animation.opts.fail )\n\t\t.always( animation.opts.always );\n\n\tjQuery.fx.timer(\n\t\tjQuery.extend( tick, {\n\t\t\telem: elem,\n\t\t\tanim: animation,\n\t\t\tqueue: animation.opts.queue\n\t\t} )\n\t);\n\n\treturn animation;\n}\n\njQuery.Animation = jQuery.extend( Animation, {\n\n\ttweeners: {\n\t\t\"*\": [ function( prop, value ) {\n\t\t\tvar tween = this.createTween( prop, value );\n\t\t\tadjustCSS( tween.elem, prop, rcssNum.exec( value ), tween );\n\t\t\treturn tween;\n\t\t} ]\n\t},\n\n\ttweener: function( props, callback ) {\n\t\tif ( jQuery.isFunction( props ) ) {\n\t\t\tcallback = props;\n\t\t\tprops = [ \"*\" ];\n\t\t} else {\n\t\t\tprops = props.match( rnothtmlwhite );\n\t\t}\n\n\t\tvar prop,\n\t\t\tindex = 0,\n\t\t\tlength = props.length;\n\n\t\tfor ( ; index < length; index++ ) {\n\t\t\tprop = props[ index ];\n\t\t\tAnimation.tweeners[ prop ] = Animation.tweeners[ prop ] || [];\n\t\t\tAnimation.tweeners[ prop ].unshift( callback );\n\t\t}\n\t},\n\n\tprefilters: [ defaultPrefilter ],\n\n\tprefilter: function( callback, prepend ) {\n\t\tif ( prepend ) {\n\t\t\tAnimation.prefilters.unshift( callback );\n\t\t} else {\n\t\t\tAnimation.prefilters.push( callback );\n\t\t}\n\t}\n} );\n\njQuery.speed = function( speed, easing, fn ) {\n\tvar opt = speed && typeof speed === \"object\" ? jQuery.extend( {}, speed ) : {\n\t\tcomplete: fn || !fn && easing ||\n\t\t\tjQuery.isFunction( speed ) && speed,\n\t\tduration: speed,\n\t\teasing: fn && easing || easing && !jQuery.isFunction( easing ) && easing\n\t};\n\n\t// Go to the end state if fx are off\n\tif ( jQuery.fx.off ) {\n\t\topt.duration = 0;\n\n\t} else {\n\t\tif ( typeof opt.duration !== \"number\" ) {\n\t\t\tif ( opt.duration in jQuery.fx.speeds ) {\n\t\t\t\topt.duration = jQuery.fx.speeds[ opt.duration ];\n\n\t\t\t} else {\n\t\t\t\topt.duration = jQuery.fx.speeds._default;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Normalize opt.queue - true/undefined/null -> \"fx\"\n\tif ( opt.queue == null || opt.queue === true ) {\n\t\topt.queue = \"fx\";\n\t}\n\n\t// Queueing\n\topt.old = opt.complete;\n\n\topt.complete = function() {\n\t\tif ( jQuery.isFunction( opt.old ) ) {\n\t\t\topt.old.call( this );\n\t\t}\n\n\t\tif ( opt.queue ) {\n\t\t\tjQuery.dequeue( this, opt.queue );\n\t\t}\n\t};\n\n\treturn opt;\n};\n\njQuery.fn.extend( {\n\tfadeTo: function( speed, to, easing, callback ) {\n\n\t\t// Show any hidden elements after setting opacity to 0\n\t\treturn this.filter( isHiddenWithinTree ).css( \"opacity\", 0 ).show()\n\n\t\t\t// Animate to the value specified\n\t\t\t.end().animate( { opacity: to }, speed, easing, callback );\n\t},\n\tanimate: function( prop, speed, easing, callback ) {\n\t\tvar empty = jQuery.isEmptyObject( prop ),\n\t\t\toptall = jQuery.speed( speed, easing, callback ),\n\t\t\tdoAnimation = function() {\n\n\t\t\t\t// Operate on a copy of prop so per-property easing won't be lost\n\t\t\t\tvar anim = Animation( this, jQuery.extend( {}, prop ), optall );\n\n\t\t\t\t// Empty animations, or finishing resolves immediately\n\t\t\t\tif ( empty || dataPriv.get( this, \"finish\" ) ) {\n\t\t\t\t\tanim.stop( true );\n\t\t\t\t}\n\t\t\t};\n\t\t\tdoAnimation.finish = doAnimation;\n\n\t\treturn empty || optall.queue === false ?\n\t\t\tthis.each( doAnimation ) :\n\t\t\tthis.queue( optall.queue, doAnimation );\n\t},\n\tstop: function( type, clearQueue, gotoEnd ) {\n\t\tvar stopQueue = function( hooks ) {\n\t\t\tvar stop = hooks.stop;\n\t\t\tdelete hooks.stop;\n\t\t\tstop( gotoEnd );\n\t\t};\n\n\t\tif ( typeof type !== \"string\" ) {\n\t\t\tgotoEnd = clearQueue;\n\t\t\tclearQueue = type;\n\t\t\ttype = undefined;\n\t\t}\n\t\tif ( clearQueue && type !== false ) {\n\t\t\tthis.queue( type || \"fx\", [] );\n\t\t}\n\n\t\treturn this.each( function() {\n\t\t\tvar dequeue = true,\n\t\t\t\tindex = type != null && type + \"queueHooks\",\n\t\t\t\ttimers = jQuery.timers,\n\t\t\t\tdata = dataPriv.get( this );\n\n\t\t\tif ( index ) {\n\t\t\t\tif ( data[ index ] && data[ index ].stop ) {\n\t\t\t\t\tstopQueue( data[ index ] );\n\t\t\t\t}\n\t\t\t} else {\n\t\t\t\tfor ( index in data ) {\n\t\t\t\t\tif ( data[ index ] && data[ index ].stop && rrun.test( index ) ) {\n\t\t\t\t\t\tstopQueue( data[ index ] );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tfor ( index = timers.length; index--; ) {\n\t\t\t\tif ( timers[ index ].elem === this &&\n\t\t\t\t\t( type == null || timers[ index ].queue === type ) ) {\n\n\t\t\t\t\ttimers[ index ].anim.stop( gotoEnd );\n\t\t\t\t\tdequeue = false;\n\t\t\t\t\ttimers.splice( index, 1 );\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Start the next in the queue if the last step wasn't forced.\n\t\t\t// Timers currently will call their complete callbacks, which\n\t\t\t// will dequeue but only if they were gotoEnd.\n\t\t\tif ( dequeue || !gotoEnd ) {\n\t\t\t\tjQuery.dequeue( this, type );\n\t\t\t}\n\t\t} );\n\t},\n\tfinish: function( type ) {\n\t\tif ( type !== false ) {\n\t\t\ttype = type || \"fx\";\n\t\t}\n\t\treturn this.each( function() {\n\t\t\tvar index,\n\t\t\t\tdata = dataPriv.get( this ),\n\t\t\t\tqueue = data[ type + \"queue\" ],\n\t\t\t\thooks = data[ type + \"queueHooks\" ],\n\t\t\t\ttimers = jQuery.timers,\n\t\t\t\tlength = queue ? queue.length : 0;\n\n\t\t\t// Enable finishing flag on private data\n\t\t\tdata.finish = true;\n\n\t\t\t// Empty the queue first\n\t\t\tjQuery.queue( this, type, [] );\n\n\t\t\tif ( hooks && hooks.stop ) {\n\t\t\t\thooks.stop.call( this, true );\n\t\t\t}\n\n\t\t\t// Look for any active animations, and finish them\n\t\t\tfor ( index = timers.length; index--; ) {\n\t\t\t\tif ( timers[ index ].elem === this && timers[ index ].queue === type ) {\n\t\t\t\t\ttimers[ index ].anim.stop( true );\n\t\t\t\t\ttimers.splice( index, 1 );\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Look for any animations in the old queue and finish them\n\t\t\tfor ( index = 0; index < length; index++ ) {\n\t\t\t\tif ( queue[ index ] && queue[ index ].finish ) {\n\t\t\t\t\tqueue[ index ].finish.call( this );\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Turn off finishing flag\n\t\t\tdelete data.finish;\n\t\t} );\n\t}\n} );\n\njQuery.each( [ \"toggle\", \"show\", \"hide\" ], function( i, name ) {\n\tvar cssFn = jQuery.fn[ name ];\n\tjQuery.fn[ name ] = function( speed, easing, callback ) {\n\t\treturn speed == null || typeof speed === \"boolean\" ?\n\t\t\tcssFn.apply( this, arguments ) :\n\t\t\tthis.animate( genFx( name, true ), speed, easing, callback );\n\t};\n} );\n\n// Generate shortcuts for custom animations\njQuery.each( {\n\tslideDown: genFx( \"show\" ),\n\tslideUp: genFx( \"hide\" ),\n\tslideToggle: genFx( \"toggle\" ),\n\tfadeIn: { opacity: \"show\" },\n\tfadeOut: { opacity: \"hide\" },\n\tfadeToggle: { opacity: \"toggle\" }\n}, function( name, props ) {\n\tjQuery.fn[ name ] = function( speed, easing, callback ) {\n\t\treturn this.animate( props, speed, easing, callback );\n\t};\n} );\n\njQuery.timers = [];\njQuery.fx.tick = function() {\n\tvar timer,\n\t\ti = 0,\n\t\ttimers = jQuery.timers;\n\n\tfxNow = jQuery.now();\n\n\tfor ( ; i < timers.length; i++ ) {\n\t\ttimer = timers[ i ];\n\n\t\t// Run the timer and safely remove it when done (allowing for external removal)\n\t\tif ( !timer() && timers[ i ] === timer ) {\n\t\t\ttimers.splice( i--, 1 );\n\t\t}\n\t}\n\n\tif ( !timers.length ) {\n\t\tjQuery.fx.stop();\n\t}\n\tfxNow = undefined;\n};\n\njQuery.fx.timer = function( timer ) {\n\tjQuery.timers.push( timer );\n\tjQuery.fx.start();\n};\n\njQuery.fx.interval = 13;\njQuery.fx.start = function() {\n\tif ( inProgress ) {\n\t\treturn;\n\t}\n\n\tinProgress = true;\n\tschedule();\n};\n\njQuery.fx.stop = function() {\n\tinProgress = null;\n};\n\njQuery.fx.speeds = {\n\tslow: 600,\n\tfast: 200,\n\n\t// Default speed\n\t_default: 400\n};\n\n\n// Based off of the plugin by Clint Helfers, with permission.\n// https://web.archive.org/web/20100324014747/http://blindsignals.com/index.php/2009/07/jquery-delay/\njQuery.fn.delay = function( time, type ) {\n\ttime = jQuery.fx ? jQuery.fx.speeds[ time ] || time : time;\n\ttype = type || \"fx\";\n\n\treturn this.queue( type, function( next, hooks ) {\n\t\tvar timeout = window.setTimeout( next, time );\n\t\thooks.stop = function() {\n\t\t\twindow.clearTimeout( timeout );\n\t\t};\n\t} );\n};\n\n\n( function() {\n\tvar input = document.createElement( \"input\" ),\n\t\tselect = document.createElement( \"select\" ),\n\t\topt = select.appendChild( document.createElement( \"option\" ) );\n\n\tinput.type = \"checkbox\";\n\n\t// Support: Android <=4.3 only\n\t// Default value for a checkbox should be \"on\"\n\tsupport.checkOn = input.value !== \"\";\n\n\t// Support: IE <=11 only\n\t// Must access selectedIndex to make default options select\n\tsupport.optSelected = opt.selected;\n\n\t// Support: IE <=11 only\n\t// An input loses its value after becoming a radio\n\tinput = document.createElement( \"input\" );\n\tinput.value = \"t\";\n\tinput.type = \"radio\";\n\tsupport.radioValue = input.value === \"t\";\n} )();\n\n\nvar boolHook,\n\tattrHandle = jQuery.expr.attrHandle;\n\njQuery.fn.extend( {\n\tattr: function( name, value ) {\n\t\treturn access( this, jQuery.attr, name, value, arguments.length > 1 );\n\t},\n\n\tremoveAttr: function( name ) {\n\t\treturn this.each( function() {\n\t\t\tjQuery.removeAttr( this, name );\n\t\t} );\n\t}\n} );\n\njQuery.extend( {\n\tattr: function( elem, name, value ) {\n\t\tvar ret, hooks,\n\t\t\tnType = elem.nodeType;\n\n\t\t// Don't get/set attributes on text, comment and attribute nodes\n\t\tif ( nType === 3 || nType === 8 || nType === 2 ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// Fallback to prop when attributes are not supported\n\t\tif ( typeof elem.getAttribute === \"undefined\" ) {\n\t\t\treturn jQuery.prop( elem, name, value );\n\t\t}\n\n\t\t// Attribute hooks are determined by the lowercase version\n\t\t// Grab necessary hook if one is defined\n\t\tif ( nType !== 1 || !jQuery.isXMLDoc( elem ) ) {\n\t\t\thooks = jQuery.attrHooks[ name.toLowerCase() ] ||\n\t\t\t\t( jQuery.expr.match.bool.test( name ) ? boolHook : undefined );\n\t\t}\n\n\t\tif ( value !== undefined ) {\n\t\t\tif ( value === null ) {\n\t\t\t\tjQuery.removeAttr( elem, name );\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif ( hooks && \"set\" in hooks &&\n\t\t\t\t( ret = hooks.set( elem, value, name ) ) !== undefined ) {\n\t\t\t\treturn ret;\n\t\t\t}\n\n\t\t\telem.setAttribute( name, value + \"\" );\n\t\t\treturn value;\n\t\t}\n\n\t\tif ( hooks && \"get\" in hooks && ( ret = hooks.get( elem, name ) ) !== null ) {\n\t\t\treturn ret;\n\t\t}\n\n\t\tret = jQuery.find.attr( elem, name );\n\n\t\t// Non-existent attributes return null, we normalize to undefined\n\t\treturn ret == null ? undefined : ret;\n\t},\n\n\tattrHooks: {\n\t\ttype: {\n\t\t\tset: function( elem, value ) {\n\t\t\t\tif ( !support.radioValue && value === \"radio\" &&\n\t\t\t\t\tnodeName( elem, \"input\" ) ) {\n\t\t\t\t\tvar val = elem.value;\n\t\t\t\t\telem.setAttribute( \"type\", value );\n\t\t\t\t\tif ( val ) {\n\t\t\t\t\t\telem.value = val;\n\t\t\t\t\t}\n\t\t\t\t\treturn value;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t},\n\n\tremoveAttr: function( elem, value ) {\n\t\tvar name,\n\t\t\ti = 0,\n\n\t\t\t// Attribute names can contain non-HTML whitespace characters\n\t\t\t// https://html.spec.whatwg.org/multipage/syntax.html#attributes-2\n\t\t\tattrNames = value && value.match( rnothtmlwhite );\n\n\t\tif ( attrNames && elem.nodeType === 1 ) {\n\t\t\twhile ( ( name = attrNames[ i++ ] ) ) {\n\t\t\t\telem.removeAttribute( name );\n\t\t\t}\n\t\t}\n\t}\n} );\n\n// Hooks for boolean attributes\nboolHook = {\n\tset: function( elem, value, name ) {\n\t\tif ( value === false ) {\n\n\t\t\t// Remove boolean attributes when set to false\n\t\t\tjQuery.removeAttr( elem, name );\n\t\t} else {\n\t\t\telem.setAttribute( name, name );\n\t\t}\n\t\treturn name;\n\t}\n};\n\njQuery.each( jQuery.expr.match.bool.source.match( /\\w+/g ), function( i, name ) {\n\tvar getter = attrHandle[ name ] || jQuery.find.attr;\n\n\tattrHandle[ name ] = function( elem, name, isXML ) {\n\t\tvar ret, handle,\n\t\t\tlowercaseName = name.toLowerCase();\n\n\t\tif ( !isXML ) {\n\n\t\t\t// Avoid an infinite loop by temporarily removing this function from the getter\n\t\t\thandle = attrHandle[ lowercaseName ];\n\t\t\tattrHandle[ lowercaseName ] = ret;\n\t\t\tret = getter( elem, name, isXML ) != null ?\n\t\t\t\tlowercaseName :\n\t\t\t\tnull;\n\t\t\tattrHandle[ lowercaseName ] = handle;\n\t\t}\n\t\treturn ret;\n\t};\n} );\n\n\n\n\nvar rfocusable = /^(?:input|select|textarea|button)$/i,\n\trclickable = /^(?:a|area)$/i;\n\njQuery.fn.extend( {\n\tprop: function( name, value ) {\n\t\treturn access( this, jQuery.prop, name, value, arguments.length > 1 );\n\t},\n\n\tremoveProp: function( name ) {\n\t\treturn this.each( function() {\n\t\t\tdelete this[ jQuery.propFix[ name ] || name ];\n\t\t} );\n\t}\n} );\n\njQuery.extend( {\n\tprop: function( elem, name, value ) {\n\t\tvar ret, hooks,\n\t\t\tnType = elem.nodeType;\n\n\t\t// Don't get/set properties on text, comment and attribute nodes\n\t\tif ( nType === 3 || nType === 8 || nType === 2 ) {\n\t\t\treturn;\n\t\t}\n\n\t\tif ( nType !== 1 || !jQuery.isXMLDoc( elem ) ) {\n\n\t\t\t// Fix name and attach hooks\n\t\t\tname = jQuery.propFix[ name ] || name;\n\t\t\thooks = jQuery.propHooks[ name ];\n\t\t}\n\n\t\tif ( value !== undefined ) {\n\t\t\tif ( hooks && \"set\" in hooks &&\n\t\t\t\t( ret = hooks.set( elem, value, name ) ) !== undefined ) {\n\t\t\t\treturn ret;\n\t\t\t}\n\n\t\t\treturn ( elem[ name ] = value );\n\t\t}\n\n\t\tif ( hooks && \"get\" in hooks && ( ret = hooks.get( elem, name ) ) !== null ) {\n\t\t\treturn ret;\n\t\t}\n\n\t\treturn elem[ name ];\n\t},\n\n\tpropHooks: {\n\t\ttabIndex: {\n\t\t\tget: function( elem ) {\n\n\t\t\t\t// Support: IE <=9 - 11 only\n\t\t\t\t// elem.tabIndex doesn't always return the\n\t\t\t\t// correct value when it hasn't been explicitly set\n\t\t\t\t// https://web.archive.org/web/20141116233347/http://fluidproject.org/blog/2008/01/09/getting-setting-and-removing-tabindex-values-with-javascript/\n\t\t\t\t// Use proper attribute retrieval(#12072)\n\t\t\t\tvar tabindex = jQuery.find.attr( elem, \"tabindex\" );\n\n\t\t\t\tif ( tabindex ) {\n\t\t\t\t\treturn parseInt( tabindex, 10 );\n\t\t\t\t}\n\n\t\t\t\tif (\n\t\t\t\t\trfocusable.test( elem.nodeName ) ||\n\t\t\t\t\trclickable.test( elem.nodeName ) &&\n\t\t\t\t\telem.href\n\t\t\t\t) {\n\t\t\t\t\treturn 0;\n\t\t\t\t}\n\n\t\t\t\treturn -1;\n\t\t\t}\n\t\t}\n\t},\n\n\tpropFix: {\n\t\t\"for\": \"htmlFor\",\n\t\t\"class\": \"className\"\n\t}\n} );\n\n// Support: IE <=11 only\n// Accessing the selectedIndex property\n// forces the browser to respect setting selected\n// on the option\n// The getter ensures a default option is selected\n// when in an optgroup\n// eslint rule \"no-unused-expressions\" is disabled for this code\n// since it considers such accessions noop\nif ( !support.optSelected ) {\n\tjQuery.propHooks.selected = {\n\t\tget: function( elem ) {\n\n\t\t\t/* eslint no-unused-expressions: \"off\" */\n\n\t\t\tvar parent = elem.parentNode;\n\t\t\tif ( parent && parent.parentNode ) {\n\t\t\t\tparent.parentNode.selectedIndex;\n\t\t\t}\n\t\t\treturn null;\n\t\t},\n\t\tset: function( elem ) {\n\n\t\t\t/* eslint no-unused-expressions: \"off\" */\n\n\t\t\tvar parent = elem.parentNode;\n\t\t\tif ( parent ) {\n\t\t\t\tparent.selectedIndex;\n\n\t\t\t\tif ( parent.parentNode ) {\n\t\t\t\t\tparent.parentNode.selectedIndex;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t};\n}\n\njQuery.each( [\n\t\"tabIndex\",\n\t\"readOnly\",\n\t\"maxLength\",\n\t\"cellSpacing\",\n\t\"cellPadding\",\n\t\"rowSpan\",\n\t\"colSpan\",\n\t\"useMap\",\n\t\"frameBorder\",\n\t\"contentEditable\"\n], function() {\n\tjQuery.propFix[ this.toLowerCase() ] = this;\n} );\n\n\n\n\n\t// Strip and collapse whitespace according to HTML spec\n\t// https://html.spec.whatwg.org/multipage/infrastructure.html#strip-and-collapse-whitespace\n\tfunction stripAndCollapse( value ) {\n\t\tvar tokens = value.match( rnothtmlwhite ) || [];\n\t\treturn tokens.join( \" \" );\n\t}\n\n\nfunction getClass( elem ) {\n\treturn elem.getAttribute && elem.getAttribute( \"class\" ) || \"\";\n}\n\njQuery.fn.extend( {\n\taddClass: function( value ) {\n\t\tvar classes, elem, cur, curValue, clazz, j, finalValue,\n\t\t\ti = 0;\n\n\t\tif ( jQuery.isFunction( value ) ) {\n\t\t\treturn this.each( function( j ) {\n\t\t\t\tjQuery( this ).addClass( value.call( this, j, getClass( this ) ) );\n\t\t\t} );\n\t\t}\n\n\t\tif ( typeof value === \"string\" && value ) {\n\t\t\tclasses = value.match( rnothtmlwhite ) || [];\n\n\t\t\twhile ( ( elem = this[ i++ ] ) ) {\n\t\t\t\tcurValue = getClass( elem );\n\t\t\t\tcur = elem.nodeType === 1 && ( \" \" + stripAndCollapse( curValue ) + \" \" );\n\n\t\t\t\tif ( cur ) {\n\t\t\t\t\tj = 0;\n\t\t\t\t\twhile ( ( clazz = classes[ j++ ] ) ) {\n\t\t\t\t\t\tif ( cur.indexOf( \" \" + clazz + \" \" ) < 0 ) {\n\t\t\t\t\t\t\tcur += clazz + \" \";\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\t// Only assign if different to avoid unneeded rendering.\n\t\t\t\t\tfinalValue = stripAndCollapse( cur );\n\t\t\t\t\tif ( curValue !== finalValue ) {\n\t\t\t\t\t\telem.setAttribute( \"class\", finalValue );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn this;\n\t},\n\n\tremoveClass: function( value ) {\n\t\tvar classes, elem, cur, curValue, clazz, j, finalValue,\n\t\t\ti = 0;\n\n\t\tif ( jQuery.isFunction( value ) ) {\n\t\t\treturn this.each( function( j ) {\n\t\t\t\tjQuery( this ).removeClass( value.call( this, j, getClass( this ) ) );\n\t\t\t} );\n\t\t}\n\n\t\tif ( !arguments.length ) {\n\t\t\treturn this.attr( \"class\", \"\" );\n\t\t}\n\n\t\tif ( typeof value === \"string\" && value ) {\n\t\t\tclasses = value.match( rnothtmlwhite ) || [];\n\n\t\t\twhile ( ( elem = this[ i++ ] ) ) {\n\t\t\t\tcurValue = getClass( elem );\n\n\t\t\t\t// This expression is here for better compressibility (see addClass)\n\t\t\t\tcur = elem.nodeType === 1 && ( \" \" + stripAndCollapse( curValue ) + \" \" );\n\n\t\t\t\tif ( cur ) {\n\t\t\t\t\tj = 0;\n\t\t\t\t\twhile ( ( clazz = classes[ j++ ] ) ) {\n\n\t\t\t\t\t\t// Remove *all* instances\n\t\t\t\t\t\twhile ( cur.indexOf( \" \" + clazz + \" \" ) > -1 ) {\n\t\t\t\t\t\t\tcur = cur.replace( \" \" + clazz + \" \", \" \" );\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\n\t\t\t\t\t// Only assign if different to avoid unneeded rendering.\n\t\t\t\t\tfinalValue = stripAndCollapse( cur );\n\t\t\t\t\tif ( curValue !== finalValue ) {\n\t\t\t\t\t\telem.setAttribute( \"class\", finalValue );\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn this;\n\t},\n\n\ttoggleClass: function( value, stateVal ) {\n\t\tvar type = typeof value;\n\n\t\tif ( typeof stateVal === \"boolean\" && type === \"string\" ) {\n\t\t\treturn stateVal ? this.addClass( value ) : this.removeClass( value );\n\t\t}\n\n\t\tif ( jQuery.isFunction( value ) ) {\n\t\t\treturn this.each( function( i ) {\n\t\t\t\tjQuery( this ).toggleClass(\n\t\t\t\t\tvalue.call( this, i, getClass( this ), stateVal ),\n\t\t\t\t\tstateVal\n\t\t\t\t);\n\t\t\t} );\n\t\t}\n\n\t\treturn this.each( function() {\n\t\t\tvar className, i, self, classNames;\n\n\t\t\tif ( type === \"string\" ) {\n\n\t\t\t\t// Toggle individual class names\n\t\t\t\ti = 0;\n\t\t\t\tself = jQuery( this );\n\t\t\t\tclassNames = value.match( rnothtmlwhite ) || [];\n\n\t\t\t\twhile ( ( className = classNames[ i++ ] ) ) {\n\n\t\t\t\t\t// Check each className given, space separated list\n\t\t\t\t\tif ( self.hasClass( className ) ) {\n\t\t\t\t\t\tself.removeClass( className );\n\t\t\t\t\t} else {\n\t\t\t\t\t\tself.addClass( className );\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t// Toggle whole class name\n\t\t\t} else if ( value === undefined || type === \"boolean\" ) {\n\t\t\t\tclassName = getClass( this );\n\t\t\t\tif ( className ) {\n\n\t\t\t\t\t// Store className if set\n\t\t\t\t\tdataPriv.set( this, \"__className__\", className );\n\t\t\t\t}\n\n\t\t\t\t// If the element has a class name or if we're passed `false`,\n\t\t\t\t// then remove the whole classname (if there was one, the above saved it).\n\t\t\t\t// Otherwise bring back whatever was previously saved (if anything),\n\t\t\t\t// falling back to the empty string if nothing was stored.\n\t\t\t\tif ( this.setAttribute ) {\n\t\t\t\t\tthis.setAttribute( \"class\",\n\t\t\t\t\t\tclassName || value === false ?\n\t\t\t\t\t\t\"\" :\n\t\t\t\t\t\tdataPriv.get( this, \"__className__\" ) || \"\"\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t}\n\t\t} );\n\t},\n\n\thasClass: function( selector ) {\n\t\tvar className, elem,\n\t\t\ti = 0;\n\n\t\tclassName = \" \" + selector + \" \";\n\t\twhile ( ( elem = this[ i++ ] ) ) {\n\t\t\tif ( elem.nodeType === 1 &&\n\t\t\t\t( \" \" + stripAndCollapse( getClass( elem ) ) + \" \" ).indexOf( className ) > -1 ) {\n\t\t\t\t\treturn true;\n\t\t\t}\n\t\t}\n\n\t\treturn false;\n\t}\n} );\n\n\n\n\nvar rreturn = /\\r/g;\n\njQuery.fn.extend( {\n\tval: function( value ) {\n\t\tvar hooks, ret, isFunction,\n\t\t\telem = this[ 0 ];\n\n\t\tif ( !arguments.length ) {\n\t\t\tif ( elem ) {\n\t\t\t\thooks = jQuery.valHooks[ elem.type ] ||\n\t\t\t\t\tjQuery.valHooks[ elem.nodeName.toLowerCase() ];\n\n\t\t\t\tif ( hooks &&\n\t\t\t\t\t\"get\" in hooks &&\n\t\t\t\t\t( ret = hooks.get( elem, \"value\" ) ) !== undefined\n\t\t\t\t) {\n\t\t\t\t\treturn ret;\n\t\t\t\t}\n\n\t\t\t\tret = elem.value;\n\n\t\t\t\t// Handle most common string cases\n\t\t\t\tif ( typeof ret === \"string\" ) {\n\t\t\t\t\treturn ret.replace( rreturn, \"\" );\n\t\t\t\t}\n\n\t\t\t\t// Handle cases where value is null/undef or number\n\t\t\t\treturn ret == null ? \"\" : ret;\n\t\t\t}\n\n\t\t\treturn;\n\t\t}\n\n\t\tisFunction = jQuery.isFunction( value );\n\n\t\treturn this.each( function( i ) {\n\t\t\tvar val;\n\n\t\t\tif ( this.nodeType !== 1 ) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tif ( isFunction ) {\n\t\t\t\tval = value.call( this, i, jQuery( this ).val() );\n\t\t\t} else {\n\t\t\t\tval = value;\n\t\t\t}\n\n\t\t\t// Treat null/undefined as \"\"; convert numbers to string\n\t\t\tif ( val == null ) {\n\t\t\t\tval = \"\";\n\n\t\t\t} else if ( typeof val === \"number\" ) {\n\t\t\t\tval += \"\";\n\n\t\t\t} else if ( Array.isArray( val ) ) {\n\t\t\t\tval = jQuery.map( val, function( value ) {\n\t\t\t\t\treturn value == null ? \"\" : value + \"\";\n\t\t\t\t} );\n\t\t\t}\n\n\t\t\thooks = jQuery.valHooks[ this.type ] || jQuery.valHooks[ this.nodeName.toLowerCase() ];\n\n\t\t\t// If set returns undefined, fall back to normal setting\n\t\t\tif ( !hooks || !( \"set\" in hooks ) || hooks.set( this, val, \"value\" ) === undefined ) {\n\t\t\t\tthis.value = val;\n\t\t\t}\n\t\t} );\n\t}\n} );\n\njQuery.extend( {\n\tvalHooks: {\n\t\toption: {\n\t\t\tget: function( elem ) {\n\n\t\t\t\tvar val = jQuery.find.attr( elem, \"value\" );\n\t\t\t\treturn val != null ?\n\t\t\t\t\tval :\n\n\t\t\t\t\t// Support: IE <=10 - 11 only\n\t\t\t\t\t// option.text throws exceptions (#14686, #14858)\n\t\t\t\t\t// Strip and collapse whitespace\n\t\t\t\t\t// https://html.spec.whatwg.org/#strip-and-collapse-whitespace\n\t\t\t\t\tstripAndCollapse( jQuery.text( elem ) );\n\t\t\t}\n\t\t},\n\t\tselect: {\n\t\t\tget: function( elem ) {\n\t\t\t\tvar value, option, i,\n\t\t\t\t\toptions = elem.options,\n\t\t\t\t\tindex = elem.selectedIndex,\n\t\t\t\t\tone = elem.type === \"select-one\",\n\t\t\t\t\tvalues = one ? null : [],\n\t\t\t\t\tmax = one ? index + 1 : options.length;\n\n\t\t\t\tif ( index < 0 ) {\n\t\t\t\t\ti = max;\n\n\t\t\t\t} else {\n\t\t\t\t\ti = one ? index : 0;\n\t\t\t\t}\n\n\t\t\t\t// Loop through all the selected options\n\t\t\t\tfor ( ; i < max; i++ ) {\n\t\t\t\t\toption = options[ i ];\n\n\t\t\t\t\t// Support: IE <=9 only\n\t\t\t\t\t// IE8-9 doesn't update selected after form reset (#2551)\n\t\t\t\t\tif ( ( option.selected || i === index ) &&\n\n\t\t\t\t\t\t\t// Don't return options that are disabled or in a disabled optgroup\n\t\t\t\t\t\t\t!option.disabled &&\n\t\t\t\t\t\t\t( !option.parentNode.disabled ||\n\t\t\t\t\t\t\t\t!nodeName( option.parentNode, \"optgroup\" ) ) ) {\n\n\t\t\t\t\t\t// Get the specific value for the option\n\t\t\t\t\t\tvalue = jQuery( option ).val();\n\n\t\t\t\t\t\t// We don't need an array for one selects\n\t\t\t\t\t\tif ( one ) {\n\t\t\t\t\t\t\treturn value;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// Multi-Selects return an array\n\t\t\t\t\t\tvalues.push( value );\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\treturn values;\n\t\t\t},\n\n\t\t\tset: function( elem, value ) {\n\t\t\t\tvar optionSet, option,\n\t\t\t\t\toptions = elem.options,\n\t\t\t\t\tvalues = jQuery.makeArray( value ),\n\t\t\t\t\ti = options.length;\n\n\t\t\t\twhile ( i-- ) {\n\t\t\t\t\toption = options[ i ];\n\n\t\t\t\t\t/* eslint-disable no-cond-assign */\n\n\t\t\t\t\tif ( option.selected =\n\t\t\t\t\t\tjQuery.inArray( jQuery.valHooks.option.get( option ), values ) > -1\n\t\t\t\t\t) {\n\t\t\t\t\t\toptionSet = true;\n\t\t\t\t\t}\n\n\t\t\t\t\t/* eslint-enable no-cond-assign */\n\t\t\t\t}\n\n\t\t\t\t// Force browsers to behave consistently when non-matching value is set\n\t\t\t\tif ( !optionSet ) {\n\t\t\t\t\telem.selectedIndex = -1;\n\t\t\t\t}\n\t\t\t\treturn values;\n\t\t\t}\n\t\t}\n\t}\n} );\n\n// Radios and checkboxes getter/setter\njQuery.each( [ \"radio\", \"checkbox\" ], function() {\n\tjQuery.valHooks[ this ] = {\n\t\tset: function( elem, value ) {\n\t\t\tif ( Array.isArray( value ) ) {\n\t\t\t\treturn ( elem.checked = jQuery.inArray( jQuery( elem ).val(), value ) > -1 );\n\t\t\t}\n\t\t}\n\t};\n\tif ( !support.checkOn ) {\n\t\tjQuery.valHooks[ this ].get = function( elem ) {\n\t\t\treturn elem.getAttribute( \"value\" ) === null ? \"on\" : elem.value;\n\t\t};\n\t}\n} );\n\n\n\n\n// Return jQuery for attributes-only inclusion\n\n\nvar rfocusMorph = /^(?:focusinfocus|focusoutblur)$/;\n\njQuery.extend( jQuery.event, {\n\n\ttrigger: function( event, data, elem, onlyHandlers ) {\n\n\t\tvar i, cur, tmp, bubbleType, ontype, handle, special,\n\t\t\teventPath = [ elem || document ],\n\t\t\ttype = hasOwn.call( event, \"type\" ) ? event.type : event,\n\t\t\tnamespaces = hasOwn.call( event, \"namespace\" ) ? event.namespace.split( \".\" ) : [];\n\n\t\tcur = tmp = elem = elem || document;\n\n\t\t// Don't do events on text and comment nodes\n\t\tif ( elem.nodeType === 3 || elem.nodeType === 8 ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// focus/blur morphs to focusin/out; ensure we're not firing them right now\n\t\tif ( rfocusMorph.test( type + jQuery.event.triggered ) ) {\n\t\t\treturn;\n\t\t}\n\n\t\tif ( type.indexOf( \".\" ) > -1 ) {\n\n\t\t\t// Namespaced trigger; create a regexp to match event type in handle()\n\t\t\tnamespaces = type.split( \".\" );\n\t\t\ttype = namespaces.shift();\n\t\t\tnamespaces.sort();\n\t\t}\n\t\tontype = type.indexOf( \":\" ) < 0 && \"on\" + type;\n\n\t\t// Caller can pass in a jQuery.Event object, Object, or just an event type string\n\t\tevent = event[ jQuery.expando ] ?\n\t\t\tevent :\n\t\t\tnew jQuery.Event( type, typeof event === \"object\" && event );\n\n\t\t// Trigger bitmask: & 1 for native handlers; & 2 for jQuery (always true)\n\t\tevent.isTrigger = onlyHandlers ? 2 : 3;\n\t\tevent.namespace = namespaces.join( \".\" );\n\t\tevent.rnamespace = event.namespace ?\n\t\t\tnew RegExp( \"(^|\\\\.)\" + namespaces.join( \"\\\\.(?:.*\\\\.|)\" ) + \"(\\\\.|$)\" ) :\n\t\t\tnull;\n\n\t\t// Clean up the event in case it is being reused\n\t\tevent.result = undefined;\n\t\tif ( !event.target ) {\n\t\t\tevent.target = elem;\n\t\t}\n\n\t\t// Clone any incoming data and prepend the event, creating the handler arg list\n\t\tdata = data == null ?\n\t\t\t[ event ] :\n\t\t\tjQuery.makeArray( data, [ event ] );\n\n\t\t// Allow special events to draw outside the lines\n\t\tspecial = jQuery.event.special[ type ] || {};\n\t\tif ( !onlyHandlers && special.trigger && special.trigger.apply( elem, data ) === false ) {\n\t\t\treturn;\n\t\t}\n\n\t\t// Determine event propagation path in advance, per W3C events spec (#9951)\n\t\t// Bubble up to document, then to window; watch for a global ownerDocument var (#9724)\n\t\tif ( !onlyHandlers && !special.noBubble && !jQuery.isWindow( elem ) ) {\n\n\t\t\tbubbleType = special.delegateType || type;\n\t\t\tif ( !rfocusMorph.test( bubbleType + type ) ) {\n\t\t\t\tcur = cur.parentNode;\n\t\t\t}\n\t\t\tfor ( ; cur; cur = cur.parentNode ) {\n\t\t\t\teventPath.push( cur );\n\t\t\t\ttmp = cur;\n\t\t\t}\n\n\t\t\t// Only add window if we got to document (e.g., not plain obj or detached DOM)\n\t\t\tif ( tmp === ( elem.ownerDocument || document ) ) {\n\t\t\t\teventPath.push( tmp.defaultView || tmp.parentWindow || window );\n\t\t\t}\n\t\t}\n\n\t\t// Fire handlers on the event path\n\t\ti = 0;\n\t\twhile ( ( cur = eventPath[ i++ ] ) && !event.isPropagationStopped() ) {\n\n\t\t\tevent.type = i > 1 ?\n\t\t\t\tbubbleType :\n\t\t\t\tspecial.bindType || type;\n\n\t\t\t// jQuery handler\n\t\t\thandle = ( dataPriv.get( cur, \"events\" ) || {} )[ event.type ] &&\n\t\t\t\tdataPriv.get( cur, \"handle\" );\n\t\t\tif ( handle ) {\n\t\t\t\thandle.apply( cur, data );\n\t\t\t}\n\n\t\t\t// Native handler\n\t\t\thandle = ontype && cur[ ontype ];\n\t\t\tif ( handle && handle.apply && acceptData( cur ) ) {\n\t\t\t\tevent.result = handle.apply( cur, data );\n\t\t\t\tif ( event.result === false ) {\n\t\t\t\t\tevent.preventDefault();\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tevent.type = type;\n\n\t\t// If nobody prevented the default action, do it now\n\t\tif ( !onlyHandlers && !event.isDefaultPrevented() ) {\n\n\t\t\tif ( ( !special._default ||\n\t\t\t\tspecial._default.apply( eventPath.pop(), data ) === false ) &&\n\t\t\t\tacceptData( elem ) ) {\n\n\t\t\t\t// Call a native DOM method on the target with the same name as the event.\n\t\t\t\t// Don't do default actions on window, that's where global variables be (#6170)\n\t\t\t\tif ( ontype && jQuery.isFunction( elem[ type ] ) && !jQuery.isWindow( elem ) ) {\n\n\t\t\t\t\t// Don't re-trigger an onFOO event when we call its FOO() method\n\t\t\t\t\ttmp = elem[ ontype ];\n\n\t\t\t\t\tif ( tmp ) {\n\t\t\t\t\t\telem[ ontype ] = null;\n\t\t\t\t\t}\n\n\t\t\t\t\t// Prevent re-triggering of the same event, since we already bubbled it above\n\t\t\t\t\tjQuery.event.triggered = type;\n\t\t\t\t\telem[ type ]();\n\t\t\t\t\tjQuery.event.triggered = undefined;\n\n\t\t\t\t\tif ( tmp ) {\n\t\t\t\t\t\telem[ ontype ] = tmp;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn event.result;\n\t},\n\n\t// Piggyback on a donor event to simulate a different one\n\t// Used only for `focus(in | out)` events\n\tsimulate: function( type, elem, event ) {\n\t\tvar e = jQuery.extend(\n\t\t\tnew jQuery.Event(),\n\t\t\tevent,\n\t\t\t{\n\t\t\t\ttype: type,\n\t\t\t\tisSimulated: true\n\t\t\t}\n\t\t);\n\n\t\tjQuery.event.trigger( e, null, elem );\n\t}\n\n} );\n\njQuery.fn.extend( {\n\n\ttrigger: function( type, data ) {\n\t\treturn this.each( function() {\n\t\t\tjQuery.event.trigger( type, data, this );\n\t\t} );\n\t},\n\ttriggerHandler: function( type, data ) {\n\t\tvar elem = this[ 0 ];\n\t\tif ( elem ) {\n\t\t\treturn jQuery.event.trigger( type, data, elem, true );\n\t\t}\n\t}\n} );\n\n\njQuery.each( ( \"blur focus focusin focusout resize scroll click dblclick \" +\n\t\"mousedown mouseup mousemove mouseover mouseout mouseenter mouseleave \" +\n\t\"change select submit keydown keypress keyup contextmenu\" ).split( \" \" ),\n\tfunction( i, name ) {\n\n\t// Handle event binding\n\tjQuery.fn[ name ] = function( data, fn ) {\n\t\treturn arguments.length > 0 ?\n\t\t\tthis.on( name, null, data, fn ) :\n\t\t\tthis.trigger( name );\n\t};\n} );\n\njQuery.fn.extend( {\n\thover: function( fnOver, fnOut ) {\n\t\treturn this.mouseenter( fnOver ).mouseleave( fnOut || fnOver );\n\t}\n} );\n\n\n\n\nsupport.focusin = \"onfocusin\" in window;\n\n\n// Support: Firefox <=44\n// Firefox doesn't have focus(in | out) events\n// Related ticket - https://bugzilla.mozilla.org/show_bug.cgi?id=687787\n//\n// Support: Chrome <=48 - 49, Safari <=9.0 - 9.1\n// focus(in | out) events fire after focus & blur events,\n// which is spec violation - http://www.w3.org/TR/DOM-Level-3-Events/#events-focusevent-event-order\n// Related ticket - https://bugs.chromium.org/p/chromium/issues/detail?id=449857\nif ( !support.focusin ) {\n\tjQuery.each( { focus: \"focusin\", blur: \"focusout\" }, function( orig, fix ) {\n\n\t\t// Attach a single capturing handler on the document while someone wants focusin/focusout\n\t\tvar handler = function( event ) {\n\t\t\tjQuery.event.simulate( fix, event.target, jQuery.event.fix( event ) );\n\t\t};\n\n\t\tjQuery.event.special[ fix ] = {\n\t\t\tsetup: function() {\n\t\t\t\tvar doc = this.ownerDocument || this,\n\t\t\t\t\tattaches = dataPriv.access( doc, fix );\n\n\t\t\t\tif ( !attaches ) {\n\t\t\t\t\tdoc.addEventListener( orig, handler, true );\n\t\t\t\t}\n\t\t\t\tdataPriv.access( doc, fix, ( attaches || 0 ) + 1 );\n\t\t\t},\n\t\t\tteardown: function() {\n\t\t\t\tvar doc = this.ownerDocument || this,\n\t\t\t\t\tattaches = dataPriv.access( doc, fix ) - 1;\n\n\t\t\t\tif ( !attaches ) {\n\t\t\t\t\tdoc.removeEventListener( orig, handler, true );\n\t\t\t\t\tdataPriv.remove( doc, fix );\n\n\t\t\t\t} else {\n\t\t\t\t\tdataPriv.access( doc, fix, attaches );\n\t\t\t\t}\n\t\t\t}\n\t\t};\n\t} );\n}\nvar location = window.location;\n\nvar nonce = jQuery.now();\n\nvar rquery = ( /\\?/ );\n\n\n\n// Cross-browser xml parsing\njQuery.parseXML = function( data ) {\n\tvar xml;\n\tif ( !data || typeof data !== \"string\" ) {\n\t\treturn null;\n\t}\n\n\t// Support: IE 9 - 11 only\n\t// IE throws on parseFromString with invalid input.\n\ttry {\n\t\txml = ( new window.DOMParser() ).parseFromString( data, \"text/xml\" );\n\t} catch ( e ) {\n\t\txml = undefined;\n\t}\n\n\tif ( !xml || xml.getElementsByTagName( \"parsererror\" ).length ) {\n\t\tjQuery.error( \"Invalid XML: \" + data );\n\t}\n\treturn xml;\n};\n\n\nvar\n\trbracket = /\\[\\]$/,\n\trCRLF = /\\r?\\n/g,\n\trsubmitterTypes = /^(?:submit|button|image|reset|file)$/i,\n\trsubmittable = /^(?:input|select|textarea|keygen)/i;\n\nfunction buildParams( prefix, obj, traditional, add ) {\n\tvar name;\n\n\tif ( Array.isArray( obj ) ) {\n\n\t\t// Serialize array item.\n\t\tjQuery.each( obj, function( i, v ) {\n\t\t\tif ( traditional || rbracket.test( prefix ) ) {\n\n\t\t\t\t// Treat each array item as a scalar.\n\t\t\t\tadd( prefix, v );\n\n\t\t\t} else {\n\n\t\t\t\t// Item is non-scalar (array or object), encode its numeric index.\n\t\t\t\tbuildParams(\n\t\t\t\t\tprefix + \"[\" + ( typeof v === \"object\" && v != null ? i : \"\" ) + \"]\",\n\t\t\t\t\tv,\n\t\t\t\t\ttraditional,\n\t\t\t\t\tadd\n\t\t\t\t);\n\t\t\t}\n\t\t} );\n\n\t} else if ( !traditional && jQuery.type( obj ) === \"object\" ) {\n\n\t\t// Serialize object item.\n\t\tfor ( name in obj ) {\n\t\t\tbuildParams( prefix + \"[\" + name + \"]\", obj[ name ], traditional, add );\n\t\t}\n\n\t} else {\n\n\t\t// Serialize scalar item.\n\t\tadd( prefix, obj );\n\t}\n}\n\n// Serialize an array of form elements or a set of\n// key/values into a query string\njQuery.param = function( a, traditional ) {\n\tvar prefix,\n\t\ts = [],\n\t\tadd = function( key, valueOrFunction ) {\n\n\t\t\t// If value is a function, invoke it and use its return value\n\t\t\tvar value = jQuery.isFunction( valueOrFunction ) ?\n\t\t\t\tvalueOrFunction() :\n\t\t\t\tvalueOrFunction;\n\n\t\t\ts[ s.length ] = encodeURIComponent( key ) + \"=\" +\n\t\t\t\tencodeURIComponent( value == null ? \"\" : value );\n\t\t};\n\n\t// If an array was passed in, assume that it is an array of form elements.\n\tif ( Array.isArray( a ) || ( a.jquery && !jQuery.isPlainObject( a ) ) ) {\n\n\t\t// Serialize the form elements\n\t\tjQuery.each( a, function() {\n\t\t\tadd( this.name, this.value );\n\t\t} );\n\n\t} else {\n\n\t\t// If traditional, encode the \"old\" way (the way 1.3.2 or older\n\t\t// did it), otherwise encode params recursively.\n\t\tfor ( prefix in a ) {\n\t\t\tbuildParams( prefix, a[ prefix ], traditional, add );\n\t\t}\n\t}\n\n\t// Return the resulting serialization\n\treturn s.join( \"&\" );\n};\n\njQuery.fn.extend( {\n\tserialize: function() {\n\t\treturn jQuery.param( this.serializeArray() );\n\t},\n\tserializeArray: function() {\n\t\treturn this.map( function() {\n\n\t\t\t// Can add propHook for \"elements\" to filter or add form elements\n\t\t\tvar elements = jQuery.prop( this, \"elements\" );\n\t\t\treturn elements ? jQuery.makeArray( elements ) : this;\n\t\t} )\n\t\t.filter( function() {\n\t\t\tvar type = this.type;\n\n\t\t\t// Use .is( \":disabled\" ) so that fieldset[disabled] works\n\t\t\treturn this.name && !jQuery( this ).is( \":disabled\" ) &&\n\t\t\t\trsubmittable.test( this.nodeName ) && !rsubmitterTypes.test( type ) &&\n\t\t\t\t( this.checked || !rcheckableType.test( type ) );\n\t\t} )\n\t\t.map( function( i, elem ) {\n\t\t\tvar val = jQuery( this ).val();\n\n\t\t\tif ( val == null ) {\n\t\t\t\treturn null;\n\t\t\t}\n\n\t\t\tif ( Array.isArray( val ) ) {\n\t\t\t\treturn jQuery.map( val, function( val ) {\n\t\t\t\t\treturn { name: elem.name, value: val.replace( rCRLF, \"\\r\\n\" ) };\n\t\t\t\t} );\n\t\t\t}\n\n\t\t\treturn { name: elem.name, value: val.replace( rCRLF, \"\\r\\n\" ) };\n\t\t} ).get();\n\t}\n} );\n\n\nvar\n\tr20 = /%20/g,\n\trhash = /#.*$/,\n\trantiCache = /([?&])_=[^&]*/,\n\trheaders = /^(.*?):[ \\t]*([^\\r\\n]*)$/mg,\n\n\t// #7653, #8125, #8152: local protocol detection\n\trlocalProtocol = /^(?:about|app|app-storage|.+-extension|file|res|widget):$/,\n\trnoContent = /^(?:GET|HEAD)$/,\n\trprotocol = /^\\/\\//,\n\n\t/* Prefilters\n\t * 1) They are useful to introduce custom dataTypes (see ajax/jsonp.js for an example)\n\t * 2) These are called:\n\t * - BEFORE asking for a transport\n\t * - AFTER param serialization (s.data is a string if s.processData is true)\n\t * 3) key is the dataType\n\t * 4) the catchall symbol \"*\" can be used\n\t * 5) execution will start with transport dataType and THEN continue down to \"*\" if needed\n\t */\n\tprefilters = {},\n\n\t/* Transports bindings\n\t * 1) key is the dataType\n\t * 2) the catchall symbol \"*\" can be used\n\t * 3) selection will start with transport dataType and THEN go to \"*\" if needed\n\t */\n\ttransports = {},\n\n\t// Avoid comment-prolog char sequence (#10098); must appease lint and evade compression\n\tallTypes = \"*/\".concat( \"*\" ),\n\n\t// Anchor tag for parsing the document origin\n\toriginAnchor = document.createElement( \"a\" );\n\toriginAnchor.href = location.href;\n\n// Base \"constructor\" for jQuery.ajaxPrefilter and jQuery.ajaxTransport\nfunction addToPrefiltersOrTransports( structure ) {\n\n\t// dataTypeExpression is optional and defaults to \"*\"\n\treturn function( dataTypeExpression, func ) {\n\n\t\tif ( typeof dataTypeExpression !== \"string\" ) {\n\t\t\tfunc = dataTypeExpression;\n\t\t\tdataTypeExpression = \"*\";\n\t\t}\n\n\t\tvar dataType,\n\t\t\ti = 0,\n\t\t\tdataTypes = dataTypeExpression.toLowerCase().match( rnothtmlwhite ) || [];\n\n\t\tif ( jQuery.isFunction( func ) ) {\n\n\t\t\t// For each dataType in the dataTypeExpression\n\t\t\twhile ( ( dataType = dataTypes[ i++ ] ) ) {\n\n\t\t\t\t// Prepend if requested\n\t\t\t\tif ( dataType[ 0 ] === \"+\" ) {\n\t\t\t\t\tdataType = dataType.slice( 1 ) || \"*\";\n\t\t\t\t\t( structure[ dataType ] = structure[ dataType ] || [] ).unshift( func );\n\n\t\t\t\t// Otherwise append\n\t\t\t\t} else {\n\t\t\t\t\t( structure[ dataType ] = structure[ dataType ] || [] ).push( func );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t};\n}\n\n// Base inspection function for prefilters and transports\nfunction inspectPrefiltersOrTransports( structure, options, originalOptions, jqXHR ) {\n\n\tvar inspected = {},\n\t\tseekingTransport = ( structure === transports );\n\n\tfunction inspect( dataType ) {\n\t\tvar selected;\n\t\tinspected[ dataType ] = true;\n\t\tjQuery.each( structure[ dataType ] || [], function( _, prefilterOrFactory ) {\n\t\t\tvar dataTypeOrTransport = prefilterOrFactory( options, originalOptions, jqXHR );\n\t\t\tif ( typeof dataTypeOrTransport === \"string\" &&\n\t\t\t\t!seekingTransport && !inspected[ dataTypeOrTransport ] ) {\n\n\t\t\t\toptions.dataTypes.unshift( dataTypeOrTransport );\n\t\t\t\tinspect( dataTypeOrTransport );\n\t\t\t\treturn false;\n\t\t\t} else if ( seekingTransport ) {\n\t\t\t\treturn !( selected = dataTypeOrTransport );\n\t\t\t}\n\t\t} );\n\t\treturn selected;\n\t}\n\n\treturn inspect( options.dataTypes[ 0 ] ) || !inspected[ \"*\" ] && inspect( \"*\" );\n}\n\n// A special extend for ajax options\n// that takes \"flat\" options (not to be deep extended)\n// Fixes #9887\nfunction ajaxExtend( target, src ) {\n\tvar key, deep,\n\t\tflatOptions = jQuery.ajaxSettings.flatOptions || {};\n\n\tfor ( key in src ) {\n\t\tif ( src[ key ] !== undefined ) {\n\t\t\t( flatOptions[ key ] ? target : ( deep || ( deep = {} ) ) )[ key ] = src[ key ];\n\t\t}\n\t}\n\tif ( deep ) {\n\t\tjQuery.extend( true, target, deep );\n\t}\n\n\treturn target;\n}\n\n/* Handles responses to an ajax request:\n * - finds the right dataType (mediates between content-type and expected dataType)\n * - returns the corresponding response\n */\nfunction ajaxHandleResponses( s, jqXHR, responses ) {\n\n\tvar ct, type, finalDataType, firstDataType,\n\t\tcontents = s.contents,\n\t\tdataTypes = s.dataTypes;\n\n\t// Remove auto dataType and get content-type in the process\n\twhile ( dataTypes[ 0 ] === \"*\" ) {\n\t\tdataTypes.shift();\n\t\tif ( ct === undefined ) {\n\t\t\tct = s.mimeType || jqXHR.getResponseHeader( \"Content-Type\" );\n\t\t}\n\t}\n\n\t// Check if we're dealing with a known content-type\n\tif ( ct ) {\n\t\tfor ( type in contents ) {\n\t\t\tif ( contents[ type ] && contents[ type ].test( ct ) ) {\n\t\t\t\tdataTypes.unshift( type );\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\t}\n\n\t// Check to see if we have a response for the expected dataType\n\tif ( dataTypes[ 0 ] in responses ) {\n\t\tfinalDataType = dataTypes[ 0 ];\n\t} else {\n\n\t\t// Try convertible dataTypes\n\t\tfor ( type in responses ) {\n\t\t\tif ( !dataTypes[ 0 ] || s.converters[ type + \" \" + dataTypes[ 0 ] ] ) {\n\t\t\t\tfinalDataType = type;\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tif ( !firstDataType ) {\n\t\t\t\tfirstDataType = type;\n\t\t\t}\n\t\t}\n\n\t\t// Or just use first one\n\t\tfinalDataType = finalDataType || firstDataType;\n\t}\n\n\t// If we found a dataType\n\t// We add the dataType to the list if needed\n\t// and return the corresponding response\n\tif ( finalDataType ) {\n\t\tif ( finalDataType !== dataTypes[ 0 ] ) {\n\t\t\tdataTypes.unshift( finalDataType );\n\t\t}\n\t\treturn responses[ finalDataType ];\n\t}\n}\n\n/* Chain conversions given the request and the original response\n * Also sets the responseXXX fields on the jqXHR instance\n */\nfunction ajaxConvert( s, response, jqXHR, isSuccess ) {\n\tvar conv2, current, conv, tmp, prev,\n\t\tconverters = {},\n\n\t\t// Work with a copy of dataTypes in case we need to modify it for conversion\n\t\tdataTypes = s.dataTypes.slice();\n\n\t// Create converters map with lowercased keys\n\tif ( dataTypes[ 1 ] ) {\n\t\tfor ( conv in s.converters ) {\n\t\t\tconverters[ conv.toLowerCase() ] = s.converters[ conv ];\n\t\t}\n\t}\n\n\tcurrent = dataTypes.shift();\n\n\t// Convert to each sequential dataType\n\twhile ( current ) {\n\n\t\tif ( s.responseFields[ current ] ) {\n\t\t\tjqXHR[ s.responseFields[ current ] ] = response;\n\t\t}\n\n\t\t// Apply the dataFilter if provided\n\t\tif ( !prev && isSuccess && s.dataFilter ) {\n\t\t\tresponse = s.dataFilter( response, s.dataType );\n\t\t}\n\n\t\tprev = current;\n\t\tcurrent = dataTypes.shift();\n\n\t\tif ( current ) {\n\n\t\t\t// There's only work to do if current dataType is non-auto\n\t\t\tif ( current === \"*\" ) {\n\n\t\t\t\tcurrent = prev;\n\n\t\t\t// Convert response if prev dataType is non-auto and differs from current\n\t\t\t} else if ( prev !== \"*\" && prev !== current ) {\n\n\t\t\t\t// Seek a direct converter\n\t\t\t\tconv = converters[ prev + \" \" + current ] || converters[ \"* \" + current ];\n\n\t\t\t\t// If none found, seek a pair\n\t\t\t\tif ( !conv ) {\n\t\t\t\t\tfor ( conv2 in converters ) {\n\n\t\t\t\t\t\t// If conv2 outputs current\n\t\t\t\t\t\ttmp = conv2.split( \" \" );\n\t\t\t\t\t\tif ( tmp[ 1 ] === current ) {\n\n\t\t\t\t\t\t\t// If prev can be converted to accepted input\n\t\t\t\t\t\t\tconv = converters[ prev + \" \" + tmp[ 0 ] ] ||\n\t\t\t\t\t\t\t\tconverters[ \"* \" + tmp[ 0 ] ];\n\t\t\t\t\t\t\tif ( conv ) {\n\n\t\t\t\t\t\t\t\t// Condense equivalence converters\n\t\t\t\t\t\t\t\tif ( conv === true ) {\n\t\t\t\t\t\t\t\t\tconv = converters[ conv2 ];\n\n\t\t\t\t\t\t\t\t// Otherwise, insert the intermediate dataType\n\t\t\t\t\t\t\t\t} else if ( converters[ conv2 ] !== true ) {\n\t\t\t\t\t\t\t\t\tcurrent = tmp[ 0 ];\n\t\t\t\t\t\t\t\t\tdataTypes.unshift( tmp[ 1 ] );\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Apply converter (if not an equivalence)\n\t\t\t\tif ( conv !== true ) {\n\n\t\t\t\t\t// Unless errors are allowed to bubble, catch and return them\n\t\t\t\t\tif ( conv && s.throws ) {\n\t\t\t\t\t\tresponse = conv( response );\n\t\t\t\t\t} else {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\tresponse = conv( response );\n\t\t\t\t\t\t} catch ( e ) {\n\t\t\t\t\t\t\treturn {\n\t\t\t\t\t\t\t\tstate: \"parsererror\",\n\t\t\t\t\t\t\t\terror: conv ? e : \"No conversion from \" + prev + \" to \" + current\n\t\t\t\t\t\t\t};\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\treturn { state: \"success\", data: response };\n}\n\njQuery.extend( {\n\n\t// Counter for holding the number of active queries\n\tactive: 0,\n\n\t// Last-Modified header cache for next request\n\tlastModified: {},\n\tetag: {},\n\n\tajaxSettings: {\n\t\turl: location.href,\n\t\ttype: \"GET\",\n\t\tisLocal: rlocalProtocol.test( location.protocol ),\n\t\tglobal: true,\n\t\tprocessData: true,\n\t\tasync: true,\n\t\tcontentType: \"application/x-www-form-urlencoded; charset=UTF-8\",\n\n\t\t/*\n\t\ttimeout: 0,\n\t\tdata: null,\n\t\tdataType: null,\n\t\tusername: null,\n\t\tpassword: null,\n\t\tcache: null,\n\t\tthrows: false,\n\t\ttraditional: false,\n\t\theaders: {},\n\t\t*/\n\n\t\taccepts: {\n\t\t\t\"*\": allTypes,\n\t\t\ttext: \"text/plain\",\n\t\t\thtml: \"text/html\",\n\t\t\txml: \"application/xml, text/xml\",\n\t\t\tjson: \"application/json, text/javascript\"\n\t\t},\n\n\t\tcontents: {\n\t\t\txml: /\\bxml\\b/,\n\t\t\thtml: /\\bhtml/,\n\t\t\tjson: /\\bjson\\b/\n\t\t},\n\n\t\tresponseFields: {\n\t\t\txml: \"responseXML\",\n\t\t\ttext: \"responseText\",\n\t\t\tjson: \"responseJSON\"\n\t\t},\n\n\t\t// Data converters\n\t\t// Keys separate source (or catchall \"*\") and destination types with a single space\n\t\tconverters: {\n\n\t\t\t// Convert anything to text\n\t\t\t\"* text\": String,\n\n\t\t\t// Text to html (true = no transformation)\n\t\t\t\"text html\": true,\n\n\t\t\t// Evaluate text as a json expression\n\t\t\t\"text json\": JSON.parse,\n\n\t\t\t// Parse text as xml\n\t\t\t\"text xml\": jQuery.parseXML\n\t\t},\n\n\t\t// For options that shouldn't be deep extended:\n\t\t// you can add your own custom options here if\n\t\t// and when you create one that shouldn't be\n\t\t// deep extended (see ajaxExtend)\n\t\tflatOptions: {\n\t\t\turl: true,\n\t\t\tcontext: true\n\t\t}\n\t},\n\n\t// Creates a full fledged settings object into target\n\t// with both ajaxSettings and settings fields.\n\t// If target is omitted, writes into ajaxSettings.\n\tajaxSetup: function( target, settings ) {\n\t\treturn settings ?\n\n\t\t\t// Building a settings object\n\t\t\tajaxExtend( ajaxExtend( target, jQuery.ajaxSettings ), settings ) :\n\n\t\t\t// Extending ajaxSettings\n\t\t\tajaxExtend( jQuery.ajaxSettings, target );\n\t},\n\n\tajaxPrefilter: addToPrefiltersOrTransports( prefilters ),\n\tajaxTransport: addToPrefiltersOrTransports( transports ),\n\n\t// Main method\n\tajax: function( url, options ) {\n\n\t\t// If url is an object, simulate pre-1.5 signature\n\t\tif ( typeof url === \"object\" ) {\n\t\t\toptions = url;\n\t\t\turl = undefined;\n\t\t}\n\n\t\t// Force options to be an object\n\t\toptions = options || {};\n\n\t\tvar transport,\n\n\t\t\t// URL without anti-cache param\n\t\t\tcacheURL,\n\n\t\t\t// Response headers\n\t\t\tresponseHeadersString,\n\t\t\tresponseHeaders,\n\n\t\t\t// timeout handle\n\t\t\ttimeoutTimer,\n\n\t\t\t// Url cleanup var\n\t\t\turlAnchor,\n\n\t\t\t// Request state (becomes false upon send and true upon completion)\n\t\t\tcompleted,\n\n\t\t\t// To know if global events are to be dispatched\n\t\t\tfireGlobals,\n\n\t\t\t// Loop variable\n\t\t\ti,\n\n\t\t\t// uncached part of the url\n\t\t\tuncached,\n\n\t\t\t// Create the final options object\n\t\t\ts = jQuery.ajaxSetup( {}, options ),\n\n\t\t\t// Callbacks context\n\t\t\tcallbackContext = s.context || s,\n\n\t\t\t// Context for global events is callbackContext if it is a DOM node or jQuery collection\n\t\t\tglobalEventContext = s.context &&\n\t\t\t\t( callbackContext.nodeType || callbackContext.jquery ) ?\n\t\t\t\t\tjQuery( callbackContext ) :\n\t\t\t\t\tjQuery.event,\n\n\t\t\t// Deferreds\n\t\t\tdeferred = jQuery.Deferred(),\n\t\t\tcompleteDeferred = jQuery.Callbacks( \"once memory\" ),\n\n\t\t\t// Status-dependent callbacks\n\t\t\tstatusCode = s.statusCode || {},\n\n\t\t\t// Headers (they are sent all at once)\n\t\t\trequestHeaders = {},\n\t\t\trequestHeadersNames = {},\n\n\t\t\t// Default abort message\n\t\t\tstrAbort = \"canceled\",\n\n\t\t\t// Fake xhr\n\t\t\tjqXHR = {\n\t\t\t\treadyState: 0,\n\n\t\t\t\t// Builds headers hashtable if needed\n\t\t\t\tgetResponseHeader: function( key ) {\n\t\t\t\t\tvar match;\n\t\t\t\t\tif ( completed ) {\n\t\t\t\t\t\tif ( !responseHeaders ) {\n\t\t\t\t\t\t\tresponseHeaders = {};\n\t\t\t\t\t\t\twhile ( ( match = rheaders.exec( responseHeadersString ) ) ) {\n\t\t\t\t\t\t\t\tresponseHeaders[ match[ 1 ].toLowerCase() ] = match[ 2 ];\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\tmatch = responseHeaders[ key.toLowerCase() ];\n\t\t\t\t\t}\n\t\t\t\t\treturn match == null ? null : match;\n\t\t\t\t},\n\n\t\t\t\t// Raw string\n\t\t\t\tgetAllResponseHeaders: function() {\n\t\t\t\t\treturn completed ? responseHeadersString : null;\n\t\t\t\t},\n\n\t\t\t\t// Caches the header\n\t\t\t\tsetRequestHeader: function( name, value ) {\n\t\t\t\t\tif ( completed == null ) {\n\t\t\t\t\t\tname = requestHeadersNames[ name.toLowerCase() ] =\n\t\t\t\t\t\t\trequestHeadersNames[ name.toLowerCase() ] || name;\n\t\t\t\t\t\trequestHeaders[ name ] = value;\n\t\t\t\t\t}\n\t\t\t\t\treturn this;\n\t\t\t\t},\n\n\t\t\t\t// Overrides response content-type header\n\t\t\t\toverrideMimeType: function( type ) {\n\t\t\t\t\tif ( completed == null ) {\n\t\t\t\t\t\ts.mimeType = type;\n\t\t\t\t\t}\n\t\t\t\t\treturn this;\n\t\t\t\t},\n\n\t\t\t\t// Status-dependent callbacks\n\t\t\t\tstatusCode: function( map ) {\n\t\t\t\t\tvar code;\n\t\t\t\t\tif ( map ) {\n\t\t\t\t\t\tif ( completed ) {\n\n\t\t\t\t\t\t\t// Execute the appropriate callbacks\n\t\t\t\t\t\t\tjqXHR.always( map[ jqXHR.status ] );\n\t\t\t\t\t\t} else {\n\n\t\t\t\t\t\t\t// Lazy-add the new callbacks in a way that preserves old ones\n\t\t\t\t\t\t\tfor ( code in map ) {\n\t\t\t\t\t\t\t\tstatusCode[ code ] = [ statusCode[ code ], map[ code ] ];\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\treturn this;\n\t\t\t\t},\n\n\t\t\t\t// Cancel the request\n\t\t\t\tabort: function( statusText ) {\n\t\t\t\t\tvar finalText = statusText || strAbort;\n\t\t\t\t\tif ( transport ) {\n\t\t\t\t\t\ttransport.abort( finalText );\n\t\t\t\t\t}\n\t\t\t\t\tdone( 0, finalText );\n\t\t\t\t\treturn this;\n\t\t\t\t}\n\t\t\t};\n\n\t\t// Attach deferreds\n\t\tdeferred.promise( jqXHR );\n\n\t\t// Add protocol if not provided (prefilters might expect it)\n\t\t// Handle falsy url in the settings object (#10093: consistency with old signature)\n\t\t// We also use the url parameter if available\n\t\ts.url = ( ( url || s.url || location.href ) + \"\" )\n\t\t\t.replace( rprotocol, location.protocol + \"//\" );\n\n\t\t// Alias method option to type as per ticket #12004\n\t\ts.type = options.method || options.type || s.method || s.type;\n\n\t\t// Extract dataTypes list\n\t\ts.dataTypes = ( s.dataType || \"*\" ).toLowerCase().match( rnothtmlwhite ) || [ \"\" ];\n\n\t\t// A cross-domain request is in order when the origin doesn't match the current origin.\n\t\tif ( s.crossDomain == null ) {\n\t\t\turlAnchor = document.createElement( \"a\" );\n\n\t\t\t// Support: IE <=8 - 11, Edge 12 - 13\n\t\t\t// IE throws exception on accessing the href property if url is malformed,\n\t\t\t// e.g. http://example.com:80x/\n\t\t\ttry {\n\t\t\t\turlAnchor.href = s.url;\n\n\t\t\t\t// Support: IE <=8 - 11 only\n\t\t\t\t// Anchor's host property isn't correctly set when s.url is relative\n\t\t\t\turlAnchor.href = urlAnchor.href;\n\t\t\t\ts.crossDomain = originAnchor.protocol + \"//\" + originAnchor.host !==\n\t\t\t\t\turlAnchor.protocol + \"//\" + urlAnchor.host;\n\t\t\t} catch ( e ) {\n\n\t\t\t\t// If there is an error parsing the URL, assume it is crossDomain,\n\t\t\t\t// it can be rejected by the transport if it is invalid\n\t\t\t\ts.crossDomain = true;\n\t\t\t}\n\t\t}\n\n\t\t// Convert data if not already a string\n\t\tif ( s.data && s.processData && typeof s.data !== \"string\" ) {\n\t\t\ts.data = jQuery.param( s.data, s.traditional );\n\t\t}\n\n\t\t// Apply prefilters\n\t\tinspectPrefiltersOrTransports( prefilters, s, options, jqXHR );\n\n\t\t// If request was aborted inside a prefilter, stop there\n\t\tif ( completed ) {\n\t\t\treturn jqXHR;\n\t\t}\n\n\t\t// We can fire global events as of now if asked to\n\t\t// Don't fire events if jQuery.event is undefined in an AMD-usage scenario (#15118)\n\t\tfireGlobals = jQuery.event && s.global;\n\n\t\t// Watch for a new set of requests\n\t\tif ( fireGlobals && jQuery.active++ === 0 ) {\n\t\t\tjQuery.event.trigger( \"ajaxStart\" );\n\t\t}\n\n\t\t// Uppercase the type\n\t\ts.type = s.type.toUpperCase();\n\n\t\t// Determine if request has content\n\t\ts.hasContent = !rnoContent.test( s.type );\n\n\t\t// Save the URL in case we're toying with the If-Modified-Since\n\t\t// and/or If-None-Match header later on\n\t\t// Remove hash to simplify url manipulation\n\t\tcacheURL = s.url.replace( rhash, \"\" );\n\n\t\t// More options handling for requests with no content\n\t\tif ( !s.hasContent ) {\n\n\t\t\t// Remember the hash so we can put it back\n\t\t\tuncached = s.url.slice( cacheURL.length );\n\n\t\t\t// If data is available, append data to url\n\t\t\tif ( s.data ) {\n\t\t\t\tcacheURL += ( rquery.test( cacheURL ) ? \"&\" : \"?\" ) + s.data;\n\n\t\t\t\t// #9682: remove data so that it's not used in an eventual retry\n\t\t\t\tdelete s.data;\n\t\t\t}\n\n\t\t\t// Add or update anti-cache param if needed\n\t\t\tif ( s.cache === false ) {\n\t\t\t\tcacheURL = cacheURL.replace( rantiCache, \"$1\" );\n\t\t\t\tuncached = ( rquery.test( cacheURL ) ? \"&\" : \"?\" ) + \"_=\" + ( nonce++ ) + uncached;\n\t\t\t}\n\n\t\t\t// Put hash and anti-cache on the URL that will be requested (gh-1732)\n\t\t\ts.url = cacheURL + uncached;\n\n\t\t// Change '%20' to '+' if this is encoded form body content (gh-2658)\n\t\t} else if ( s.data && s.processData &&\n\t\t\t( s.contentType || \"\" ).indexOf( \"application/x-www-form-urlencoded\" ) === 0 ) {\n\t\t\ts.data = s.data.replace( r20, \"+\" );\n\t\t}\n\n\t\t// Set the If-Modified-Since and/or If-None-Match header, if in ifModified mode.\n\t\tif ( s.ifModified ) {\n\t\t\tif ( jQuery.lastModified[ cacheURL ] ) {\n\t\t\t\tjqXHR.setRequestHeader( \"If-Modified-Since\", jQuery.lastModified[ cacheURL ] );\n\t\t\t}\n\t\t\tif ( jQuery.etag[ cacheURL ] ) {\n\t\t\t\tjqXHR.setRequestHeader( \"If-None-Match\", jQuery.etag[ cacheURL ] );\n\t\t\t}\n\t\t}\n\n\t\t// Set the correct header, if data is being sent\n\t\tif ( s.data && s.hasContent && s.contentType !== false || options.contentType ) {\n\t\t\tjqXHR.setRequestHeader( \"Content-Type\", s.contentType );\n\t\t}\n\n\t\t// Set the Accepts header for the server, depending on the dataType\n\t\tjqXHR.setRequestHeader(\n\t\t\t\"Accept\",\n\t\t\ts.dataTypes[ 0 ] && s.accepts[ s.dataTypes[ 0 ] ] ?\n\t\t\t\ts.accepts[ s.dataTypes[ 0 ] ] +\n\t\t\t\t\t( s.dataTypes[ 0 ] !== \"*\" ? \", \" + allTypes + \"; q=0.01\" : \"\" ) :\n\t\t\t\ts.accepts[ \"*\" ]\n\t\t);\n\n\t\t// Check for headers option\n\t\tfor ( i in s.headers ) {\n\t\t\tjqXHR.setRequestHeader( i, s.headers[ i ] );\n\t\t}\n\n\t\t// Allow custom headers/mimetypes and early abort\n\t\tif ( s.beforeSend &&\n\t\t\t( s.beforeSend.call( callbackContext, jqXHR, s ) === false || completed ) ) {\n\n\t\t\t// Abort if not done already and return\n\t\t\treturn jqXHR.abort();\n\t\t}\n\n\t\t// Aborting is no longer a cancellation\n\t\tstrAbort = \"abort\";\n\n\t\t// Install callbacks on deferreds\n\t\tcompleteDeferred.add( s.complete );\n\t\tjqXHR.done( s.success );\n\t\tjqXHR.fail( s.error );\n\n\t\t// Get transport\n\t\ttransport = inspectPrefiltersOrTransports( transports, s, options, jqXHR );\n\n\t\t// If no transport, we auto-abort\n\t\tif ( !transport ) {\n\t\t\tdone( -1, \"No Transport\" );\n\t\t} else {\n\t\t\tjqXHR.readyState = 1;\n\n\t\t\t// Send global event\n\t\t\tif ( fireGlobals ) {\n\t\t\t\tglobalEventContext.trigger( \"ajaxSend\", [ jqXHR, s ] );\n\t\t\t}\n\n\t\t\t// If request was aborted inside ajaxSend, stop there\n\t\t\tif ( completed ) {\n\t\t\t\treturn jqXHR;\n\t\t\t}\n\n\t\t\t// Timeout\n\t\t\tif ( s.async && s.timeout > 0 ) {\n\t\t\t\ttimeoutTimer = window.setTimeout( function() {\n\t\t\t\t\tjqXHR.abort( \"timeout\" );\n\t\t\t\t}, s.timeout );\n\t\t\t}\n\n\t\t\ttry {\n\t\t\t\tcompleted = false;\n\t\t\t\ttransport.send( requestHeaders, done );\n\t\t\t} catch ( e ) {\n\n\t\t\t\t// Rethrow post-completion exceptions\n\t\t\t\tif ( completed ) {\n\t\t\t\t\tthrow e;\n\t\t\t\t}\n\n\t\t\t\t// Propagate others as results\n\t\t\t\tdone( -1, e );\n\t\t\t}\n\t\t}\n\n\t\t// Callback for when everything is done\n\t\tfunction done( status, nativeStatusText, responses, headers ) {\n\t\t\tvar isSuccess, success, error, response, modified,\n\t\t\t\tstatusText = nativeStatusText;\n\n\t\t\t// Ignore repeat invocations\n\t\t\tif ( completed ) {\n\t\t\t\treturn;\n\t\t\t}\n\n\t\t\tcompleted = true;\n\n\t\t\t// Clear timeout if it exists\n\t\t\tif ( timeoutTimer ) {\n\t\t\t\twindow.clearTimeout( timeoutTimer );\n\t\t\t}\n\n\t\t\t// Dereference transport for early garbage collection\n\t\t\t// (no matter how long the jqXHR object will be used)\n\t\t\ttransport = undefined;\n\n\t\t\t// Cache response headers\n\t\t\tresponseHeadersString = headers || \"\";\n\n\t\t\t// Set readyState\n\t\t\tjqXHR.readyState = status > 0 ? 4 : 0;\n\n\t\t\t// Determine if successful\n\t\t\tisSuccess = status >= 200 && status < 300 || status === 304;\n\n\t\t\t// Get response data\n\t\t\tif ( responses ) {\n\t\t\t\tresponse = ajaxHandleResponses( s, jqXHR, responses );\n\t\t\t}\n\n\t\t\t// Convert no matter what (that way responseXXX fields are always set)\n\t\t\tresponse = ajaxConvert( s, response, jqXHR, isSuccess );\n\n\t\t\t// If successful, handle type chaining\n\t\t\tif ( isSuccess ) {\n\n\t\t\t\t// Set the If-Modified-Since and/or If-None-Match header, if in ifModified mode.\n\t\t\t\tif ( s.ifModified ) {\n\t\t\t\t\tmodified = jqXHR.getResponseHeader( \"Last-Modified\" );\n\t\t\t\t\tif ( modified ) {\n\t\t\t\t\t\tjQuery.lastModified[ cacheURL ] = modified;\n\t\t\t\t\t}\n\t\t\t\t\tmodified = jqXHR.getResponseHeader( \"etag\" );\n\t\t\t\t\tif ( modified ) {\n\t\t\t\t\t\tjQuery.etag[ cacheURL ] = modified;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// if no content\n\t\t\t\tif ( status === 204 || s.type === \"HEAD\" ) {\n\t\t\t\t\tstatusText = \"nocontent\";\n\n\t\t\t\t// if not modified\n\t\t\t\t} else if ( status === 304 ) {\n\t\t\t\t\tstatusText = \"notmodified\";\n\n\t\t\t\t// If we have data, let's convert it\n\t\t\t\t} else {\n\t\t\t\t\tstatusText = response.state;\n\t\t\t\t\tsuccess = response.data;\n\t\t\t\t\terror = response.error;\n\t\t\t\t\tisSuccess = !error;\n\t\t\t\t}\n\t\t\t} else {\n\n\t\t\t\t// Extract error from statusText and normalize for non-aborts\n\t\t\t\terror = statusText;\n\t\t\t\tif ( status || !statusText ) {\n\t\t\t\t\tstatusText = \"error\";\n\t\t\t\t\tif ( status < 0 ) {\n\t\t\t\t\t\tstatus = 0;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Set data for the fake xhr object\n\t\t\tjqXHR.status = status;\n\t\t\tjqXHR.statusText = ( nativeStatusText || statusText ) + \"\";\n\n\t\t\t// Success/Error\n\t\t\tif ( isSuccess ) {\n\t\t\t\tdeferred.resolveWith( callbackContext, [ success, statusText, jqXHR ] );\n\t\t\t} else {\n\t\t\t\tdeferred.rejectWith( callbackContext, [ jqXHR, statusText, error ] );\n\t\t\t}\n\n\t\t\t// Status-dependent callbacks\n\t\t\tjqXHR.statusCode( statusCode );\n\t\t\tstatusCode = undefined;\n\n\t\t\tif ( fireGlobals ) {\n\t\t\t\tglobalEventContext.trigger( isSuccess ? \"ajaxSuccess\" : \"ajaxError\",\n\t\t\t\t\t[ jqXHR, s, isSuccess ? success : error ] );\n\t\t\t}\n\n\t\t\t// Complete\n\t\t\tcompleteDeferred.fireWith( callbackContext, [ jqXHR, statusText ] );\n\n\t\t\tif ( fireGlobals ) {\n\t\t\t\tglobalEventContext.trigger( \"ajaxComplete\", [ jqXHR, s ] );\n\n\t\t\t\t// Handle the global AJAX counter\n\t\t\t\tif ( !( --jQuery.active ) ) {\n\t\t\t\t\tjQuery.event.trigger( \"ajaxStop\" );\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\treturn jqXHR;\n\t},\n\n\tgetJSON: function( url, data, callback ) {\n\t\treturn jQuery.get( url, data, callback, \"json\" );\n\t},\n\n\tgetScript: function( url, callback ) {\n\t\treturn jQuery.get( url, undefined, callback, \"script\" );\n\t}\n} );\n\njQuery.each( [ \"get\", \"post\" ], function( i, method ) {\n\tjQuery[ method ] = function( url, data, callback, type ) {\n\n\t\t// Shift arguments if data argument was omitted\n\t\tif ( jQuery.isFunction( data ) ) {\n\t\t\ttype = type || callback;\n\t\t\tcallback = data;\n\t\t\tdata = undefined;\n\t\t}\n\n\t\t// The url can be an options object (which then must have .url)\n\t\treturn jQuery.ajax( jQuery.extend( {\n\t\t\turl: url,\n\t\t\ttype: method,\n\t\t\tdataType: type,\n\t\t\tdata: data,\n\t\t\tsuccess: callback\n\t\t}, jQuery.isPlainObject( url ) && url ) );\n\t};\n} );\n\n\njQuery._evalUrl = function( url ) {\n\treturn jQuery.ajax( {\n\t\turl: url,\n\n\t\t// Make this explicit, since user can override this through ajaxSetup (#11264)\n\t\ttype: \"GET\",\n\t\tdataType: \"script\",\n\t\tcache: true,\n\t\tasync: false,\n\t\tglobal: false,\n\t\t\"throws\": true\n\t} );\n};\n\n\njQuery.fn.extend( {\n\twrapAll: function( html ) {\n\t\tvar wrap;\n\n\t\tif ( this[ 0 ] ) {\n\t\t\tif ( jQuery.isFunction( html ) ) {\n\t\t\t\thtml = html.call( this[ 0 ] );\n\t\t\t}\n\n\t\t\t// The elements to wrap the target around\n\t\t\twrap = jQuery( html, this[ 0 ].ownerDocument ).eq( 0 ).clone( true );\n\n\t\t\tif ( this[ 0 ].parentNode ) {\n\t\t\t\twrap.insertBefore( this[ 0 ] );\n\t\t\t}\n\n\t\t\twrap.map( function() {\n\t\t\t\tvar elem = this;\n\n\t\t\t\twhile ( elem.firstElementChild ) {\n\t\t\t\t\telem = elem.firstElementChild;\n\t\t\t\t}\n\n\t\t\t\treturn elem;\n\t\t\t} ).append( this );\n\t\t}\n\n\t\treturn this;\n\t},\n\n\twrapInner: function( html ) {\n\t\tif ( jQuery.isFunction( html ) ) {\n\t\t\treturn this.each( function( i ) {\n\t\t\t\tjQuery( this ).wrapInner( html.call( this, i ) );\n\t\t\t} );\n\t\t}\n\n\t\treturn this.each( function() {\n\t\t\tvar self = jQuery( this ),\n\t\t\t\tcontents = self.contents();\n\n\t\t\tif ( contents.length ) {\n\t\t\t\tcontents.wrapAll( html );\n\n\t\t\t} else {\n\t\t\t\tself.append( html );\n\t\t\t}\n\t\t} );\n\t},\n\n\twrap: function( html ) {\n\t\tvar isFunction = jQuery.isFunction( html );\n\n\t\treturn this.each( function( i ) {\n\t\t\tjQuery( this ).wrapAll( isFunction ? html.call( this, i ) : html );\n\t\t} );\n\t},\n\n\tunwrap: function( selector ) {\n\t\tthis.parent( selector ).not( \"body\" ).each( function() {\n\t\t\tjQuery( this ).replaceWith( this.childNodes );\n\t\t} );\n\t\treturn this;\n\t}\n} );\n\n\njQuery.expr.pseudos.hidden = function( elem ) {\n\treturn !jQuery.expr.pseudos.visible( elem );\n};\njQuery.expr.pseudos.visible = function( elem ) {\n\treturn !!( elem.offsetWidth || elem.offsetHeight || elem.getClientRects().length );\n};\n\n\n\n\njQuery.ajaxSettings.xhr = function() {\n\ttry {\n\t\treturn new window.XMLHttpRequest();\n\t} catch ( e ) {}\n};\n\nvar xhrSuccessStatus = {\n\n\t\t// File protocol always yields status code 0, assume 200\n\t\t0: 200,\n\n\t\t// Support: IE <=9 only\n\t\t// #1450: sometimes IE returns 1223 when it should be 204\n\t\t1223: 204\n\t},\n\txhrSupported = jQuery.ajaxSettings.xhr();\n\nsupport.cors = !!xhrSupported && ( \"withCredentials\" in xhrSupported );\nsupport.ajax = xhrSupported = !!xhrSupported;\n\njQuery.ajaxTransport( function( options ) {\n\tvar callback, errorCallback;\n\n\t// Cross domain only allowed if supported through XMLHttpRequest\n\tif ( support.cors || xhrSupported && !options.crossDomain ) {\n\t\treturn {\n\t\t\tsend: function( headers, complete ) {\n\t\t\t\tvar i,\n\t\t\t\t\txhr = options.xhr();\n\n\t\t\t\txhr.open(\n\t\t\t\t\toptions.type,\n\t\t\t\t\toptions.url,\n\t\t\t\t\toptions.async,\n\t\t\t\t\toptions.username,\n\t\t\t\t\toptions.password\n\t\t\t\t);\n\n\t\t\t\t// Apply custom fields if provided\n\t\t\t\tif ( options.xhrFields ) {\n\t\t\t\t\tfor ( i in options.xhrFields ) {\n\t\t\t\t\t\txhr[ i ] = options.xhrFields[ i ];\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// Override mime type if needed\n\t\t\t\tif ( options.mimeType && xhr.overrideMimeType ) {\n\t\t\t\t\txhr.overrideMimeType( options.mimeType );\n\t\t\t\t}\n\n\t\t\t\t// X-Requested-With header\n\t\t\t\t// For cross-domain requests, seeing as conditions for a preflight are\n\t\t\t\t// akin to a jigsaw puzzle, we simply never set it to be sure.\n\t\t\t\t// (it can always be set on a per-request basis or even using ajaxSetup)\n\t\t\t\t// For same-domain requests, won't change header if already provided.\n\t\t\t\tif ( !options.crossDomain && !headers[ \"X-Requested-With\" ] ) {\n\t\t\t\t\theaders[ \"X-Requested-With\" ] = \"XMLHttpRequest\";\n\t\t\t\t}\n\n\t\t\t\t// Set headers\n\t\t\t\tfor ( i in headers ) {\n\t\t\t\t\txhr.setRequestHeader( i, headers[ i ] );\n\t\t\t\t}\n\n\t\t\t\t// Callback\n\t\t\t\tcallback = function( type ) {\n\t\t\t\t\treturn function() {\n\t\t\t\t\t\tif ( callback ) {\n\t\t\t\t\t\t\tcallback = errorCallback = xhr.onload =\n\t\t\t\t\t\t\t\txhr.onerror = xhr.onabort = xhr.onreadystatechange = null;\n\n\t\t\t\t\t\t\tif ( type === \"abort\" ) {\n\t\t\t\t\t\t\t\txhr.abort();\n\t\t\t\t\t\t\t} else if ( type === \"error\" ) {\n\n\t\t\t\t\t\t\t\t// Support: IE <=9 only\n\t\t\t\t\t\t\t\t// On a manual native abort, IE9 throws\n\t\t\t\t\t\t\t\t// errors on any property access that is not readyState\n\t\t\t\t\t\t\t\tif ( typeof xhr.status !== \"number\" ) {\n\t\t\t\t\t\t\t\t\tcomplete( 0, \"error\" );\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tcomplete(\n\n\t\t\t\t\t\t\t\t\t\t// File: protocol always yields status 0; see #8605, #14207\n\t\t\t\t\t\t\t\t\t\txhr.status,\n\t\t\t\t\t\t\t\t\t\txhr.statusText\n\t\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tcomplete(\n\t\t\t\t\t\t\t\t\txhrSuccessStatus[ xhr.status ] || xhr.status,\n\t\t\t\t\t\t\t\t\txhr.statusText,\n\n\t\t\t\t\t\t\t\t\t// Support: IE <=9 only\n\t\t\t\t\t\t\t\t\t// IE9 has no XHR2 but throws on binary (trac-11426)\n\t\t\t\t\t\t\t\t\t// For XHR2 non-text, let the caller handle it (gh-2498)\n\t\t\t\t\t\t\t\t\t( xhr.responseType || \"text\" ) !== \"text\" ||\n\t\t\t\t\t\t\t\t\ttypeof xhr.responseText !== \"string\" ?\n\t\t\t\t\t\t\t\t\t\t{ binary: xhr.response } :\n\t\t\t\t\t\t\t\t\t\t{ text: xhr.responseText },\n\t\t\t\t\t\t\t\t\txhr.getAllResponseHeaders()\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t};\n\t\t\t\t};\n\n\t\t\t\t// Listen to events\n\t\t\t\txhr.onload = callback();\n\t\t\t\terrorCallback = xhr.onerror = callback( \"error\" );\n\n\t\t\t\t// Support: IE 9 only\n\t\t\t\t// Use onreadystatechange to replace onabort\n\t\t\t\t// to handle uncaught aborts\n\t\t\t\tif ( xhr.onabort !== undefined ) {\n\t\t\t\t\txhr.onabort = errorCallback;\n\t\t\t\t} else {\n\t\t\t\t\txhr.onreadystatechange = function() {\n\n\t\t\t\t\t\t// Check readyState before timeout as it changes\n\t\t\t\t\t\tif ( xhr.readyState === 4 ) {\n\n\t\t\t\t\t\t\t// Allow onerror to be called first,\n\t\t\t\t\t\t\t// but that will not handle a native abort\n\t\t\t\t\t\t\t// Also, save errorCallback to a variable\n\t\t\t\t\t\t\t// as xhr.onerror cannot be accessed\n\t\t\t\t\t\t\twindow.setTimeout( function() {\n\t\t\t\t\t\t\t\tif ( callback ) {\n\t\t\t\t\t\t\t\t\terrorCallback();\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} );\n\t\t\t\t\t\t}\n\t\t\t\t\t};\n\t\t\t\t}\n\n\t\t\t\t// Create the abort callback\n\t\t\t\tcallback = callback( \"abort\" );\n\n\t\t\t\ttry {\n\n\t\t\t\t\t// Do send the request (this may raise an exception)\n\t\t\t\t\txhr.send( options.hasContent && options.data || null );\n\t\t\t\t} catch ( e ) {\n\n\t\t\t\t\t// #14683: Only rethrow if this hasn't been notified as an error yet\n\t\t\t\t\tif ( callback ) {\n\t\t\t\t\t\tthrow e;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t},\n\n\t\t\tabort: function() {\n\t\t\t\tif ( callback ) {\n\t\t\t\t\tcallback();\n\t\t\t\t}\n\t\t\t}\n\t\t};\n\t}\n} );\n\n\n\n\n// Prevent auto-execution of scripts when no explicit dataType was provided (See gh-2432)\njQuery.ajaxPrefilter( function( s ) {\n\tif ( s.crossDomain ) {\n\t\ts.contents.script = false;\n\t}\n} );\n\n// Install script dataType\njQuery.ajaxSetup( {\n\taccepts: {\n\t\tscript: \"text/javascript, application/javascript, \" +\n\t\t\t\"application/ecmascript, application/x-ecmascript\"\n\t},\n\tcontents: {\n\t\tscript: /\\b(?:java|ecma)script\\b/\n\t},\n\tconverters: {\n\t\t\"text script\": function( text ) {\n\t\t\tjQuery.globalEval( text );\n\t\t\treturn text;\n\t\t}\n\t}\n} );\n\n// Handle cache's special case and crossDomain\njQuery.ajaxPrefilter( \"script\", function( s ) {\n\tif ( s.cache === undefined ) {\n\t\ts.cache = false;\n\t}\n\tif ( s.crossDomain ) {\n\t\ts.type = \"GET\";\n\t}\n} );\n\n// Bind script tag hack transport\njQuery.ajaxTransport( \"script\", function( s ) {\n\n\t// This transport only deals with cross domain requests\n\tif ( s.crossDomain ) {\n\t\tvar script, callback;\n\t\treturn {\n\t\t\tsend: function( _, complete ) {\n\t\t\t\tscript = jQuery( \"\n \n \n \n \n \n \n\n\n
\n
\n
\n \n
\n
\n
Error Encountered
\n
\n
\n
\n
\n

\n \n
\n
\n
\n\t\t\t

Network Analysis

\n\t\t\t
\n\t\t\t
\n
\n
\n
\n\n\n" }, { "path": "presets/SSD300.prototxt", "content": "name: \"VGG_VOC0712_SSD_300x300_deploy\"\ninput: \"data\"\ninput_shape {\n dim: 1\n dim: 3\n dim: 300\n dim: 300\n}\nlayer {\n name: \"conv1_1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu1_1\"\n type: \"ReLU\"\n bottom: \"conv1_1\"\n top: \"conv1_1\"\n}\nlayer {\n name: \"conv1_2\"\n type: \"Convolution\"\n bottom: \"conv1_1\"\n top: \"conv1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu1_2\"\n type: \"ReLU\"\n bottom: \"conv1_2\"\n top: \"conv1_2\"\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv1_2\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv2_1\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"conv2_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu2_1\"\n type: \"ReLU\"\n bottom: \"conv2_1\"\n top: \"conv2_1\"\n}\nlayer {\n name: \"conv2_2\"\n type: \"Convolution\"\n bottom: \"conv2_1\"\n top: \"conv2_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu2_2\"\n type: \"ReLU\"\n bottom: \"conv2_2\"\n top: \"conv2_2\"\n}\nlayer {\n name: \"pool2\"\n type: \"Pooling\"\n bottom: \"conv2_2\"\n top: \"pool2\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv3_1\"\n type: \"Convolution\"\n bottom: \"pool2\"\n top: \"conv3_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu3_1\"\n type: \"ReLU\"\n bottom: \"conv3_1\"\n top: \"conv3_1\"\n}\nlayer {\n name: \"conv3_2\"\n type: \"Convolution\"\n bottom: \"conv3_1\"\n top: \"conv3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu3_2\"\n type: \"ReLU\"\n bottom: \"conv3_2\"\n top: \"conv3_2\"\n}\nlayer {\n name: \"conv3_3\"\n type: \"Convolution\"\n bottom: \"conv3_2\"\n top: \"conv3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu3_3\"\n type: \"ReLU\"\n bottom: \"conv3_3\"\n top: \"conv3_3\"\n}\nlayer {\n name: \"pool3\"\n type: \"Pooling\"\n bottom: \"conv3_3\"\n top: \"pool3\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv4_1\"\n type: \"Convolution\"\n bottom: \"pool3\"\n top: \"conv4_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu4_1\"\n type: \"ReLU\"\n bottom: \"conv4_1\"\n top: \"conv4_1\"\n}\nlayer {\n name: \"conv4_2\"\n type: \"Convolution\"\n bottom: \"conv4_1\"\n top: \"conv4_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu4_2\"\n type: \"ReLU\"\n bottom: \"conv4_2\"\n top: \"conv4_2\"\n}\nlayer {\n name: \"conv4_3\"\n type: \"Convolution\"\n bottom: \"conv4_2\"\n top: \"conv4_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu4_3\"\n type: \"ReLU\"\n bottom: \"conv4_3\"\n top: \"conv4_3\"\n}\nlayer {\n name: \"pool4\"\n type: \"Pooling\"\n bottom: \"conv4_3\"\n top: \"pool4\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv5_1\"\n type: \"Convolution\"\n bottom: \"pool4\"\n top: \"conv5_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n dilation: 1\n }\n}\nlayer {\n name: \"relu5_1\"\n type: \"ReLU\"\n bottom: \"conv5_1\"\n top: \"conv5_1\"\n}\nlayer {\n name: \"conv5_2\"\n type: \"Convolution\"\n bottom: \"conv5_1\"\n top: \"conv5_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n dilation: 1\n }\n}\nlayer {\n name: \"relu5_2\"\n type: \"ReLU\"\n bottom: \"conv5_2\"\n top: \"conv5_2\"\n}\nlayer {\n name: \"conv5_3\"\n type: \"Convolution\"\n bottom: \"conv5_2\"\n top: \"conv5_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n dilation: 1\n }\n}\nlayer {\n name: \"relu5_3\"\n type: \"ReLU\"\n bottom: \"conv5_3\"\n top: \"conv5_3\"\n}\nlayer {\n name: \"pool5\"\n type: \"Pooling\"\n bottom: \"conv5_3\"\n top: \"pool5\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"fc6\"\n type: \"Convolution\"\n bottom: \"pool5\"\n top: \"fc6\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1024\n pad: 6\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n dilation: 6\n }\n}\nlayer {\n name: \"relu6\"\n type: \"ReLU\"\n bottom: \"fc6\"\n top: \"fc6\"\n}\nlayer {\n name: \"fc7\"\n type: \"Convolution\"\n bottom: \"fc6\"\n top: \"fc7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1024\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu7\"\n type: \"ReLU\"\n bottom: \"fc7\"\n top: \"fc7\"\n}\nlayer {\n name: \"conv6_1\"\n type: \"Convolution\"\n bottom: \"fc7\"\n top: \"conv6_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv6_1_relu\"\n type: \"ReLU\"\n bottom: \"conv6_1\"\n top: \"conv6_1\"\n}\nlayer {\n name: \"conv6_2\"\n type: \"Convolution\"\n bottom: \"conv6_1\"\n top: \"conv6_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv6_2_relu\"\n type: \"ReLU\"\n bottom: \"conv6_2\"\n top: \"conv6_2\"\n}\nlayer {\n name: \"conv7_1\"\n type: \"Convolution\"\n bottom: \"conv6_2\"\n top: \"conv7_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv7_1_relu\"\n type: \"ReLU\"\n bottom: \"conv7_1\"\n top: \"conv7_1\"\n}\nlayer {\n name: \"conv7_2\"\n type: \"Convolution\"\n bottom: \"conv7_1\"\n top: \"conv7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv7_2_relu\"\n type: \"ReLU\"\n bottom: \"conv7_2\"\n top: \"conv7_2\"\n}\nlayer {\n name: \"conv8_1\"\n type: \"Convolution\"\n bottom: \"conv7_2\"\n top: \"conv8_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv8_1_relu\"\n type: \"ReLU\"\n bottom: \"conv8_1\"\n top: \"conv8_1\"\n}\nlayer {\n name: \"conv8_2\"\n type: \"Convolution\"\n bottom: \"conv8_1\"\n top: \"conv8_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv8_2_relu\"\n type: \"ReLU\"\n bottom: \"conv8_2\"\n top: \"conv8_2\"\n}\nlayer {\n name: \"conv9_1\"\n type: \"Convolution\"\n bottom: \"conv8_2\"\n top: \"conv9_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv9_1_relu\"\n type: \"ReLU\"\n bottom: \"conv9_1\"\n top: \"conv9_1\"\n}\nlayer {\n name: \"conv9_2\"\n type: \"Convolution\"\n bottom: \"conv9_1\"\n top: \"conv9_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv9_2_relu\"\n type: \"ReLU\"\n bottom: \"conv9_2\"\n top: \"conv9_2\"\n}\nlayer {\n name: \"conv4_3_norm\"\n type: \"Normalize\"\n bottom: \"conv4_3\"\n top: \"conv4_3_norm\"\n norm_param {\n across_spatial: false\n scale_filler {\n type: \"constant\"\n value: 20\n }\n channel_shared: false\n }\n}\nlayer {\n name: \"conv4_3_norm_mbox_loc\"\n type: \"Convolution\"\n bottom: \"conv4_3_norm\"\n top: \"conv4_3_norm_mbox_loc\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 16\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv4_3_norm_mbox_loc_perm\"\n type: \"Permute\"\n bottom: \"conv4_3_norm_mbox_loc\"\n top: \"conv4_3_norm_mbox_loc_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv4_3_norm_mbox_loc_flat\"\n type: \"Flatten\"\n bottom: \"conv4_3_norm_mbox_loc_perm\"\n top: \"conv4_3_norm_mbox_loc_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv4_3_norm_mbox_conf\"\n type: \"Convolution\"\n bottom: \"conv4_3_norm\"\n top: \"conv4_3_norm_mbox_conf\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 84\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv4_3_norm_mbox_conf_perm\"\n type: \"Permute\"\n bottom: \"conv4_3_norm_mbox_conf\"\n top: \"conv4_3_norm_mbox_conf_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv4_3_norm_mbox_conf_flat\"\n type: \"Flatten\"\n bottom: \"conv4_3_norm_mbox_conf_perm\"\n top: \"conv4_3_norm_mbox_conf_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv4_3_norm_mbox_priorbox\"\n type: \"PriorBox\"\n bottom: \"conv4_3_norm\"\n bottom: \"data\"\n top: \"conv4_3_norm_mbox_priorbox\"\n prior_box_param {\n min_size: 30.0\n max_size: 60.0\n aspect_ratio: 2\n flip: true\n clip: false\n variance: 0.1\n variance: 0.1\n variance: 0.2\n variance: 0.2\n step: 8\n offset: 0.5\n }\n}\nlayer {\n name: \"fc7_mbox_loc\"\n type: \"Convolution\"\n bottom: \"fc7\"\n top: \"fc7_mbox_loc\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 24\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"fc7_mbox_loc_perm\"\n type: \"Permute\"\n bottom: \"fc7_mbox_loc\"\n top: \"fc7_mbox_loc_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"fc7_mbox_loc_flat\"\n type: \"Flatten\"\n bottom: \"fc7_mbox_loc_perm\"\n top: \"fc7_mbox_loc_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"fc7_mbox_conf\"\n type: \"Convolution\"\n bottom: \"fc7\"\n top: \"fc7_mbox_conf\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 126\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"fc7_mbox_conf_perm\"\n type: \"Permute\"\n bottom: \"fc7_mbox_conf\"\n top: \"fc7_mbox_conf_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"fc7_mbox_conf_flat\"\n type: \"Flatten\"\n bottom: \"fc7_mbox_conf_perm\"\n top: \"fc7_mbox_conf_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"fc7_mbox_priorbox\"\n type: \"PriorBox\"\n bottom: \"fc7\"\n bottom: \"data\"\n top: \"fc7_mbox_priorbox\"\n prior_box_param {\n min_size: 60.0\n max_size: 111.0\n aspect_ratio: 2\n aspect_ratio: 3\n flip: true\n clip: false\n variance: 0.1\n variance: 0.1\n variance: 0.2\n variance: 0.2\n step: 16\n offset: 0.5\n }\n}\nlayer {\n name: \"conv6_2_mbox_loc\"\n type: \"Convolution\"\n bottom: \"conv6_2\"\n top: \"conv6_2_mbox_loc\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 24\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv6_2_mbox_loc_perm\"\n type: \"Permute\"\n bottom: \"conv6_2_mbox_loc\"\n top: \"conv6_2_mbox_loc_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv6_2_mbox_loc_flat\"\n type: \"Flatten\"\n bottom: \"conv6_2_mbox_loc_perm\"\n top: \"conv6_2_mbox_loc_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv6_2_mbox_conf\"\n type: \"Convolution\"\n bottom: \"conv6_2\"\n top: \"conv6_2_mbox_conf\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 126\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv6_2_mbox_conf_perm\"\n type: \"Permute\"\n bottom: \"conv6_2_mbox_conf\"\n top: \"conv6_2_mbox_conf_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv6_2_mbox_conf_flat\"\n type: \"Flatten\"\n bottom: \"conv6_2_mbox_conf_perm\"\n top: \"conv6_2_mbox_conf_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv6_2_mbox_priorbox\"\n type: \"PriorBox\"\n bottom: \"conv6_2\"\n bottom: \"data\"\n top: \"conv6_2_mbox_priorbox\"\n prior_box_param {\n min_size: 111.0\n max_size: 162.0\n aspect_ratio: 2\n aspect_ratio: 3\n flip: true\n clip: false\n variance: 0.1\n variance: 0.1\n variance: 0.2\n variance: 0.2\n step: 32\n offset: 0.5\n }\n}\nlayer {\n name: \"conv7_2_mbox_loc\"\n type: \"Convolution\"\n bottom: \"conv7_2\"\n top: \"conv7_2_mbox_loc\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 24\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv7_2_mbox_loc_perm\"\n type: \"Permute\"\n bottom: \"conv7_2_mbox_loc\"\n top: \"conv7_2_mbox_loc_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv7_2_mbox_loc_flat\"\n type: \"Flatten\"\n bottom: \"conv7_2_mbox_loc_perm\"\n top: \"conv7_2_mbox_loc_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv7_2_mbox_conf\"\n type: \"Convolution\"\n bottom: \"conv7_2\"\n top: \"conv7_2_mbox_conf\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 126\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv7_2_mbox_conf_perm\"\n type: \"Permute\"\n bottom: \"conv7_2_mbox_conf\"\n top: \"conv7_2_mbox_conf_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv7_2_mbox_conf_flat\"\n type: \"Flatten\"\n bottom: \"conv7_2_mbox_conf_perm\"\n top: \"conv7_2_mbox_conf_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv7_2_mbox_priorbox\"\n type: \"PriorBox\"\n bottom: \"conv7_2\"\n bottom: \"data\"\n top: \"conv7_2_mbox_priorbox\"\n prior_box_param {\n min_size: 162.0\n max_size: 213.0\n aspect_ratio: 2\n aspect_ratio: 3\n flip: true\n clip: false\n variance: 0.1\n variance: 0.1\n variance: 0.2\n variance: 0.2\n step: 64\n offset: 0.5\n }\n}\nlayer {\n name: \"conv8_2_mbox_loc\"\n type: \"Convolution\"\n bottom: \"conv8_2\"\n top: \"conv8_2_mbox_loc\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 16\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv8_2_mbox_loc_perm\"\n type: \"Permute\"\n bottom: \"conv8_2_mbox_loc\"\n top: \"conv8_2_mbox_loc_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv8_2_mbox_loc_flat\"\n type: \"Flatten\"\n bottom: \"conv8_2_mbox_loc_perm\"\n top: \"conv8_2_mbox_loc_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv8_2_mbox_conf\"\n type: \"Convolution\"\n bottom: \"conv8_2\"\n top: \"conv8_2_mbox_conf\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 84\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv8_2_mbox_conf_perm\"\n type: \"Permute\"\n bottom: \"conv8_2_mbox_conf\"\n top: \"conv8_2_mbox_conf_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv8_2_mbox_conf_flat\"\n type: \"Flatten\"\n bottom: \"conv8_2_mbox_conf_perm\"\n top: \"conv8_2_mbox_conf_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv8_2_mbox_priorbox\"\n type: \"PriorBox\"\n bottom: \"conv8_2\"\n bottom: \"data\"\n top: \"conv8_2_mbox_priorbox\"\n prior_box_param {\n min_size: 213.0\n max_size: 264.0\n aspect_ratio: 2\n flip: true\n clip: false\n variance: 0.1\n variance: 0.1\n variance: 0.2\n variance: 0.2\n step: 100\n offset: 0.5\n }\n}\nlayer {\n name: \"conv9_2_mbox_loc\"\n type: \"Convolution\"\n bottom: \"conv9_2\"\n top: \"conv9_2_mbox_loc\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 16\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv9_2_mbox_loc_perm\"\n type: \"Permute\"\n bottom: \"conv9_2_mbox_loc\"\n top: \"conv9_2_mbox_loc_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv9_2_mbox_loc_flat\"\n type: \"Flatten\"\n bottom: \"conv9_2_mbox_loc_perm\"\n top: \"conv9_2_mbox_loc_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv9_2_mbox_conf\"\n type: \"Convolution\"\n bottom: \"conv9_2\"\n top: \"conv9_2_mbox_conf\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 84\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"conv9_2_mbox_conf_perm\"\n type: \"Permute\"\n bottom: \"conv9_2_mbox_conf\"\n top: \"conv9_2_mbox_conf_perm\"\n permute_param {\n order: 0\n order: 2\n order: 3\n order: 1\n }\n}\nlayer {\n name: \"conv9_2_mbox_conf_flat\"\n type: \"Flatten\"\n bottom: \"conv9_2_mbox_conf_perm\"\n top: \"conv9_2_mbox_conf_flat\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"conv9_2_mbox_priorbox\"\n type: \"PriorBox\"\n bottom: \"conv9_2\"\n bottom: \"data\"\n top: \"conv9_2_mbox_priorbox\"\n prior_box_param {\n min_size: 264.0\n max_size: 315.0\n aspect_ratio: 2\n flip: true\n clip: false\n variance: 0.1\n variance: 0.1\n variance: 0.2\n variance: 0.2\n step: 300\n offset: 0.5\n }\n}\nlayer {\n name: \"mbox_loc\"\n type: \"Concat\"\n bottom: \"conv4_3_norm_mbox_loc_flat\"\n bottom: \"fc7_mbox_loc_flat\"\n bottom: \"conv6_2_mbox_loc_flat\"\n bottom: \"conv7_2_mbox_loc_flat\"\n bottom: \"conv8_2_mbox_loc_flat\"\n bottom: \"conv9_2_mbox_loc_flat\"\n top: \"mbox_loc\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mbox_conf\"\n type: \"Concat\"\n bottom: \"conv4_3_norm_mbox_conf_flat\"\n bottom: \"fc7_mbox_conf_flat\"\n bottom: \"conv6_2_mbox_conf_flat\"\n bottom: \"conv7_2_mbox_conf_flat\"\n bottom: \"conv8_2_mbox_conf_flat\"\n bottom: \"conv9_2_mbox_conf_flat\"\n top: \"mbox_conf\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mbox_priorbox\"\n type: \"Concat\"\n bottom: \"conv4_3_norm_mbox_priorbox\"\n bottom: \"fc7_mbox_priorbox\"\n bottom: \"conv6_2_mbox_priorbox\"\n bottom: \"conv7_2_mbox_priorbox\"\n bottom: \"conv8_2_mbox_priorbox\"\n bottom: \"conv9_2_mbox_priorbox\"\n top: \"mbox_priorbox\"\n concat_param {\n axis: 2\n }\n}\nlayer {\n name: \"mbox_conf_reshape\"\n type: \"Reshape\"\n bottom: \"mbox_conf\"\n top: \"mbox_conf_reshape\"\n reshape_param {\n shape {\n dim: 0\n dim: -1\n dim: 21\n }\n }\n}\nlayer {\n name: \"mbox_conf_softmax\"\n type: \"Softmax\"\n bottom: \"mbox_conf_reshape\"\n top: \"mbox_conf_softmax\"\n softmax_param {\n axis: 2\n }\n}\nlayer {\n name: \"mbox_conf_flatten\"\n type: \"Flatten\"\n bottom: \"mbox_conf_softmax\"\n top: \"mbox_conf_flatten\"\n flatten_param {\n axis: 1\n }\n}\nlayer {\n name: \"detection_out\"\n type: \"DetectionOutput\"\n bottom: \"mbox_loc\"\n bottom: \"mbox_conf_flatten\"\n bottom: \"mbox_priorbox\"\n top: \"detection_out\"\n include {\n phase: TEST\n }\n detection_output_param {\n num_classes: 21\n share_location: true\n background_label_id: 0\n nms_param {\n nms_threshold: 0.45\n top_k: 400\n }\n save_output_param {\n output_directory: \"/home-2/wliu/data/VOCdevkit/results/VOC2007/SSD_300x300/Main\"\n output_name_prefix: \"comp4_det_test_\"\n output_format: \"VOC\"\n label_map_file: \"data/VOC0712/labelmap_voc.prototxt\"\n name_size_file: \"data/VOC0712/test_name_size.txt\"\n num_test_image: 4952\n }\n code_type: CENTER_SIZE\n keep_top_k: 200\n confidence_threshold: 0.01\n }\n}\n\n" }, { "path": "presets/YOLO.prototxt", "content": "name: \"YOLONet\"\ninput: \"data\"\ninput_shape {\n dim: 1\n dim: 3\n dim: 448\n dim: 448\n}\n\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n convolution_param {\n num_output: 64\n kernel_size: 7\n pad: 3\n stride: 2\n }\n}\nlayer {\n name: \"relu1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv1\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\n\nlayer{\n name: \"conv2\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"conv2\"\n convolution_param {\n num_output: 192\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu2\"\n type: \"ReLU\"\n bottom: \"conv2\"\n top: \"conv2\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\nlayer {\n name: \"pool2\"\n type: \"Pooling\"\n bottom: \"conv2\"\n top: \"pool2\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\n\nlayer{\n name: \"conv3\"\n type: \"Convolution\"\n bottom: \"pool2\"\n top: \"conv3\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu3\"\n type: \"ReLU\"\n bottom: \"conv3\"\n top: \"conv3\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"conv4\"\n type: \"Convolution\"\n bottom: \"conv3\"\n top: \"conv4\"\n convolution_param {\n num_output: 256\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu4\"\n type: \"ReLU\"\n bottom: \"conv4\"\n top: \"conv4\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv5\"\n type: \"Convolution\"\n bottom: \"conv4\"\n top: \"conv5\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu5\"\n type: \"ReLU\"\n bottom: \"conv5\"\n top: \"conv5\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv6\"\n type: \"Convolution\"\n bottom: \"conv5\"\n top: \"conv6\"\n convolution_param {\n num_output: 512\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu6\"\n type: \"ReLU\"\n bottom: \"conv6\"\n top: \"conv6\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\nlayer {\n name: \"pool6\"\n type: \"Pooling\"\n bottom: \"conv6\"\n top: \"pool6\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\n\nlayer{\n name: \"conv7\"\n type: \"Convolution\"\n bottom: \"pool6\"\n top: \"conv7\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu7\"\n type: \"ReLU\"\n bottom: \"conv7\"\n top: \"conv7\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv8\"\n type: \"Convolution\"\n bottom: \"conv7\"\n top: \"conv8\"\n convolution_param {\n num_output: 512\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu8\"\n type: \"ReLU\"\n bottom: \"conv8\"\n top: \"conv8\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv9\"\n type: \"Convolution\"\n bottom: \"conv8\"\n top: \"conv9\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu9\"\n type: \"ReLU\"\n bottom: \"conv9\"\n top: \"conv9\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv10\"\n type: \"Convolution\"\n bottom: \"conv9\"\n top: \"conv10\"\n convolution_param {\n num_output: 512\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu10\"\n type: \"ReLU\"\n bottom: \"conv10\"\n top: \"conv10\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv11\"\n type: \"Convolution\"\n bottom: \"conv10\"\n top: \"conv11\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu11\"\n type: \"ReLU\"\n bottom: \"conv11\"\n top: \"conv11\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"conv12\"\n type: \"Convolution\"\n bottom: \"conv11\"\n top: \"conv12\"\n convolution_param {\n num_output: 512\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu12\"\n type: \"ReLU\"\n bottom: \"conv12\"\n top: \"conv12\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"conv13\"\n type: \"Convolution\"\n bottom: \"conv12\"\n top: \"conv13\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu13\"\n type: \"ReLU\"\n bottom: \"conv13\"\n top: \"conv13\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv14\"\n type: \"Convolution\"\n bottom: \"conv13\"\n top: \"conv14\"\n convolution_param {\n num_output: 512\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu14\"\n type: \"ReLU\"\n bottom: \"conv14\"\n top: \"conv14\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer{\n name: \"conv15\"\n type: \"Convolution\"\n bottom: \"conv14\"\n top: \"conv15\"\n convolution_param {\n num_output: 512\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu15\"\n type: \"ReLU\"\n bottom: \"conv15\"\n top: \"conv15\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"conv16\"\n type: \"Convolution\"\n bottom: \"conv15\"\n top: \"conv16\"\n convolution_param {\n num_output: 1024\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu16\"\n type: \"ReLU\"\n bottom: \"conv16\"\n top: \"conv16\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\nlayer {\n name: \"pool16\"\n type: \"Pooling\"\n bottom: \"conv16\"\n top: \"pool16\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\n\n\nlayer{\n name: \"conv17\"\n type: \"Convolution\"\n bottom: \"pool16\"\n top: \"conv17\"\n convolution_param {\n num_output: 512\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu17\"\n type: \"ReLU\"\n bottom: \"conv17\"\n top: \"conv17\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"conv18\"\n type: \"Convolution\"\n bottom: \"conv17\"\n top: \"conv18\"\n convolution_param {\n num_output: 1024\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu18\"\n type: \"ReLU\"\n bottom: \"conv18\"\n top: \"conv18\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\n\nlayer{\n name: \"conv19\"\n type: \"Convolution\"\n bottom: \"conv18\"\n top: \"conv19\"\n convolution_param {\n num_output: 512\n kernel_size: 1\n pad: 0\n stride: 1\n }\n}\nlayer {\n name: \"relu19\"\n type: \"ReLU\"\n bottom: \"conv19\"\n top: \"conv19\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\n\nlayer{\n name: \"conv20\"\n type: \"Convolution\"\n bottom: \"conv19\"\n top: \"conv20\"\n convolution_param {\n num_output: 1024\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu20\"\n type: \"ReLU\"\n bottom: \"conv20\"\n top: \"conv20\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\n\nlayer{\n name: \"conv21\"\n type: \"Convolution\"\n bottom: \"conv20\"\n top: \"conv21\"\n convolution_param {\n num_output: 1024\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu21\"\n type: \"ReLU\"\n bottom: \"conv21\"\n top: \"conv21\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"conv22\"\n type: \"Convolution\"\n bottom: \"conv21\"\n top: \"conv22\"\n convolution_param {\n num_output: 1024\n kernel_size: 3\n pad: 1\n stride: 2\n }\n}\nlayer {\n name: \"relu22\"\n type: \"ReLU\"\n bottom: \"conv22\"\n top: \"conv22\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\n\nlayer{\n name: \"conv23\"\n type: \"Convolution\"\n bottom: \"conv22\"\n top: \"conv23\"\n convolution_param {\n num_output: 1024\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu23\"\n type: \"ReLU\"\n bottom: \"conv23\"\n top: \"conv23\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"conv24\"\n type: \"Convolution\"\n bottom: \"conv23\"\n top: \"conv24\"\n convolution_param {\n num_output: 1024\n kernel_size: 3\n pad: 1\n stride: 1\n }\n}\nlayer {\n name: \"relu24\"\n type: \"ReLU\"\n bottom: \"conv24\"\n top: \"conv24\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\n\n\nlayer{\n name: \"fc25\"\n type: \"InnerProduct\"\n bottom: \"conv24\"\n top: \"fc25\"\n inner_product_param {\n num_output: 4096\n }\n}\nlayer {\n name: \"relu25\"\n type: \"ReLU\"\n bottom: \"fc25\"\n top: \"fc25\"\n relu_param{\n negative_slope: 0.1\n }\t\t\n}\n\n\nlayer{\n name: \"fc26\"\n type: \"InnerProduct\"\n bottom: \"fc25\"\n top: \"result\"\n inner_product_param {\n num_output: 1470\n }\n}\n\n" }, { "path": "presets/alexnet.prototxt", "content": "name: \"AlexNet\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n top: \"label\"\n include {\n phase: TRAIN\n }\n transform_param {\n mirror: true\n crop_size: 227\n mean_file: \"data/ilsvrc12/imagenet_mean.binaryproto\"\n }\n data_param {\n source: \"examples/imagenet/ilsvrc12_train_lmdb\"\n batch_size: 256\n backend: LMDB\n }\n}\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n top: \"label\"\n include {\n phase: TEST\n }\n transform_param {\n mirror: false\n crop_size: 227\n mean_file: \"data/ilsvrc12/imagenet_mean.binaryproto\"\n }\n data_param {\n source: \"examples/imagenet/ilsvrc12_val_lmdb\"\n batch_size: 50\n backend: LMDB\n }\n}\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n kernel_size: 11\n stride: 4\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n}\nlayer {\n name: \"norm1\"\n type: \"LRN\"\n bottom: \"conv1\"\n top: \"norm1\"\n lrn_param {\n local_size: 5\n alpha: 0.0001\n beta: 0.75\n }\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"norm1\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"conv2\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"conv2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 2\n kernel_size: 5\n group: 2\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.1\n }\n }\n}\nlayer {\n name: \"relu2\"\n type: \"ReLU\"\n bottom: \"conv2\"\n top: \"conv2\"\n}\nlayer {\n name: \"norm2\"\n type: \"LRN\"\n bottom: \"conv2\"\n top: \"norm2\"\n lrn_param {\n local_size: 5\n alpha: 0.0001\n beta: 0.75\n }\n}\nlayer {\n name: \"pool2\"\n type: \"Pooling\"\n bottom: \"norm2\"\n top: \"pool2\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"conv3\"\n type: \"Convolution\"\n bottom: \"pool2\"\n top: \"conv3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu3\"\n type: \"ReLU\"\n bottom: \"conv3\"\n top: \"conv3\"\n}\nlayer {\n name: \"conv4\"\n type: \"Convolution\"\n bottom: \"conv3\"\n top: \"conv4\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 1\n kernel_size: 3\n group: 2\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.1\n }\n }\n}\nlayer {\n name: \"relu4\"\n type: \"ReLU\"\n bottom: \"conv4\"\n top: \"conv4\"\n}\nlayer {\n name: \"conv5\"\n type: \"Convolution\"\n bottom: \"conv4\"\n top: \"conv5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n group: 2\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.1\n }\n }\n}\nlayer {\n name: \"relu5\"\n type: \"ReLU\"\n bottom: \"conv5\"\n top: \"conv5\"\n}\nlayer {\n name: \"pool5\"\n type: \"Pooling\"\n bottom: \"conv5\"\n top: \"pool5\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fc6\"\n type: \"InnerProduct\"\n bottom: \"pool5\"\n top: \"fc6\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 4096\n weight_filler {\n type: \"gaussian\"\n std: 0.005\n }\n bias_filler {\n type: \"constant\"\n value: 0.1\n }\n }\n}\nlayer {\n name: \"relu6\"\n type: \"ReLU\"\n bottom: \"fc6\"\n top: \"fc6\"\n}\nlayer {\n name: \"drop6\"\n type: \"Dropout\"\n bottom: \"fc6\"\n top: \"fc6\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"fc7\"\n type: \"InnerProduct\"\n bottom: \"fc6\"\n top: \"fc7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 4096\n weight_filler {\n type: \"gaussian\"\n std: 0.005\n }\n bias_filler {\n type: \"constant\"\n value: 0.1\n }\n }\n}\nlayer {\n name: \"relu7\"\n type: \"ReLU\"\n bottom: \"fc7\"\n top: \"fc7\"\n}\nlayer {\n name: \"drop7\"\n type: \"Dropout\"\n bottom: \"fc7\"\n top: \"fc7\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"fc8\"\n type: \"InnerProduct\"\n bottom: \"fc7\"\n top: \"fc8\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 1000\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"accuracy\"\n type: \"Accuracy\"\n bottom: \"fc8\"\n bottom: \"label\"\n top: \"accuracy\"\n include {\n phase: TEST\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"fc8\"\n bottom: \"label\"\n top: \"loss\"\n}\n" }, { "path": "presets/caffenet.prototxt", "content": "name: \"CaffeNet\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n top: \"label\"\n include {\n phase: TRAIN\n }\n transform_param {\n mirror: true\n crop_size: 227\n mean_file: \"data/ilsvrc12/imagenet_mean.binaryproto\"\n }\n# mean pixel / channel-wise mean instead of mean image\n# transform_param {\n# crop_size: 227\n# mean_value: 104\n# mean_value: 117\n# mean_value: 123\n# mirror: true\n# }\n data_param {\n source: \"examples/imagenet/ilsvrc12_train_lmdb\"\n batch_size: 256\n backend: LMDB\n }\n}\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n top: \"label\"\n include {\n phase: TEST\n }\n transform_param {\n mirror: false\n crop_size: 227\n mean_file: \"data/ilsvrc12/imagenet_mean.binaryproto\"\n }\n# mean pixel / channel-wise mean instead of mean image\n# transform_param {\n# crop_size: 227\n# mean_value: 104\n# mean_value: 117\n# mean_value: 123\n# mirror: true\n# }\n data_param {\n source: \"examples/imagenet/ilsvrc12_val_lmdb\"\n batch_size: 50\n backend: LMDB\n }\n}\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n kernel_size: 11\n stride: 4\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv1\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"norm1\"\n type: \"LRN\"\n bottom: \"pool1\"\n top: \"norm1\"\n lrn_param {\n local_size: 5\n alpha: 0.0001\n beta: 0.75\n }\n}\nlayer {\n name: \"conv2\"\n type: \"Convolution\"\n bottom: \"norm1\"\n top: \"conv2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 2\n kernel_size: 5\n group: 2\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 1\n }\n }\n}\nlayer {\n name: \"relu2\"\n type: \"ReLU\"\n bottom: \"conv2\"\n top: \"conv2\"\n}\nlayer {\n name: \"pool2\"\n type: \"Pooling\"\n bottom: \"conv2\"\n top: \"pool2\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"norm2\"\n type: \"LRN\"\n bottom: \"pool2\"\n top: \"norm2\"\n lrn_param {\n local_size: 5\n alpha: 0.0001\n beta: 0.75\n }\n}\nlayer {\n name: \"conv3\"\n type: \"Convolution\"\n bottom: \"norm2\"\n top: \"conv3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"relu3\"\n type: \"ReLU\"\n bottom: \"conv3\"\n top: \"conv3\"\n}\nlayer {\n name: \"conv4\"\n type: \"Convolution\"\n bottom: \"conv3\"\n top: \"conv4\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 1\n kernel_size: 3\n group: 2\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 1\n }\n }\n}\nlayer {\n name: \"relu4\"\n type: \"ReLU\"\n bottom: \"conv4\"\n top: \"conv4\"\n}\nlayer {\n name: \"conv5\"\n type: \"Convolution\"\n bottom: \"conv4\"\n top: \"conv5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n group: 2\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 1\n }\n }\n}\nlayer {\n name: \"relu5\"\n type: \"ReLU\"\n bottom: \"conv5\"\n top: \"conv5\"\n}\nlayer {\n name: \"pool5\"\n type: \"Pooling\"\n bottom: \"conv5\"\n top: \"pool5\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fc6\"\n type: \"InnerProduct\"\n bottom: \"pool5\"\n top: \"fc6\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 4096\n weight_filler {\n type: \"gaussian\"\n std: 0.005\n }\n bias_filler {\n type: \"constant\"\n value: 1\n }\n }\n}\nlayer {\n name: \"relu6\"\n type: \"ReLU\"\n bottom: \"fc6\"\n top: \"fc6\"\n}\nlayer {\n name: \"drop6\"\n type: \"Dropout\"\n bottom: \"fc6\"\n top: \"fc6\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"fc7\"\n type: \"InnerProduct\"\n bottom: \"fc6\"\n top: \"fc7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 4096\n weight_filler {\n type: \"gaussian\"\n std: 0.005\n }\n bias_filler {\n type: \"constant\"\n value: 1\n }\n }\n}\nlayer {\n name: \"relu7\"\n type: \"ReLU\"\n bottom: \"fc7\"\n top: \"fc7\"\n}\nlayer {\n name: \"drop7\"\n type: \"Dropout\"\n bottom: \"fc7\"\n top: \"fc7\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"fc8\"\n type: \"InnerProduct\"\n bottom: \"fc7\"\n top: \"fc8\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 1000\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"accuracy\"\n type: \"Accuracy\"\n bottom: \"fc8\"\n bottom: \"label\"\n top: \"accuracy\"\n include {\n phase: TEST\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"fc8\"\n bottom: \"label\"\n top: \"loss\"\n}\n" }, { "path": "presets/fasterRCNN_AlexNet.prototxt", "content": "name: \"Faster R-CNN (AlexNet)\"\n\ninput: \"data\"\ninput_shape {\n dim: 1\n dim: 3\n dim: 400\n dim: 500\n}\n\ninput: \"im_info\"\ninput_shape {\n dim: 1\n dim: 3\n}\n\n#========= conv1-conv5 ============\n\nlayer {\n\tname: \"conv1\"\n\ttype: \"Convolution\"\n\tbottom: \"data\"\n\ttop: \"conv1\"\n\tconvolution_param {\n\t\tnum_output: 96\n\t\tkernel_size: 7\n\t\tpad: 3\n\t\tstride: 2\n\t}\n}\nlayer {\n\tname: \"relu1\"\n\ttype: \"ReLU\"\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n}\nlayer {\n\tname: \"norm1\"\n\ttype: \"LRN\"\n\tbottom: \"conv1\"\n\ttop: \"norm1\"\n\tlrn_param {\n\t\tlocal_size: 3\n\t\talpha: 0.00005\n\t\tbeta: 0.75\n\t\tnorm_region: WITHIN_CHANNEL\n engine: CAFFE\n\t}\n}\nlayer {\n\tname: \"pool1\"\n\ttype: \"Pooling\"\n\tbottom: \"norm1\"\n\ttop: \"pool1\"\n\tpooling_param {\n\t\tkernel_size: 3\n\t\tstride: 2\n\t\tpad: 1\n\t\tpool: MAX\n\t}\n}\nlayer {\n\tname: \"conv2\"\n\ttype: \"Convolution\"\n\tbottom: \"pool1\"\n\ttop: \"conv2\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 5\n\t\tpad: 2\n\t\tstride: 2\n\t}\n}\nlayer {\n\tname: \"relu2\"\n\ttype: \"ReLU\"\n\tbottom: \"conv2\"\n\ttop: \"conv2\"\n}\nlayer {\n\tname: \"norm2\"\n\ttype: \"LRN\"\n\tbottom: \"conv2\"\n\ttop: \"norm2\"\n\tlrn_param {\n\t\tlocal_size: 3\n\t\talpha: 0.00005\n\t\tbeta: 0.75\n\t\tnorm_region: WITHIN_CHANNEL\n engine: CAFFE\n\t}\n}\nlayer {\n\tname: \"pool2\"\n\ttype: \"Pooling\"\n\tbottom: \"norm2\"\n\ttop: \"pool2\"\n\tpooling_param {\n\t\tkernel_size: 3\n\t\tstride: 2\n\t\tpad: 1\n\t\tpool: MAX\n\t}\n}\nlayer {\n\tname: \"conv3\"\n\ttype: \"Convolution\"\n\tbottom: \"pool2\"\n\ttop: \"conv3\"\n\tconvolution_param {\n\t\tnum_output: 384\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t}\n}\nlayer {\n\tname: \"relu3\"\n\ttype: \"ReLU\"\n\tbottom: \"conv3\"\n\ttop: \"conv3\"\n}\nlayer {\n\tname: \"conv4\"\n\ttype: \"Convolution\"\n\tbottom: \"conv3\"\n\ttop: \"conv4\"\n\tconvolution_param {\n\t\tnum_output: 384\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t}\n}\nlayer {\n\tname: \"relu4\"\n\ttype: \"ReLU\"\n\tbottom: \"conv4\"\n\ttop: \"conv4\"\n}\nlayer {\n\tname: \"conv5\"\n\ttype: \"Convolution\"\n\tbottom: \"conv4\"\n\ttop: \"conv5\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t}\n}\nlayer {\n\tname: \"relu5\"\n\ttype: \"ReLU\"\n\tbottom: \"conv5\"\n\ttop: \"conv5\"\n}\n\n#========= RPN ============\n\nlayer {\n name: \"rpn_conv/3x3\"\n type: \"Convolution\"\n bottom: \"conv5\"\n top: \"rpn_conv/3x3\"\n convolution_param {\n num_output: 256\n kernel_size: 3 pad: 1 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_relu/3x3\"\n type: \"ReLU\"\n bottom: \"rpn_conv/3x3\"\n top: \"rpn_conv/3x3\"\n}\n#layer {\n# name: \"rpn_conv/3x3\"\n# type: \"Convolution\"\n# bottom: \"conv5\"\n# top: \"rpn_conv/3x3\"\n# param { lr_mult: 1.0 decay_mult: 1.0 }\n# param { lr_mult: 2.0 decay_mult: 0 }\n# convolution_param {\n# num_output: 192\n# kernel_size: 3 pad: 1 stride: 1\n# weight_filler { type: \"gaussian\" std: 0.01 }\n# bias_filler { type: \"constant\" value: 0 }\n# }\n#}\n#layer {\n# name: \"rpn_conv/5x5\"\n# type: \"Convolution\"\n# bottom: \"conv5\"\n# top: \"rpn_conv/5x5\"\n# param { lr_mult: 1.0 decay_mult: 1.0 }\n# param { lr_mult: 2.0 decay_mult: 0 }\n# convolution_param {\n# num_output: 64\n# kernel_size: 5 pad: 2 stride: 1\n# weight_filler { type: \"gaussian\" std: 0.0036 }\n# bias_filler { type: \"constant\" value: 0 }\n# }\n#}\n#layer {\n# name: \"rpn/output\"\n# type: \"Concat\"\n# bottom: \"rpn_conv/3x3\"\n# bottom: \"rpn_conv/5x5\"\n# top: \"rpn/output\"\n#}\n#layer {\n# name: \"rpn_relu/output\"\n# type: \"ReLU\"\n# bottom: \"rpn/output\"\n# top: \"rpn/output\"\n#}\nlayer {\n name: \"rpn_cls_score\"\n type: \"Convolution\"\n bottom: \"rpn_conv/3x3\"\n top: \"rpn_cls_score\"\n convolution_param {\n num_output: 18 # 2(bg/fg) * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_bbox_pred\"\n type: \"Convolution\"\n bottom: \"rpn_conv/3x3\"\n top: \"rpn_bbox_pred\"\n convolution_param {\n num_output: 36 # 4 * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n bottom: \"rpn_cls_score\"\n top: \"rpn_cls_score_reshape\"\n name: \"rpn_cls_score_reshape\"\n type: \"Reshape\"\n reshape_param { shape { dim: 0 dim: 2 dim: -1 dim: 0 } }\n}\n\n#========= RoI Proposal ============\n\nlayer {\n name: \"rpn_cls_prob\"\n type: \"Softmax\"\n bottom: \"rpn_cls_score_reshape\"\n top: \"rpn_cls_prob\"\n}\nlayer {\n name: 'rpn_cls_prob_reshape'\n type: 'Reshape'\n bottom: 'rpn_cls_prob'\n top: 'rpn_cls_prob_reshape'\n reshape_param { shape { dim: 0 dim: 18 dim: -1 dim: 0 } }\n}\nlayer {\n name: 'rpn_proposals'\n type: 'Python'\n bottom: 'rpn_cls_prob_reshape'\n bottom: 'rpn_bbox_pred'\n bottom: 'im_info'\n top: 'rpn_proposals'\n python_param {\n module: 'rpn.proposal_layer'\n layer: 'ProposalLayer'\n param_str: \"'feat_stride': 16\"\n }\n}\n\n#========= RCNN ============\n\nlayer {\n name: \"roi_pool_conv5\"\n type: \"ROIPooling\"\n bottom: \"conv5\"\n bottom: \"rpn_proposals\"\n top: \"roi_pool_conv5\"\n roi_pooling_param {\n pooled_w: 6\n pooled_h: 6\n spatial_scale: 0.0625 # 1/16\n }\n}\nlayer {\n name: \"fc6\"\n type: \"InnerProduct\"\n bottom: \"roi_pool_conv5\"\n top: \"fc6\"\n inner_product_param {\n num_output: 4096\n }\n}\nlayer {\n name: \"relu6\"\n type: \"ReLU\"\n bottom: \"fc6\"\n top: \"fc6\"\n}\nlayer {\n name: \"drop6\"\n type: \"Dropout\"\n bottom: \"fc6\"\n top: \"fc6\"\n dropout_param {\n dropout_ratio: 0.5\n scale_train: false\n }\n}\nlayer {\n name: \"fc7\"\n type: \"InnerProduct\"\n bottom: \"fc6\"\n top: \"fc7\"\n inner_product_param {\n num_output: 4096\n }\n}\nlayer {\n name: \"relu7\"\n type: \"ReLU\"\n bottom: \"fc7\"\n top: \"fc7\"\n}\nlayer {\n name: \"drop7\"\n type: \"Dropout\"\n bottom: \"fc7\"\n top: \"fc7\"\n dropout_param {\n dropout_ratio: 0.5\n scale_train: false\n }\n}\nlayer {\n name: \"cls_score\"\n type: \"InnerProduct\"\n bottom: \"fc7\"\n top: \"cls_score\"\n inner_product_param {\n num_output: 21\n }\n}\nlayer {\n name: \"bbox_pred\"\n type: \"InnerProduct\"\n bottom: \"fc7\"\n top: \"bbox_pred\"\n inner_product_param {\n num_output: 84\n }\n}\nlayer {\n name: \"bbox_out\"\n type: \"implicit\"\n bottom: \"bbox_pred\"\n}\nlayer {\n name: \"cls_prob\"\n type: \"Softmax\"\n bottom: \"cls_score\"\n top: \"cls_prob\"\n loss_param {\n ignore_label: -1\n normalize: true\n }\n}\n" }, { "path": "presets/fasterRCNN_ResNet.prototxt", "content": "name: \"Faster R-CNN (ResNet-50)\"\n\ninput: \"data\"\ninput_shape {\n dim: 1\n dim: 3\n dim: 500\n dim: 400\n}\n\ninput: \"im_info\"\ninput_shape {\n dim: 1\n dim: 3\n}\n\n#========= BASE CNN, FULLY_CONVOLUTIONAL ============\nlayer {\n\tbottom: \"data\"\n\ttop: \"conv1\"\n\tname: \"conv1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 7\n\t\tpad: 3\n\t\tstride: 2\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"bn_conv1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"scale_conv1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"conv1_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"pool1\"\n\tname: \"pool1\"\n\ttype: \"Pooling\"\n\tpooling_param {\n\t\tkernel_size: 3\n\t\tstride: 2\n\t\tpool: MAX\n\t}\n}\n\nlayer {\n\tbottom: \"pool1\"\n\ttop: \"res2a_branch1\"\n\tname: \"res2a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\ttop: \"res2a_branch1\"\n\tname: \"bn2a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\ttop: \"res2a_branch1\"\n\tname: \"scale2a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"pool1\"\n\ttop: \"res2a_branch2a\"\n\tname: \"res2a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"bn2a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"scale2a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"res2a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2b\"\n\tname: \"res2a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"bn2a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"scale2a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"res2a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2c\"\n\tname: \"res2a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a_branch2c\"\n\tname: \"bn2a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a_branch2c\"\n\tname: \"scale2a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a\"\n\tname: \"res2a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2a\"\n\ttop: \"res2a\"\n\tname: \"res2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"res2b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"bn2b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"scale2b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"res2b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2b\"\n\tname: \"res2b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"bn2b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"scale2b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"res2b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2c\"\n\tname: \"res2b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b_branch2c\"\n\tname: \"bn2b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b_branch2c\"\n\tname: \"scale2b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a\"\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b\"\n\tname: \"res2b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2b\"\n\ttop: \"res2b\"\n\tname: \"res2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b\"\n\ttop: \"res2c_branch2a\"\n\tname: \"res2c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"bn2c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"scale2c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"res2c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2b\"\n\tname: \"res2c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"bn2c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"scale2c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"res2c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2c\"\n\tname: \"res2c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c_branch2c\"\n\tname: \"bn2c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c_branch2c\"\n\tname: \"scale2c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b\"\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c\"\n\tname: \"res2c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res2c\"\n\tname: \"res2c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res3a_branch1\"\n\tname: \"res3a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\ttop: \"res3a_branch1\"\n\tname: \"bn3a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\ttop: \"res3a_branch1\"\n\tname: \"scale3a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res3a_branch2a\"\n\tname: \"res3a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"bn3a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"scale3a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"res3a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2b\"\n\tname: \"res3a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"bn3a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"scale3a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"res3a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2c\"\n\tname: \"res3a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a_branch2c\"\n\tname: \"bn3a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a_branch2c\"\n\tname: \"scale3a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a\"\n\tname: \"res3a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3a\"\n\ttop: \"res3a\"\n\tname: \"res3a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"res3b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"bn3b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"scale3b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"res3b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2b\"\n\tname: \"res3b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2b\"\n\tname: \"bn3b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2b\"\n\tname: \"scale3b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2b\"\n\tname: \"res3b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2c\"\n\tname: \"res3b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2c\"\n\ttop: \"res3b_branch2c\"\n\tname: \"bn3b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2c\"\n\ttop: \"res3b_branch2c\"\n\tname: \"scale3b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a\"\n\tbottom: \"res3b_branch2c\"\n\ttop: \"res3b\"\n\tname: \"res3b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b\"\n\ttop: \"res3b\"\n\tname: \"res3b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b\"\n\ttop: \"res3c_branch2a\"\n\tname: \"res3c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2a\"\n\tname: \"bn3c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2a\"\n\tname: \"scale3c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2a\"\n\tname: \"res3c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2b\"\n\tname: \"res3c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2b\"\n\tname: \"bn3c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2b\"\n\tname: \"scale3c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2b\"\n\tname: \"res3c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2c\"\n\tname: \"res3c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2c\"\n\ttop: \"res3c_branch2c\"\n\tname: \"bn3c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2c\"\n\ttop: \"res3c_branch2c\"\n\tname: \"scale3c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b\"\n\tbottom: \"res3c_branch2c\"\n\ttop: \"res3c\"\n\tname: \"res3c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3c\"\n\ttop: \"res3c\"\n\tname: \"res3c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3c\"\n\ttop: \"res3d_branch2a\"\n\tname: \"res3d_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2a\"\n\tname: \"bn3d_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2a\"\n\tname: \"scale3d_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2a\"\n\tname: \"res3d_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2b\"\n\tname: \"res3d_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2b\"\n\tname: \"bn3d_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2b\"\n\tname: \"scale3d_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2b\"\n\tname: \"res3d_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2c\"\n\tname: \"res3d_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2c\"\n\ttop: \"res3d_branch2c\"\n\tname: \"bn3d_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2c\"\n\ttop: \"res3d_branch2c\"\n\tname: \"scale3d_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c\"\n\tbottom: \"res3d_branch2c\"\n\ttop: \"res3d\"\n\tname: \"res3d\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3d\"\n\ttop: \"res3d\"\n\tname: \"res3d_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3d\"\n\ttop: \"res4a_branch1\"\n\tname: \"res4a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\ttop: \"res4a_branch1\"\n\tname: \"bn4a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\ttop: \"res4a_branch1\"\n\tname: \"scale4a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d\"\n\ttop: \"res4a_branch2a\"\n\tname: \"res4a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"bn4a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"scale4a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"res4a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2b\"\n\tname: \"res4a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"bn4a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"scale4a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"res4a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2c\"\n\tname: \"res4a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a_branch2c\"\n\tname: \"bn4a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a_branch2c\"\n\tname: \"scale4a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a\"\n\tname: \"res4a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4a\"\n\ttop: \"res4a\"\n\tname: \"res4a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"res4b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"bn4b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"scale4b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"res4b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2b\"\n\tname: \"res4b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2b\"\n\tname: \"bn4b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2b\"\n\tname: \"scale4b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2b\"\n\tname: \"res4b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2c\"\n\tname: \"res4b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2c\"\n\ttop: \"res4b_branch2c\"\n\tname: \"bn4b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2c\"\n\ttop: \"res4b_branch2c\"\n\tname: \"scale4b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a\"\n\tbottom: \"res4b_branch2c\"\n\ttop: \"res4b\"\n\tname: \"res4b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b\"\n\ttop: \"res4b\"\n\tname: \"res4b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b\"\n\ttop: \"res4c_branch2a\"\n\tname: \"res4c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2a\"\n\tname: \"bn4c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2a\"\n\tname: \"scale4c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2a\"\n\tname: \"res4c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2b\"\n\tname: \"res4c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2b\"\n\tname: \"bn4c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2b\"\n\tname: \"scale4c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2b\"\n\tname: \"res4c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2c\"\n\tname: \"res4c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2c\"\n\ttop: \"res4c_branch2c\"\n\tname: \"bn4c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2c\"\n\ttop: \"res4c_branch2c\"\n\tname: \"scale4c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b\"\n\tbottom: \"res4c_branch2c\"\n\ttop: \"res4c\"\n\tname: \"res4c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4c\"\n\ttop: \"res4c\"\n\tname: \"res4c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4c\"\n\ttop: \"res4d_branch2a\"\n\tname: \"res4d_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2a\"\n\tname: \"bn4d_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2a\"\n\tname: \"scale4d_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2a\"\n\tname: \"res4d_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2b\"\n\tname: \"res4d_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2b\"\n\tname: \"bn4d_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2b\"\n\tname: \"scale4d_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2b\"\n\tname: \"res4d_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2c\"\n\tname: \"res4d_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2c\"\n\ttop: \"res4d_branch2c\"\n\tname: \"bn4d_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2c\"\n\ttop: \"res4d_branch2c\"\n\tname: \"scale4d_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c\"\n\tbottom: \"res4d_branch2c\"\n\ttop: \"res4d\"\n\tname: \"res4d\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4d\"\n\ttop: \"res4d\"\n\tname: \"res4d_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4d\"\n\ttop: \"res4e_branch2a\"\n\tname: \"res4e_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2a\"\n\tname: \"bn4e_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2a\"\n\tname: \"scale4e_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2a\"\n\tname: \"res4e_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2b\"\n\tname: \"res4e_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2b\"\n\tname: \"bn4e_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2b\"\n\tname: \"scale4e_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2b\"\n\tname: \"res4e_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2c\"\n\tname: \"res4e_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2c\"\n\ttop: \"res4e_branch2c\"\n\tname: \"bn4e_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2c\"\n\ttop: \"res4e_branch2c\"\n\tname: \"scale4e_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d\"\n\tbottom: \"res4e_branch2c\"\n\ttop: \"res4e\"\n\tname: \"res4e\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4e\"\n\ttop: \"res4e\"\n\tname: \"res4e_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4e\"\n\ttop: \"res4f_branch2a\"\n\tname: \"res4f_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2a\"\n\tname: \"bn4f_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2a\"\n\tname: \"scale4f_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2a\"\n\tname: \"res4f_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2b\"\n\tname: \"res4f_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2b\"\n\tname: \"bn4f_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2b\"\n\tname: \"scale4f_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2b\"\n\tname: \"res4f_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2c\"\n\tname: \"res4f_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2c\"\n\ttop: \"res4f_branch2c\"\n\tname: \"bn4f_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2c\"\n\ttop: \"res4f_branch2c\"\n\tname: \"scale4f_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e\"\n\tbottom: \"res4f_branch2c\"\n\ttop: \"res4f\"\n\tname: \"res4f\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4f\"\n\ttop: \"res4f\"\n\tname: \"res4f_relu\"\n\ttype: \"ReLU\"\n}\n\n\nlayer {\n\tbottom: \"res4f\"\n\ttop: \"res5a_branch1\"\n\tname: \"res5a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\ttop: \"res5a_branch1\"\n\tname: \"bn5a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\ttop: \"res5a_branch1\"\n\tname: \"scale5a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f\"\n\ttop: \"res5a_branch2a\"\n\tname: \"res5a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"bn5a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"scale5a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"res5a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2b\"\n\tname: \"res5a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"bn5a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"scale5a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"res5a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2c\"\n\tname: \"res5a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a_branch2c\"\n\tname: \"bn5a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a_branch2c\"\n\tname: \"scale5a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a\"\n\tname: \"res5a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5a\"\n\ttop: \"res5a\"\n\tname: \"res5a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"res5b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"bn5b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"scale5b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"res5b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2b\"\n\tname: \"res5b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"bn5b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"scale5b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"res5b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2c\"\n\tname: \"res5b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b_branch2c\"\n\tname: \"bn5b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b_branch2c\"\n\tname: \"scale5b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a\"\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b\"\n\tname: \"res5b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5b\"\n\ttop: \"res5b\"\n\tname: \"res5b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b\"\n\ttop: \"res5c_branch2a\"\n\tname: \"res5c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"bn5c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"scale5c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"res5c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2b\"\n\tname: \"res5c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"bn5c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"scale5c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"res5c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2c\"\n\tname: \"res5c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c_branch2c\"\n\tname: \"bn5c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c_branch2c\"\n\tname: \"scale5c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b\"\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c\"\n\tname: \"res5c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5c\"\n\ttop: \"res5c\"\n\tname: \"res5c_relu\"\n\ttype: \"ReLU\"\n}\n\n\n#========= RPN ============\n\nlayer {\n name: \"rpn_conv/3x3/output\"\n type: \"Convolution\"\n bottom: \"res5c\"\n top: \"rpn_conv/3x3/output\"\n convolution_param {\n num_output: 256\n kernel_size: 3 pad: 1 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_relu/3x3\"\n type: \"ReLU\"\n bottom: \"rpn_conv/3x3/output\"\n top: \"rpn_conv/3x3/output\"\n}\nlayer {\n name: \"rpn_cls_score\"\n type: \"Convolution\"\n bottom: \"rpn_conv/3x3/output\"\n top: \"rpn_cls_score\"\n convolution_param {\n num_output: 18 # 2(bg/fg) * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_bbox_pred\"\n type: \"Convolution\"\n bottom: \"rpn_conv/3x3/output\"\n top: \"rpn_bbox_pred\"\n convolution_param {\n num_output: 36 # 4 * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n bottom: \"rpn_cls_score\"\n top: \"rpn_cls_score_reshape\"\n name: \"rpn_cls_score_reshape\"\n type: \"Reshape\"\n reshape_param { shape { dim: 0 dim: 2 dim: -1 dim: 0 } }\n}\n\n#========= RoI Proposal ============\n\nlayer {\n name: \"rpn_cls_prob\"\n type: \"Softmax\"\n bottom: \"rpn_cls_score_reshape\"\n top: \"rpn_cls_prob\"\n}\nlayer {\n name: 'rpn_cls_prob_reshape'\n type: 'Reshape'\n bottom: 'rpn_cls_prob'\n top: 'rpn_cls_prob_reshape'\n reshape_param { shape { dim: 0 dim: 18 dim: -1 dim: 0 } }\n}\nlayer {\n name: 'rpn_proposals'\n type: 'Python'\n bottom: 'rpn_cls_prob_reshape'\n bottom: 'rpn_bbox_pred'\n bottom: 'im_info'\n top: 'rpn_proposals'\n python_param {\n module: 'rpn.proposal_layer'\n layer: 'ProposalLayer'\n param_str: \"'feat_stride': 16\"\n }\n}\n\n#========= ROI POOLING ============\n\nlayer {\n name: \"roi_pool\"\n type: \"ROIPooling\"\n bottom: \"res5c\"\n bottom: \"rpn_proposals\"\n top: \"roi_pool\"\n roi_pooling_param {\n pooled_w: 6\n pooled_h: 6\n spatial_scale: 0.0625 # 1/16\n }\n}\n\n# ====== BASE CNN, PART B =======\n\nlayer {\n\tbottom: \"roi_pool\"\n\ttop: \"pool5\"\n\tname: \"pool5\"\n\ttype: \"Pooling\"\n\tpooling_param {\n\t#\tglobal_pooling: 1\n kernel_size: 6\n\t\tstride: 1\n\t\tpool: AVE\n\t}\n}\n\nlayer {\n\tbottom: \"pool5\"\n\ttop: \"fc1000\"\n\tname: \"fc1000\"\n\ttype: \"InnerProduct\"\n\tinner_product_param {\n\t\tnum_output: 1000\n\t}\n}\n\n# ====== CLASSIFICATION =======\n\nlayer {\n name: \"cls_score\"\n type: \"InnerProduct\"\n bottom: \"fc1000\"\n top: \"cls_score\"\n inner_product_param {\n num_output: 21\n }\n}\nlayer {\n name: \"bbox_pred\"\n type: \"InnerProduct\"\n bottom: \"fc1000\"\n top: \"bbox_pred\"\n inner_product_param {\n num_output: 84\n }\n}\nlayer {\n name: \"bbox_out\"\n type: \"implicit\"\n bottom: \"bbox_pred\"\n}\nlayer {\n name: \"cls_prob\"\n type: \"Softmax\"\n bottom: \"cls_score\"\n top: \"cls_prob\"\n loss_param {\n ignore_label: -1\n normalize: true\n }\n}\n" }, { "path": "presets/fasterRCNN_VGG.prototxt", "content": "name: \"Faster R-CNN (VGG16)\"\n\ninput: \"data\"\ninput_shape {\n dim: 1\n dim: 3\n dim: 400\n dim: 500\n}\n\ninput: \"im_info\"\ninput_shape {\n dim: 1\n dim: 3\n}\n\nlayer {\n name: \"conv1_1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1_1\"\n param {\n lr_mult: 0\n decay_mult: 0\n }\n param {\n lr_mult: 0\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu1_1\"\n type: \"ReLU\"\n bottom: \"conv1_1\"\n top: \"conv1_1\"\n}\nlayer {\n name: \"conv1_2\"\n type: \"Convolution\"\n bottom: \"conv1_1\"\n top: \"conv1_2\"\n param {\n lr_mult: 0\n decay_mult: 0\n }\n param {\n lr_mult: 0\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu1_2\"\n type: \"ReLU\"\n bottom: \"conv1_2\"\n top: \"conv1_2\"\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv1_2\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv2_1\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"conv2_1\"\n param {\n lr_mult: 0\n decay_mult: 0\n }\n param {\n lr_mult: 0\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu2_1\"\n type: \"ReLU\"\n bottom: \"conv2_1\"\n top: \"conv2_1\"\n}\nlayer {\n name: \"conv2_2\"\n type: \"Convolution\"\n bottom: \"conv2_1\"\n top: \"conv2_2\"\n param {\n lr_mult: 0\n decay_mult: 0\n }\n param {\n lr_mult: 0\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu2_2\"\n type: \"ReLU\"\n bottom: \"conv2_2\"\n top: \"conv2_2\"\n}\nlayer {\n name: \"pool2\"\n type: \"Pooling\"\n bottom: \"conv2_2\"\n top: \"pool2\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv3_1\"\n type: \"Convolution\"\n bottom: \"pool2\"\n top: \"conv3_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu3_1\"\n type: \"ReLU\"\n bottom: \"conv3_1\"\n top: \"conv3_1\"\n}\nlayer {\n name: \"conv3_2\"\n type: \"Convolution\"\n bottom: \"conv3_1\"\n top: \"conv3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu3_2\"\n type: \"ReLU\"\n bottom: \"conv3_2\"\n top: \"conv3_2\"\n}\nlayer {\n name: \"conv3_3\"\n type: \"Convolution\"\n bottom: \"conv3_2\"\n top: \"conv3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu3_3\"\n type: \"ReLU\"\n bottom: \"conv3_3\"\n top: \"conv3_3\"\n}\nlayer {\n name: \"pool3\"\n type: \"Pooling\"\n bottom: \"conv3_3\"\n top: \"pool3\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv4_1\"\n type: \"Convolution\"\n bottom: \"pool3\"\n top: \"conv4_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu4_1\"\n type: \"ReLU\"\n bottom: \"conv4_1\"\n top: \"conv4_1\"\n}\nlayer {\n name: \"conv4_2\"\n type: \"Convolution\"\n bottom: \"conv4_1\"\n top: \"conv4_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu4_2\"\n type: \"ReLU\"\n bottom: \"conv4_2\"\n top: \"conv4_2\"\n}\nlayer {\n name: \"conv4_3\"\n type: \"Convolution\"\n bottom: \"conv4_2\"\n top: \"conv4_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu4_3\"\n type: \"ReLU\"\n bottom: \"conv4_3\"\n top: \"conv4_3\"\n}\nlayer {\n name: \"pool4\"\n type: \"Pooling\"\n bottom: \"conv4_3\"\n top: \"pool4\"\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayer {\n name: \"conv5_1\"\n type: \"Convolution\"\n bottom: \"pool4\"\n top: \"conv5_1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu5_1\"\n type: \"ReLU\"\n bottom: \"conv5_1\"\n top: \"conv5_1\"\n}\nlayer {\n name: \"conv5_2\"\n type: \"Convolution\"\n bottom: \"conv5_1\"\n top: \"conv5_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu5_2\"\n type: \"ReLU\"\n bottom: \"conv5_2\"\n top: \"conv5_2\"\n}\nlayer {\n name: \"conv5_3\"\n type: \"Convolution\"\n bottom: \"conv5_2\"\n top: \"conv5_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayer {\n name: \"relu5_3\"\n type: \"ReLU\"\n bottom: \"conv5_3\"\n top: \"conv5_3\"\n}\n\n#========= RPN ============\n\nlayer {\n name: \"rpn/conv3x3/output\"\n type: \"Convolution\"\n bottom: \"conv5_3\"\n top: \"rpn/conv3x3/output\"\n param { lr_mult: 1.0 decay_mult: 1.0 }\n param { lr_mult: 2.0 decay_mult: 0 }\n convolution_param {\n num_output: 512\n kernel_size: 3 pad: 1 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_relu/3x3\"\n type: \"ReLU\"\n bottom: \"rpn/conv3x3/output\"\n top: \"rpn/conv3x3/output\"\n}\n\nlayer {\n name: \"rpn_cls_score\"\n type: \"Convolution\"\n bottom: \"rpn/conv3x3/output\"\n top: \"rpn_cls_score\"\n param { lr_mult: 1.0 decay_mult: 1.0 }\n param { lr_mult: 2.0 decay_mult: 0 }\n convolution_param {\n num_output: 18 # 2(bg/fg) * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_bbox_pred\"\n type: \"Convolution\"\n bottom: \"rpn/conv3x3/output\"\n top: \"rpn_bbox_pred\"\n param { lr_mult: 1.0 decay_mult: 1.0 }\n param { lr_mult: 2.0 decay_mult: 0 }\n convolution_param {\n num_output: 36 # 4 * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n bottom: \"rpn_cls_score\"\n top: \"rpn_cls_score_reshape\"\n name: \"rpn_cls_score_reshape\"\n type: \"Reshape\"\n reshape_param { shape { dim: 0 dim: 2 dim: -1 dim: 0 } }\n}\n\n#========= RoI Proposal ============\n\nlayer {\n name: \"rpn_cls_prob\"\n type: \"Softmax\"\n bottom: \"rpn_cls_score_reshape\"\n top: \"rpn_cls_prob\"\n}\nlayer {\n name: \"rpn_cls_prob_reshape\"\n type: \"Reshape\"\n bottom: \"rpn_cls_prob\"\n top: \"rpn_cls_prob_reshape\"\n reshape_param { shape { dim: 0 dim: 18 dim: -1 dim: 0 } }\n}\nlayer {\n name: \"roi_proposals\"\n type: \"Python\"\n bottom: \"rpn_cls_prob_reshape\"\n bottom: \"rpn_bbox_pred\"\n bottom: \"im_info\"\n top: \"roi_proposals\"\n python_param {\n module: \"rpn.proposal_layer\"\n layer: \"ProposalLayer\"\n param_str: \"'feat_stride': 16\"\n }\n}\n\n#========= RCNN ============\n\nlayer {\n name: \"roi_pool5\"\n type: \"ROIPooling\"\n bottom: \"conv5_3\"\n bottom: \"roi_proposals\"\n top: \"roi_pool5\"\n roi_pooling_param {\n pooled_w: 7\n pooled_h: 7\n spatial_scale: 0.0625 # 1/16\n }\n}\nlayer {\n name: \"fc6\"\n type: \"InnerProduct\"\n bottom: \"roi_pool5\"\n top: \"fc6\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 4096\n }\n}\nlayer {\n name: \"relu6\"\n type: \"ReLU\"\n bottom: \"fc6\"\n top: \"fc6\"\n}\nlayer {\n name: \"drop6\"\n type: \"Dropout\"\n bottom: \"fc6\"\n top: \"fc6\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"fc7\"\n type: \"InnerProduct\"\n bottom: \"fc6\"\n top: \"fc7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 4096\n }\n}\nlayer {\n name: \"relu7\"\n type: \"ReLU\"\n bottom: \"fc7\"\n top: \"fc7\"\n}\nlayer {\n name: \"drop7\"\n type: \"Dropout\"\n bottom: \"fc7\"\n top: \"fc7\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"cls_score\"\n type: \"InnerProduct\"\n bottom: \"fc7\"\n top: \"cls_score\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 21\n weight_filler {\n type: \"gaussian\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"bbox_pred\"\n type: \"InnerProduct\"\n bottom: \"fc7\"\n top: \"bbox_pred\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 84\n weight_filler {\n type: \"gaussian\"\n std: 0.001\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"bbox_out\"\n type: \"implicit\"\n bottom: \"bbox_pred\"\n}\nlayer {\n name: \"cls_prob\"\n type: \"Softmax\"\n bottom: \"cls_score\"\n top: \"cls_prob\"\n}\n" }, { "path": "presets/fasterRCNN_ZynqNet.prototxt", "content": "name: \"Faster R-CNN (ZynqNet)\"\n\ninput: \"data\"\ninput_shape {\n dim: 1\n dim: 3\n dim: 500\n dim: 400\n}\n\ninput: \"im_info\"\ninput_shape {\n dim: 1\n dim: 3\n}\n\n#========= BASE CNN, FULLY_CONVOLUTIONAL ============\n\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"relu_conv1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n}\nlayer {\n name: \"fire2/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"conv1\"\n top: \"fire2/squeeze3x3\"\n convolution_param {\n num_output: 16\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/squeeze3x3\"\n}\nlayer {\n name: \"fire2/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire2/expand1x1\"\n top: \"fire2/expand1x1\"\n}\nlayer {\n name: \"fire2/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/expand3x3\"\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/expand3x3\"\n}\nlayer {\n name: \"fire2/concat\"\n type: \"Concat\"\n bottom: \"fire2/expand1x1\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/concat\"\n}\nlayer {\n name: \"fire3/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire2/concat\"\n top: \"fire3/squeeze1x1\"\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/squeeze1x1\"\n}\nlayer {\n name: \"fire3/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire3/expand1x1\"\n top: \"fire3/expand1x1\"\n}\nlayer {\n name: \"fire3/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand3x3\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/expand3x3\"\n}\nlayer {\n name: \"fire3/concat\"\n type: \"Concat\"\n bottom: \"fire3/expand1x1\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/concat\"\n}\nlayer {\n name: \"fire4/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire3/concat\"\n top: \"fire4/squeeze3x3\"\n convolution_param {\n num_output: 32\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/squeeze3x3\"\n}\nlayer {\n name: \"fire4/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire4/expand1x1\"\n top: \"fire4/expand1x1\"\n}\nlayer {\n name: \"fire4/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/expand3x3\"\n convolution_param {\n num_output: 128\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/expand3x3\"\n}\nlayer {\n name: \"fire4/concat\"\n type: \"Concat\"\n bottom: \"fire4/expand1x1\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/concat\"\n}\nlayer {\n name: \"fire5/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire4/concat\"\n top: \"fire5/squeeze1x1\"\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/squeeze1x1\"\n}\nlayer {\n name: \"fire5/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire5/expand1x1\"\n top: \"fire5/expand1x1\"\n}\nlayer {\n name: \"fire5/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand3x3\"\n convolution_param {\n num_output: 128\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/expand3x3\"\n}\nlayer {\n name: \"fire5/concat\"\n type: \"Concat\"\n bottom: \"fire5/expand1x1\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/concat\"\n}\nlayer {\n name: \"fire6/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire5/concat\"\n top: \"fire6/squeeze3x3\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/squeeze3x3\"\n}\nlayer {\n name: \"fire6/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire6/expand1x1\"\n top: \"fire6/expand1x1\"\n}\nlayer {\n name: \"fire6/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/expand3x3\"\n convolution_param {\n num_output: 256\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/expand3x3\"\n}\nlayer {\n name: \"fire6/concat\"\n type: \"Concat\"\n bottom: \"fire6/expand1x1\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/concat\"\n}\nlayer {\n name: \"fire7/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire6/concat\"\n top: \"fire7/squeeze1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/squeeze1x1\"\n}\nlayer {\n name: \"fire7/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand1x1\"\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire7/expand1x1\"\n top: \"fire7/expand1x1\"\n}\nlayer {\n name: \"fire7/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand3x3\"\n convolution_param {\n num_output: 192\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/expand3x3\"\n}\nlayer {\n name: \"fire7/concat\"\n type: \"Concat\"\n bottom: \"fire7/expand1x1\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/concat\"\n}\nlayer {\n name: \"fire8/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire7/concat\"\n top: \"fire8/squeeze3x3\"\n convolution_param {\n num_output: 112\n kernel_size: 3\n pad: 1\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/squeeze3x3\"\n}\nlayer {\n name: \"fire8/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire8/expand1x1\"\n top: \"fire8/expand1x1\"\n}\nlayer {\n name: \"fire8/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/expand3x3\"\n convolution_param {\n num_output: 256\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/expand3x3\"\n}\nlayer {\n name: \"fire8/concat\"\n type: \"Concat\"\n bottom: \"fire8/expand1x1\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/concat\"\n}\n\nlayer {\n name: \"fire9/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire8/concat\"\n top: \"fire9/squeeze1x1\"\n convolution_param {\n num_output: 112\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/squeeze1x1\"\n}\nlayer {\n name: \"fire9/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand1x1\"\n convolution_param {\n num_output: 368\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire9/expand1x1\"\n top: \"fire9/expand1x1\"\n}\nlayer {\n name: \"fire9/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand3x3\"\n convolution_param {\n num_output: 368\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/expand3x3\"\n}\nlayer {\n name: \"fire9/concat\"\n type: \"Concat\"\n bottom: \"fire9/expand1x1\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/concat\"\n}\nlayer {\n name: \"drop9\"\n type: \"Dropout\"\n bottom: \"fire9/concat\"\n top: \"fire9/concat\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\n\n#========= RPN ============\n\nlayer {\n name: \"rpn_conv/3x3/output\"\n type: \"Convolution\"\n bottom: \"fire9/concat\"\n top: \"rpn_conv/3x3/output\"\n convolution_param {\n num_output: 256\n kernel_size: 3 pad: 1 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_relu/3x3\"\n type: \"ReLU\"\n bottom: \"rpn_conv/3x3/output\"\n top: \"rpn_conv/3x3/output\"\n}\nlayer {\n name: \"rpn_cls_score\"\n type: \"Convolution\"\n bottom: \"rpn_conv/3x3/output\"\n top: \"rpn_cls_score\"\n convolution_param {\n num_output: 18 # 2(bg/fg) * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n name: \"rpn_bbox_pred\"\n type: \"Convolution\"\n bottom: \"rpn_conv/3x3/output\"\n top: \"rpn_bbox_pred\"\n convolution_param {\n num_output: 36 # 4 * 9(anchors)\n kernel_size: 1 pad: 0 stride: 1\n weight_filler { type: \"gaussian\" std: 0.01 }\n bias_filler { type: \"constant\" value: 0 }\n }\n}\nlayer {\n bottom: \"rpn_cls_score\"\n top: \"rpn_cls_score_reshape\"\n name: \"rpn_cls_score_reshape\"\n type: \"Reshape\"\n reshape_param { shape { dim: 0 dim: 2 dim: -1 dim: 0 } }\n}\n\n#========= RoI Proposal ============\n\nlayer {\n name: \"rpn_cls_prob\"\n type: \"Softmax\"\n bottom: \"rpn_cls_score_reshape\"\n top: \"rpn_cls_prob\"\n}\nlayer {\n name: 'rpn_cls_prob_reshape'\n type: 'Reshape'\n bottom: 'rpn_cls_prob'\n top: 'rpn_cls_prob_reshape'\n reshape_param { shape { dim: 0 dim: 18 dim: -1 dim: 0 } }\n}\nlayer {\n name: 'rpn_proposals'\n type: 'Python'\n bottom: 'rpn_cls_prob_reshape'\n bottom: 'rpn_bbox_pred'\n bottom: 'im_info'\n top: 'rpn_proposals'\n python_param {\n module: 'rpn.proposal_layer'\n layer: 'ProposalLayer'\n param_str: \"'feat_stride': 16\"\n }\n}\n\n#========= ROI POOLING ============\n\nlayer {\n name: \"roi_pool\"\n type: \"ROIPooling\"\n bottom: \"fire9/concat\"\n bottom: \"rpn_proposals\"\n top: \"roi_pool\"\n roi_pooling_param {\n pooled_w: 6\n pooled_h: 6\n spatial_scale: 0.0625 # 1/16\n }\n}\n\n# ====== BASE CNN, PART B =======\n\nlayer {\n name: \"conv10\"\n type: \"Convolution\"\n bottom: \"roi_pool\"\n top: \"conv10\"\n convolution_param {\n num_output: 1024\n kernel_size: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0.0\n std: 0.01\n }\n }\n}\n\n# ====== CLASSIFICATION =======\n\nlayer {\n name: \"cls_score\"\n type: \"InnerProduct\"\n bottom: \"conv10\"\n top: \"cls_score\"\n inner_product_param {\n num_output: 21\n }\n}\nlayer {\n name: \"bbox_pred\"\n type: \"InnerProduct\"\n bottom: \"conv10\"\n top: \"bbox_pred\"\n inner_product_param {\n num_output: 84\n }\n}\nlayer {\n name: \"bbox_out\"\n type: \"implicit\"\n bottom: \"bbox_pred\"\n}\nlayer {\n name: \"cls_prob\"\n type: \"Softmax\"\n bottom: \"cls_score\"\n top: \"cls_prob\"\n loss_param {\n ignore_label: -1\n normalize: true\n }\n}\n" }, { "path": "presets/fcn-16s.prototxt", "content": "name: 'FCN 16s'\n\nlayer {\n type: 'data'\n name: 'data'\n top: 'data'\n input_param {\n shape {\n dim: 1\n dim: 3\n dim: 500\n dim: 500\n }\n }\nforce_backward: true\n}\n\nlayers { bottom: 'data' top: 'conv1_1' name: 'conv1_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 64 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv1_1' top: 'conv1_1' name: 'relu1_1' type: RELU }\nlayers { bottom: 'conv1_1' top: 'conv1_2' name: 'conv1_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 64 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv1_2' top: 'conv1_2' name: 'relu1_2' type: RELU }\nlayers { name: 'pool1' bottom: 'conv1_2' top: 'pool1' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { name: 'conv2_1' bottom: 'pool1' top: 'conv2_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 128 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv2_1' top: 'conv2_1' name: 'relu2_1' type: RELU }\nlayers { bottom: 'conv2_1' top: 'conv2_2' name: 'conv2_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 128 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv2_2' top: 'conv2_2' name: 'relu2_2' type: RELU }\nlayers { bottom: 'conv2_2' top: 'pool2' name: 'pool2' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool2' top: 'conv3_1' name: 'conv3_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv3_1' top: 'conv3_1' name: 'relu3_1' type: RELU }\nlayers { bottom: 'conv3_1' top: 'conv3_2' name: 'conv3_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv3_2' top: 'conv3_2' name: 'relu3_2' type: RELU }\nlayers { bottom: 'conv3_2' top: 'conv3_3' name: 'conv3_3' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv3_3' top: 'conv3_3' name: 'relu3_3' type: RELU }\nlayers { bottom: 'conv3_3' top: 'pool3' name: 'pool3' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool3' top: 'conv4_1' name: 'conv4_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv4_1' top: 'conv4_1' name: 'relu4_1' type: RELU }\nlayers { bottom: 'conv4_1' top: 'conv4_2' name: 'conv4_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv4_2' top: 'conv4_2' name: 'relu4_2' type: RELU }\nlayers { bottom: 'conv4_2' top: 'conv4_3' name: 'conv4_3' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv4_3' top: 'conv4_3' name: 'relu4_3' type: RELU }\nlayers { bottom: 'conv4_3' top: 'pool4' name: 'pool4' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool4' top: 'conv5_1' name: 'conv5_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv5_1' top: 'conv5_1' name: 'relu5_1' type: RELU }\nlayers { bottom: 'conv5_1' top: 'conv5_2' name: 'conv5_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv5_2' top: 'conv5_2' name: 'relu5_2' type: RELU }\nlayers { bottom: 'conv5_2' top: 'conv5_3' name: 'conv5_3' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv5_3' top: 'conv5_3' name: 'relu5_3' type: RELU }\nlayers { bottom: 'conv5_3' top: 'pool5' name: 'pool5' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool5' top: 'fc6' name: 'fc6' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE kernel_size: 7 num_output: 4096 } }\nlayers { bottom: 'fc6' top: 'fc6' name: 'relu6' type: RELU }\nlayers { bottom: 'fc6' top: 'fc6' name: 'drop6' type: DROPOUT\n dropout_param { dropout_ratio: 0.5 } }\nlayers { bottom: 'fc6' top: 'fc7' name: 'fc7' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE kernel_size: 1 num_output: 4096 } }\nlayers { bottom: 'fc7' top: 'fc7' name: 'relu7' type: RELU }\nlayers { bottom: 'fc7' top: 'fc7' name: 'drop7' type: DROPOUT\n dropout_param { dropout_ratio: 0.5 } }\nlayers { name: 'score' type: CONVOLUTION bottom: 'fc7' top: 'score'\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 21 kernel_size: 1 } }\nlayers { type: DECONVOLUTION name: 'score2' bottom: 'score' top: 'score2'\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { kernel_size: 4 stride: 2 num_output: 21 } }\n \n\nlayers { name: 'score-pool4' type: CONVOLUTION bottom: 'pool4' top: 'score-pool4'\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 21 kernel_size: 1 } }\n\nlayers { type: CROP name: 'score-pool4c' bottom: 'score-pool4' top: 'score-pool44c' bottom: 'score2' }\n \nlayers { type: ELTWISE name: 'score-fuse' bottom: 'score2' bottom: 'score-pool4c' top: 'score-fuse'\neltwise_param { operation: SUM } }\n\nlayers { type: DECONVOLUTION name: 'bigscore'\n bottom: 'score-fuse' top: 'bigscore'\n blobs_lr: 0 blobs_lr: 0\n convolution_param { num_output: 21 kernel_size: 32 stride: 16 } }\n\nlayers { type: CROP name: 'upscore' bottom: 'bigscore' bottom: 'data' top: 'upscore' }\n\nlayers { type: SOFTMAX name: 'output' bottom: 'upscore'}" }, { "path": "presets/fcn-8s-pascal.prototxt", "content": "name: 'FCN'\n\ninput: 'data'\ninput_dim: 1\ninput_dim: 3\ninput_dim: 500\ninput_dim: 500\nforce_backward: true\n\nlayers { bottom: 'data' top: 'conv1_1' name: 'conv1_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 64 pad: 100 kernel_size: 3 } }\nlayers { bottom: 'conv1_1' top: 'conv1_1' name: 'relu1_1' type: RELU }\nlayers { bottom: 'conv1_1' top: 'conv1_2' name: 'conv1_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 64 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv1_2' top: 'conv1_2' name: 'relu1_2' type: RELU }\nlayers { name: 'pool1' bottom: 'conv1_2' top: 'pool1' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { name: 'conv2_1' bottom: 'pool1' top: 'conv2_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 128 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv2_1' top: 'conv2_1' name: 'relu2_1' type: RELU }\nlayers { bottom: 'conv2_1' top: 'conv2_2' name: 'conv2_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 128 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv2_2' top: 'conv2_2' name: 'relu2_2' type: RELU }\nlayers { bottom: 'conv2_2' top: 'pool2' name: 'pool2' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool2' top: 'conv3_1' name: 'conv3_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv3_1' top: 'conv3_1' name: 'relu3_1' type: RELU }\nlayers { bottom: 'conv3_1' top: 'conv3_2' name: 'conv3_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv3_2' top: 'conv3_2' name: 'relu3_2' type: RELU }\nlayers { bottom: 'conv3_2' top: 'conv3_3' name: 'conv3_3' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 256 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv3_3' top: 'conv3_3' name: 'relu3_3' type: RELU }\nlayers { bottom: 'conv3_3' top: 'pool3' name: 'pool3' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool3' top: 'conv4_1' name: 'conv4_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv4_1' top: 'conv4_1' name: 'relu4_1' type: RELU }\nlayers { bottom: 'conv4_1' top: 'conv4_2' name: 'conv4_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv4_2' top: 'conv4_2' name: 'relu4_2' type: RELU }\nlayers { bottom: 'conv4_2' top: 'conv4_3' name: 'conv4_3' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv4_3' top: 'conv4_3' name: 'relu4_3' type: RELU }\nlayers { bottom: 'conv4_3' top: 'pool4' name: 'pool4' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool4' top: 'conv5_1' name: 'conv5_1' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv5_1' top: 'conv5_1' name: 'relu5_1' type: RELU }\nlayers { bottom: 'conv5_1' top: 'conv5_2' name: 'conv5_2' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv5_2' top: 'conv5_2' name: 'relu5_2' type: RELU }\nlayers { bottom: 'conv5_2' top: 'conv5_3' name: 'conv5_3' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 512 pad: 1 kernel_size: 3 } }\nlayers { bottom: 'conv5_3' top: 'conv5_3' name: 'relu5_3' type: RELU }\nlayers { bottom: 'conv5_3' top: 'pool5' name: 'pool5' type: POOLING\n pooling_param { pool: MAX kernel_size: 2 stride: 2 } }\nlayers { bottom: 'pool5' top: 'fc6' name: 'fc6' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE kernel_size: 7 num_output: 4096 } }\nlayers { bottom: 'fc6' top: 'fc6' name: 'relu6' type: RELU }\nlayers { bottom: 'fc6' top: 'fc6' name: 'drop6' type: DROPOUT\n dropout_param { dropout_ratio: 0.5 } }\nlayers { bottom: 'fc6' top: 'fc7' name: 'fc7' type: CONVOLUTION\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE kernel_size: 1 num_output: 4096 } }\nlayers { bottom: 'fc7' top: 'fc7' name: 'relu7' type: RELU }\nlayers { bottom: 'fc7' top: 'fc7' name: 'drop7' type: DROPOUT\n dropout_param { dropout_ratio: 0.5 } }\nlayers { name: 'score-fr' type: CONVOLUTION bottom: 'fc7' top: 'score'\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 21 kernel_size: 1 } }\n\nlayers { type: DECONVOLUTION name: 'score2' bottom: 'score' top: 'score2'\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { kernel_size: 4 stride: 2 num_output: 21 } }\n\nlayers { name: 'score-pool4' type: CONVOLUTION bottom: 'pool4' top: 'score-pool4'\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 21 kernel_size: 1 } }\n\nlayers { type: CROP name: 'crop' bottom: 'score-pool4' bottom: 'score2'\n top: 'score-pool4c' }\n\nlayers { type: ELTWISE name: 'fuse' bottom: 'score2' bottom: 'score-pool4c'\n top: 'score-fused'\n eltwise_param { operation: SUM } }\n\nlayers { type: DECONVOLUTION name: 'score4' bottom: 'score-fused'\n top: 'score4'\n blobs_lr: 1 weight_decay: 1\n convolution_param { bias_term: false kernel_size: 4 stride: 2 num_output: 21 } }\n\nlayers { name: 'score-pool3' type: CONVOLUTION bottom: 'pool3' top: 'score-pool3'\n blobs_lr: 1 blobs_lr: 2 weight_decay: 1 weight_decay: 0\n convolution_param { engine: CAFFE num_output: 21 kernel_size: 1 } }\n\nlayers { type: CROP name: 'crop' bottom: 'score-pool3' bottom: 'score4'\n top: 'score-pool3c' }\n\nlayers { type: ELTWISE name: 'fuse' bottom: 'score4' bottom: 'score-pool3c'\n top: 'score-final'\n eltwise_param { operation: SUM } }\n\nlayers { type: DECONVOLUTION name: 'upsample'\n bottom: 'score-final' top: 'bigscore'\n blobs_lr: 0\n convolution_param { bias_term: false num_output: 21 kernel_size: 16 stride: 8 } }\n\nlayers { type: CROP name: 'crop' bottom: 'bigscore' bottom: 'data' top: 'upscore' }\n" }, { "path": "presets/googlenet.prototxt", "content": "name: \"GoogleNet\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n input_param {\n shape: {\n dim: 10\n dim: 3\n dim: 227\n dim: 227\n }\n }\n}\nlayer {\n name: \"conv1/7x7_s2\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1/7x7_s2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 3\n kernel_size: 7\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv1/relu_7x7\"\n type: \"ReLU\"\n bottom: \"conv1/7x7_s2\"\n top: \"conv1/7x7_s2\"\n}\nlayer {\n name: \"pool1/3x3_s2\"\n type: \"Pooling\"\n bottom: \"conv1/7x7_s2\"\n top: \"pool1/3x3_s2\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"pool1/norm1\"\n type: \"LRN\"\n bottom: \"pool1/3x3_s2\"\n top: \"pool1/norm1\"\n lrn_param {\n local_size: 5\n alpha: 0.0001\n beta: 0.75\n }\n}\nlayer {\n name: \"conv2/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"pool1/norm1\"\n top: \"conv2/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv2/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"conv2/3x3_reduce\"\n top: \"conv2/3x3_reduce\"\n}\nlayer {\n name: \"conv2/3x3\"\n type: \"Convolution\"\n bottom: \"conv2/3x3_reduce\"\n top: \"conv2/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv2/relu_3x3\"\n type: \"ReLU\"\n bottom: \"conv2/3x3\"\n top: \"conv2/3x3\"\n}\nlayer {\n name: \"conv2/norm2\"\n type: \"LRN\"\n bottom: \"conv2/3x3\"\n top: \"conv2/norm2\"\n lrn_param {\n local_size: 5\n alpha: 0.0001\n beta: 0.75\n }\n}\nlayer {\n name: \"pool2/3x3_s2\"\n type: \"Pooling\"\n bottom: \"conv2/norm2\"\n top: \"pool2/3x3_s2\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"inception_3a/1x1\"\n type: \"Convolution\"\n bottom: \"pool2/3x3_s2\"\n top: \"inception_3a/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3a/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_3a/1x1\"\n top: \"inception_3a/1x1\"\n}\nlayer {\n name: \"inception_3a/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"pool2/3x3_s2\"\n top: \"inception_3a/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3a/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_3a/3x3_reduce\"\n top: \"inception_3a/3x3_reduce\"\n}\nlayer {\n name: \"inception_3a/3x3\"\n type: \"Convolution\"\n bottom: \"inception_3a/3x3_reduce\"\n top: \"inception_3a/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3a/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_3a/3x3\"\n top: \"inception_3a/3x3\"\n}\nlayer {\n name: \"inception_3a/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"pool2/3x3_s2\"\n top: \"inception_3a/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3a/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_3a/5x5_reduce\"\n top: \"inception_3a/5x5_reduce\"\n}\nlayer {\n name: \"inception_3a/5x5\"\n type: \"Convolution\"\n bottom: \"inception_3a/5x5_reduce\"\n top: \"inception_3a/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3a/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_3a/5x5\"\n top: \"inception_3a/5x5\"\n}\nlayer {\n name: \"inception_3a/pool\"\n type: \"Pooling\"\n bottom: \"pool2/3x3_s2\"\n top: \"inception_3a/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_3a/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_3a/pool\"\n top: \"inception_3a/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3a/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_3a/pool_proj\"\n top: \"inception_3a/pool_proj\"\n}\nlayer {\n name: \"inception_3a/output\"\n type: \"Concat\"\n bottom: \"inception_3a/1x1\"\n bottom: \"inception_3a/3x3\"\n bottom: \"inception_3a/5x5\"\n bottom: \"inception_3a/pool_proj\"\n top: \"inception_3a/output\"\n}\nlayer {\n name: \"inception_3b/1x1\"\n type: \"Convolution\"\n bottom: \"inception_3a/output\"\n top: \"inception_3b/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3b/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_3b/1x1\"\n top: \"inception_3b/1x1\"\n}\nlayer {\n name: \"inception_3b/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_3a/output\"\n top: \"inception_3b/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3b/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_3b/3x3_reduce\"\n top: \"inception_3b/3x3_reduce\"\n}\nlayer {\n name: \"inception_3b/3x3\"\n type: \"Convolution\"\n bottom: \"inception_3b/3x3_reduce\"\n top: \"inception_3b/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3b/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_3b/3x3\"\n top: \"inception_3b/3x3\"\n}\nlayer {\n name: \"inception_3b/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"inception_3a/output\"\n top: \"inception_3b/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3b/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_3b/5x5_reduce\"\n top: \"inception_3b/5x5_reduce\"\n}\nlayer {\n name: \"inception_3b/5x5\"\n type: \"Convolution\"\n bottom: \"inception_3b/5x5_reduce\"\n top: \"inception_3b/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3b/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_3b/5x5\"\n top: \"inception_3b/5x5\"\n}\nlayer {\n name: \"inception_3b/pool\"\n type: \"Pooling\"\n bottom: \"inception_3a/output\"\n top: \"inception_3b/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_3b/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_3b/pool\"\n top: \"inception_3b/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_3b/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_3b/pool_proj\"\n top: \"inception_3b/pool_proj\"\n}\nlayer {\n name: \"inception_3b/output\"\n type: \"Concat\"\n bottom: \"inception_3b/1x1\"\n bottom: \"inception_3b/3x3\"\n bottom: \"inception_3b/5x5\"\n bottom: \"inception_3b/pool_proj\"\n top: \"inception_3b/output\"\n}\nlayer {\n name: \"pool3/3x3_s2\"\n type: \"Pooling\"\n bottom: \"inception_3b/output\"\n top: \"pool3/3x3_s2\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"inception_4a/1x1\"\n type: \"Convolution\"\n bottom: \"pool3/3x3_s2\"\n top: \"inception_4a/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4a/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_4a/1x1\"\n top: \"inception_4a/1x1\"\n}\nlayer {\n name: \"inception_4a/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"pool3/3x3_s2\"\n top: \"inception_4a/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4a/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4a/3x3_reduce\"\n top: \"inception_4a/3x3_reduce\"\n}\nlayer {\n name: \"inception_4a/3x3\"\n type: \"Convolution\"\n bottom: \"inception_4a/3x3_reduce\"\n top: \"inception_4a/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 208\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4a/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_4a/3x3\"\n top: \"inception_4a/3x3\"\n}\nlayer {\n name: \"inception_4a/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"pool3/3x3_s2\"\n top: \"inception_4a/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4a/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4a/5x5_reduce\"\n top: \"inception_4a/5x5_reduce\"\n}\nlayer {\n name: \"inception_4a/5x5\"\n type: \"Convolution\"\n bottom: \"inception_4a/5x5_reduce\"\n top: \"inception_4a/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 48\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4a/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_4a/5x5\"\n top: \"inception_4a/5x5\"\n}\nlayer {\n name: \"inception_4a/pool\"\n type: \"Pooling\"\n bottom: \"pool3/3x3_s2\"\n top: \"inception_4a/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_4a/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_4a/pool\"\n top: \"inception_4a/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4a/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_4a/pool_proj\"\n top: \"inception_4a/pool_proj\"\n}\nlayer {\n name: \"inception_4a/output\"\n type: \"Concat\"\n bottom: \"inception_4a/1x1\"\n bottom: \"inception_4a/3x3\"\n bottom: \"inception_4a/5x5\"\n bottom: \"inception_4a/pool_proj\"\n top: \"inception_4a/output\"\n}\nlayer {\n name: \"inception_4b/1x1\"\n type: \"Convolution\"\n bottom: \"inception_4a/output\"\n top: \"inception_4b/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4b/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_4b/1x1\"\n top: \"inception_4b/1x1\"\n}\nlayer {\n name: \"inception_4b/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4a/output\"\n top: \"inception_4b/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 112\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4b/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4b/3x3_reduce\"\n top: \"inception_4b/3x3_reduce\"\n}\nlayer {\n name: \"inception_4b/3x3\"\n type: \"Convolution\"\n bottom: \"inception_4b/3x3_reduce\"\n top: \"inception_4b/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4b/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_4b/3x3\"\n top: \"inception_4b/3x3\"\n}\nlayer {\n name: \"inception_4b/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4a/output\"\n top: \"inception_4b/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 24\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4b/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4b/5x5_reduce\"\n top: \"inception_4b/5x5_reduce\"\n}\nlayer {\n name: \"inception_4b/5x5\"\n type: \"Convolution\"\n bottom: \"inception_4b/5x5_reduce\"\n top: \"inception_4b/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4b/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_4b/5x5\"\n top: \"inception_4b/5x5\"\n}\nlayer {\n name: \"inception_4b/pool\"\n type: \"Pooling\"\n bottom: \"inception_4a/output\"\n top: \"inception_4b/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_4b/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_4b/pool\"\n top: \"inception_4b/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4b/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_4b/pool_proj\"\n top: \"inception_4b/pool_proj\"\n}\nlayer {\n name: \"inception_4b/output\"\n type: \"Concat\"\n bottom: \"inception_4b/1x1\"\n bottom: \"inception_4b/3x3\"\n bottom: \"inception_4b/5x5\"\n bottom: \"inception_4b/pool_proj\"\n top: \"inception_4b/output\"\n}\nlayer {\n name: \"inception_4c/1x1\"\n type: \"Convolution\"\n bottom: \"inception_4b/output\"\n top: \"inception_4c/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4c/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_4c/1x1\"\n top: \"inception_4c/1x1\"\n}\nlayer {\n name: \"inception_4c/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4b/output\"\n top: \"inception_4c/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4c/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4c/3x3_reduce\"\n top: \"inception_4c/3x3_reduce\"\n}\nlayer {\n name: \"inception_4c/3x3\"\n type: \"Convolution\"\n bottom: \"inception_4c/3x3_reduce\"\n top: \"inception_4c/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4c/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_4c/3x3\"\n top: \"inception_4c/3x3\"\n}\nlayer {\n name: \"inception_4c/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4b/output\"\n top: \"inception_4c/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 24\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4c/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4c/5x5_reduce\"\n top: \"inception_4c/5x5_reduce\"\n}\nlayer {\n name: \"inception_4c/5x5\"\n type: \"Convolution\"\n bottom: \"inception_4c/5x5_reduce\"\n top: \"inception_4c/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4c/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_4c/5x5\"\n top: \"inception_4c/5x5\"\n}\nlayer {\n name: \"inception_4c/pool\"\n type: \"Pooling\"\n bottom: \"inception_4b/output\"\n top: \"inception_4c/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_4c/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_4c/pool\"\n top: \"inception_4c/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4c/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_4c/pool_proj\"\n top: \"inception_4c/pool_proj\"\n}\nlayer {\n name: \"inception_4c/output\"\n type: \"Concat\"\n bottom: \"inception_4c/1x1\"\n bottom: \"inception_4c/3x3\"\n bottom: \"inception_4c/5x5\"\n bottom: \"inception_4c/pool_proj\"\n top: \"inception_4c/output\"\n}\nlayer {\n name: \"inception_4d/1x1\"\n type: \"Convolution\"\n bottom: \"inception_4c/output\"\n top: \"inception_4d/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 112\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4d/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_4d/1x1\"\n top: \"inception_4d/1x1\"\n}\nlayer {\n name: \"inception_4d/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4c/output\"\n top: \"inception_4d/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 144\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4d/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4d/3x3_reduce\"\n top: \"inception_4d/3x3_reduce\"\n}\nlayer {\n name: \"inception_4d/3x3\"\n type: \"Convolution\"\n bottom: \"inception_4d/3x3_reduce\"\n top: \"inception_4d/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 288\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4d/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_4d/3x3\"\n top: \"inception_4d/3x3\"\n}\nlayer {\n name: \"inception_4d/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4c/output\"\n top: \"inception_4d/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4d/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4d/5x5_reduce\"\n top: \"inception_4d/5x5_reduce\"\n}\nlayer {\n name: \"inception_4d/5x5\"\n type: \"Convolution\"\n bottom: \"inception_4d/5x5_reduce\"\n top: \"inception_4d/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4d/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_4d/5x5\"\n top: \"inception_4d/5x5\"\n}\nlayer {\n name: \"inception_4d/pool\"\n type: \"Pooling\"\n bottom: \"inception_4c/output\"\n top: \"inception_4d/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_4d/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_4d/pool\"\n top: \"inception_4d/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4d/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_4d/pool_proj\"\n top: \"inception_4d/pool_proj\"\n}\nlayer {\n name: \"inception_4d/output\"\n type: \"Concat\"\n bottom: \"inception_4d/1x1\"\n bottom: \"inception_4d/3x3\"\n bottom: \"inception_4d/5x5\"\n bottom: \"inception_4d/pool_proj\"\n top: \"inception_4d/output\"\n}\nlayer {\n name: \"inception_4e/1x1\"\n type: \"Convolution\"\n bottom: \"inception_4d/output\"\n top: \"inception_4e/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4e/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_4e/1x1\"\n top: \"inception_4e/1x1\"\n}\nlayer {\n name: \"inception_4e/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4d/output\"\n top: \"inception_4e/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4e/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4e/3x3_reduce\"\n top: \"inception_4e/3x3_reduce\"\n}\nlayer {\n name: \"inception_4e/3x3\"\n type: \"Convolution\"\n bottom: \"inception_4e/3x3_reduce\"\n top: \"inception_4e/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 320\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4e/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_4e/3x3\"\n top: \"inception_4e/3x3\"\n}\nlayer {\n name: \"inception_4e/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"inception_4d/output\"\n top: \"inception_4e/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4e/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_4e/5x5_reduce\"\n top: \"inception_4e/5x5_reduce\"\n}\nlayer {\n name: \"inception_4e/5x5\"\n type: \"Convolution\"\n bottom: \"inception_4e/5x5_reduce\"\n top: \"inception_4e/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4e/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_4e/5x5\"\n top: \"inception_4e/5x5\"\n}\nlayer {\n name: \"inception_4e/pool\"\n type: \"Pooling\"\n bottom: \"inception_4d/output\"\n top: \"inception_4e/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_4e/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_4e/pool\"\n top: \"inception_4e/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_4e/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_4e/pool_proj\"\n top: \"inception_4e/pool_proj\"\n}\nlayer {\n name: \"inception_4e/output\"\n type: \"Concat\"\n bottom: \"inception_4e/1x1\"\n bottom: \"inception_4e/3x3\"\n bottom: \"inception_4e/5x5\"\n bottom: \"inception_4e/pool_proj\"\n top: \"inception_4e/output\"\n}\nlayer {\n name: \"pool4/3x3_s2\"\n type: \"Pooling\"\n bottom: \"inception_4e/output\"\n top: \"pool4/3x3_s2\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"inception_5a/1x1\"\n type: \"Convolution\"\n bottom: \"pool4/3x3_s2\"\n top: \"inception_5a/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5a/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_5a/1x1\"\n top: \"inception_5a/1x1\"\n}\nlayer {\n name: \"inception_5a/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"pool4/3x3_s2\"\n top: \"inception_5a/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5a/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_5a/3x3_reduce\"\n top: \"inception_5a/3x3_reduce\"\n}\nlayer {\n name: \"inception_5a/3x3\"\n type: \"Convolution\"\n bottom: \"inception_5a/3x3_reduce\"\n top: \"inception_5a/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 320\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5a/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_5a/3x3\"\n top: \"inception_5a/3x3\"\n}\nlayer {\n name: \"inception_5a/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"pool4/3x3_s2\"\n top: \"inception_5a/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5a/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_5a/5x5_reduce\"\n top: \"inception_5a/5x5_reduce\"\n}\nlayer {\n name: \"inception_5a/5x5\"\n type: \"Convolution\"\n bottom: \"inception_5a/5x5_reduce\"\n top: \"inception_5a/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5a/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_5a/5x5\"\n top: \"inception_5a/5x5\"\n}\nlayer {\n name: \"inception_5a/pool\"\n type: \"Pooling\"\n bottom: \"pool4/3x3_s2\"\n top: \"inception_5a/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_5a/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_5a/pool\"\n top: \"inception_5a/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5a/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_5a/pool_proj\"\n top: \"inception_5a/pool_proj\"\n}\nlayer {\n name: \"inception_5a/output\"\n type: \"Concat\"\n bottom: \"inception_5a/1x1\"\n bottom: \"inception_5a/3x3\"\n bottom: \"inception_5a/5x5\"\n bottom: \"inception_5a/pool_proj\"\n top: \"inception_5a/output\"\n}\nlayer {\n name: \"inception_5b/1x1\"\n type: \"Convolution\"\n bottom: \"inception_5a/output\"\n top: \"inception_5b/1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5b/relu_1x1\"\n type: \"ReLU\"\n bottom: \"inception_5b/1x1\"\n top: \"inception_5b/1x1\"\n}\nlayer {\n name: \"inception_5b/3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_5a/output\"\n top: \"inception_5b/3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.09\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5b/relu_3x3_reduce\"\n type: \"ReLU\"\n bottom: \"inception_5b/3x3_reduce\"\n top: \"inception_5b/3x3_reduce\"\n}\nlayer {\n name: \"inception_5b/3x3\"\n type: \"Convolution\"\n bottom: \"inception_5b/3x3_reduce\"\n top: \"inception_5b/3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5b/relu_3x3\"\n type: \"ReLU\"\n bottom: \"inception_5b/3x3\"\n top: \"inception_5b/3x3\"\n}\nlayer {\n name: \"inception_5b/5x5_reduce\"\n type: \"Convolution\"\n bottom: \"inception_5a/output\"\n top: \"inception_5b/5x5_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 48\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.2\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5b/relu_5x5_reduce\"\n type: \"ReLU\"\n bottom: \"inception_5b/5x5_reduce\"\n top: \"inception_5b/5x5_reduce\"\n}\nlayer {\n name: \"inception_5b/5x5\"\n type: \"Convolution\"\n bottom: \"inception_5b/5x5_reduce\"\n top: \"inception_5b/5x5\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 2\n kernel_size: 5\n weight_filler {\n type: \"xavier\"\n std: 0.03\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5b/relu_5x5\"\n type: \"ReLU\"\n bottom: \"inception_5b/5x5\"\n top: \"inception_5b/5x5\"\n}\nlayer {\n name: \"inception_5b/pool\"\n type: \"Pooling\"\n bottom: \"inception_5a/output\"\n top: \"inception_5b/pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_5b/pool_proj\"\n type: \"Convolution\"\n bottom: \"inception_5b/pool\"\n top: \"inception_5b/pool_proj\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n std: 0.1\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_5b/relu_pool_proj\"\n type: \"ReLU\"\n bottom: \"inception_5b/pool_proj\"\n top: \"inception_5b/pool_proj\"\n}\nlayer {\n name: \"inception_5b/output\"\n type: \"Concat\"\n bottom: \"inception_5b/1x1\"\n bottom: \"inception_5b/3x3\"\n bottom: \"inception_5b/5x5\"\n bottom: \"inception_5b/pool_proj\"\n top: \"inception_5b/output\"\n}\nlayer {\n name: \"pool5/7x7_s1\"\n type: \"Pooling\"\n bottom: \"inception_5b/output\"\n top: \"pool5/7x7_s1\"\n pooling_param {\n pool: AVE\n kernel_size: 7\n stride: 1\n }\n}\nlayer {\n name: \"pool5/drop_7x7_s1\"\n type: \"Dropout\"\n bottom: \"pool5/7x7_s1\"\n top: \"pool5/7x7_s1\"\n dropout_param {\n dropout_ratio: 0.4\n }\n}\nlayer {\n name: \"loss3/classifier\"\n type: \"InnerProduct\"\n bottom: \"pool5/7x7_s1\"\n top: \"loss3/classifier\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 1000\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"prob\"\n type: \"Softmax\"\n bottom: \"loss3/classifier\"\n top: \"prob\"\n}\n" }, { "path": "presets/inceptionv3.prototxt", "content": "name: \"Inception_v3\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n include {\n phase: TRAIN\n }\n transform_param {\n mirror: true\n crop_size: 299\n mean_value: 104\n mean_value: 117\n mean_value: 123\n }\n data_param {\n source: \"/mnt/disk/ILSVRC2012/300px_ilsvrc12_train_lmdb\"\n batch_size: 20\n backend: LMDB\n }\n}\nlayer {\n name: \"conv_conv2d\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_conv2d\"\n top: \"conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_conv2d_bn\"\n top: \"conv_conv2d_bn\"\n}\nlayer {\n name: \"conv_1_1/conv2d\"\n type: \"Convolution\"\n bottom: \"conv_conv2d_bn\"\n top: \"conv_1_1/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_1_1/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_1_1/conv2d\"\n top: \"conv_1_1/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_1_1/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_1_1/conv2d_bn\"\n top: \"conv_1_1/conv2d_bn\"\n}\nlayer {\n name: \"conv_2_2/conv2d\"\n type: \"Convolution\"\n bottom: \"conv_1_1/conv2d_bn\"\n top: \"conv_2_2/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_2_2/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_2_2/conv2d\"\n top: \"conv_2_2/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_2_2/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_2_2/conv2d_bn\"\n top: \"conv_2_2/conv2d_bn\"\n}\nlayer {\n name: \"pool\"\n type: \"Pooling\"\n bottom: \"conv_2_2/conv2d_bn\"\n top: \"pool\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"conv_3_3/conv2d\"\n type: \"Convolution\"\n bottom: \"pool\"\n top: \"conv_3_3/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 80\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_3_3/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_3_3/conv2d\"\n top: \"conv_3_3/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_3_3/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_3_3/conv2d_bn\"\n top: \"conv_3_3/conv2d_bn\"\n}\nlayer {\n name: \"conv_4_4/conv2d\"\n type: \"Convolution\"\n bottom: \"conv_3_3/conv2d_bn\"\n top: \"conv_4_4/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_4_4/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_4_4/conv2d\"\n top: \"conv_4_4/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_4_4/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_4_4/conv2d_bn\"\n top: \"conv_4_4/conv2d_bn\"\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv_4_4/conv2d_bn\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_conv/conv2d\"\n top: \"mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_conv/conv2d_bn\"\n top: \"mixed_conv/conv2d_bn\"\n}\nlayer {\n name: \"mixed_tower/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"mixed_tower/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 48\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/conv_conv2d\"\n top: \"mixed_tower/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/conv_conv2d_bn\"\n top: \"mixed_tower/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_tower/conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/conv_conv2d_bn\"\n top: \"mixed_tower/conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 2\n kernel_size: 5\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/conv_1_conv2d\"\n top: \"mixed_tower/conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/conv_1_conv2d_bn\"\n top: \"mixed_tower/conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_tower/1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"mixed_tower/1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/1_conv_conv2d\"\n top: \"mixed_tower/1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/1_conv_conv2d_bn\"\n top: \"mixed_tower/1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_tower/1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/1_conv_conv2d_bn\"\n top: \"mixed_tower/1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/1_conv_1_conv2d\"\n top: \"mixed_tower/1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/1_conv_1_conv2d_bn\"\n top: \"mixed_tower/1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_tower/1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/1_conv_1_conv2d_bn\"\n top: \"mixed_tower/1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/1_conv_2_conv2d\"\n top: \"mixed_tower/1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/1_conv_2_conv2d_bn\"\n top: \"mixed_tower/1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_tower/AVG_pool\"\n type: \"Pooling\"\n bottom: \"pool1\"\n top: \"mixed_tower/AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_tower/2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/AVG_pool\"\n top: \"mixed_tower/2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/2_conv_conv2d\"\n top: \"mixed_tower/2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/2_conv_conv2d_bn\"\n top: \"mixed_tower/2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_tower/chconcat\"\n bottom: \"mixed_conv/conv2d_bn\"\n bottom: \"mixed_tower/conv_1_conv2d_bn\"\n bottom: \"mixed_tower/1_conv_2_conv2d_bn\"\n bottom: \"mixed_tower/2_conv_conv2d_bn\"\n top: \"mixed_tower/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_1/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/conv_conv2d\"\n top: \"mixed_1/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/conv_conv2d_bn\"\n top: \"mixed_1/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_1/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 48\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_conv_conv2d\"\n top: \"mixed_1/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_conv_conv2d_bn\"\n top: \"mixed_1/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_1/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_conv_conv2d_bn\"\n top: \"mixed_1/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 2\n kernel_size: 5\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_conv_1_conv2d\"\n top: \"mixed_1/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_conv_1_conv2d_bn\"\n top: \"mixed_1/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_1/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_1_conv_conv2d\"\n top: \"mixed_1/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_1_conv_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_1/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_1_conv_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_1_conv_1_conv2d\"\n top: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_1/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_1_conv_2_conv2d\"\n top: \"mixed_1/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_1/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_1/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_2_AVG_pool\"\n top: \"mixed_1/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_2_conv_conv2d\"\n top: \"mixed_1/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_2_conv_conv2d_bn\"\n top: \"mixed_1/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_1/chconcat\"\n bottom: \"mixed_1/conv_conv2d_bn\"\nbottom: \"mixed_1/tower_conv_1_conv2d_bn\"\nbottom: \"mixed_1/tower_1_conv_2_conv2d_bn\"\nbottom: \"mixed_1/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_1/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_2/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/conv_conv2d\"\n top: \"mixed_2/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/conv_conv2d_bn\"\n top: \"mixed_2/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_2/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 48\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_conv_conv2d\"\n top: \"mixed_2/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_conv_conv2d_bn\"\n top: \"mixed_2/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_2/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_conv_conv2d_bn\"\n top: \"mixed_2/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 2\n kernel_size: 5\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_conv_1_conv2d\"\n top: \"mixed_2/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_conv_1_conv2d_bn\"\n top: \"mixed_2/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_2/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_1_conv_conv2d\"\n top: \"mixed_2/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_1_conv_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_2/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_1_conv_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_1_conv_1_conv2d\"\n top: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_2/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_1_conv_2_conv2d\"\n top: \"mixed_2/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_2/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_2/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_2_AVG_pool\"\n top: \"mixed_2/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_2_conv_conv2d\"\n top: \"mixed_2/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_2_conv_conv2d_bn\"\n top: \"mixed_2/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_2/chconcat\"\n bottom: \"mixed_2/conv_conv2d_bn\"\nbottom: \"mixed_2/tower_conv_1_conv2d_bn\"\nbottom: \"mixed_2/tower_1_conv_2_conv2d_bn\"\nbottom: \"mixed_2/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_2/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_3/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/chconcat\"\n top: \"mixed_3/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/conv_conv2d\"\n top: \"mixed_3/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/conv_conv2d_bn\"\n top: \"mixed_3/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_3/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/chconcat\"\n top: \"mixed_3/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/tower_conv_conv2d\"\n top: \"mixed_3/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/tower_conv_conv2d_bn\"\n top: \"mixed_3/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_3/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/tower_conv_conv2d_bn\"\n top: \"mixed_3/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/tower_conv_1_conv2d\"\n top: \"mixed_3/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/tower_conv_1_conv2d_bn\"\n top: \"mixed_3/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_3/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/tower_conv_1_conv2d_bn\"\n top: \"mixed_3/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/tower_conv_2_conv2d\"\n top: \"mixed_3/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/tower_conv_2_conv2d_bn\"\n top: \"mixed_3/tower_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_3/max_pool\"\n type: \"Pooling\"\n bottom: \"mixed_2/chconcat\"\n top: \"mixed_3/max_pool\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"mixed_3/chconcat\"\n bottom: \"mixed_3/max_pool\"\nbottom: \"mixed_3/conv_conv2d_bn\"\nbottom: \"mixed_3/tower_conv_2_conv2d_bn\"\n\n top: \"mixed_3/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_4/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/conv_conv2d\"\n top: \"mixed_4/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/conv_conv2d_bn\"\n top: \"mixed_4/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_conv_conv2d\"\n top: \"mixed_4/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_conv_conv2d_bn\"\n top: \"mixed_4/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_conv_conv2d_bn\"\n top: \"mixed_4/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_conv_1_conv2d\"\n top: \"mixed_4/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_conv_1_conv2d_bn\"\n top: \"mixed_4/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_conv_1_conv2d_bn\"\n top: \"mixed_4/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_conv_2_conv2d\"\n top: \"mixed_4/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_conv_2_conv2d_bn\"\n top: \"mixed_4/tower_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_conv2d\"\n top: \"mixed_4/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_1_conv2d\"\n top: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_2_conv2d\"\n top: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_3_conv2d\"\n top: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_4_conv2d\"\n top: \"mixed_4/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_4_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_4/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_2_AVG_pool\"\n top: \"mixed_4/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_2_conv_conv2d\"\n top: \"mixed_4/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_2_conv_conv2d_bn\"\n top: \"mixed_4/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_4/chconcat\"\n bottom: \"mixed_4/conv_conv2d_bn\"\nbottom: \"mixed_4/tower_conv_2_conv2d_bn\"\nbottom: \"mixed_4/tower_1_conv_4_conv2d_bn\"\nbottom: \"mixed_4/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_4/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_5/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/conv_conv2d\"\n top: \"mixed_5/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/conv_conv2d_bn\"\n top: \"mixed_5/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_conv_conv2d\"\n top: \"mixed_5/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_conv_conv2d_bn\"\n top: \"mixed_5/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_conv_conv2d_bn\"\n top: \"mixed_5/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_conv_1_conv2d\"\n top: \"mixed_5/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_conv_1_conv2d_bn\"\n top: \"mixed_5/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_conv_1_conv2d_bn\"\n top: \"mixed_5/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_conv_2_conv2d\"\n top: \"mixed_5/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_conv_2_conv2d_bn\"\n top: \"mixed_5/tower_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_conv2d\"\n top: \"mixed_5/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_1_conv2d\"\n top: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_2_conv2d\"\n top: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_3_conv2d\"\n top: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_4_conv2d\"\n top: \"mixed_5/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_4_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_5/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_2_AVG_pool\"\n top: \"mixed_5/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_2_conv_conv2d\"\n top: \"mixed_5/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_2_conv_conv2d_bn\"\n top: \"mixed_5/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_5/chconcat\"\n bottom: \"mixed_5/conv_conv2d_bn\"\nbottom: \"mixed_5/tower_conv_2_conv2d_bn\"\nbottom: \"mixed_5/tower_1_conv_4_conv2d_bn\"\nbottom: \"mixed_5/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_5/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_6/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/conv_conv2d\"\n top: \"mixed_6/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/conv_conv2d_bn\"\n top: \"mixed_6/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_conv_conv2d\"\n top: \"mixed_6/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_conv_conv2d_bn\"\n top: \"mixed_6/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_conv_conv2d_bn\"\n top: \"mixed_6/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_conv_1_conv2d\"\n top: \"mixed_6/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_conv_1_conv2d_bn\"\n top: \"mixed_6/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_conv_1_conv2d_bn\"\n top: \"mixed_6/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_conv_2_conv2d\"\n top: \"mixed_6/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_conv_2_conv2d_bn\"\n top: \"mixed_6/tower_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_conv2d\"\n top: \"mixed_6/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_1_conv2d\"\n top: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_2_conv2d\"\n top: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_3_conv2d\"\n top: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_4_conv2d\"\n top: \"mixed_6/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_4_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_6/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_2_AVG_pool\"\n top: \"mixed_6/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_2_conv_conv2d\"\n top: \"mixed_6/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_2_conv_conv2d_bn\"\n top: \"mixed_6/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_6/chconcat\"\n bottom: \"mixed_6/conv_conv2d_bn\"\nbottom: \"mixed_6/tower_conv_2_conv2d_bn\"\nbottom: \"mixed_6/tower_1_conv_4_conv2d_bn\"\nbottom: \"mixed_6/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_6/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_7/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/conv_conv2d\"\n top: \"mixed_7/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/conv_conv2d_bn\"\n top: \"mixed_7/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_conv_conv2d\"\n top: \"mixed_7/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_conv_conv2d_bn\"\n top: \"mixed_7/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_conv_conv2d_bn\"\n top: \"mixed_7/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_conv_1_conv2d\"\n top: \"mixed_7/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_conv_1_conv2d_bn\"\n top: \"mixed_7/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_conv_1_conv2d_bn\"\n top: \"mixed_7/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_conv_2_conv2d\"\n top: \"mixed_7/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_conv_2_conv2d_bn\"\n top: \"mixed_7/tower_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_conv2d\"\n top: \"mixed_7/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_1_conv2d\"\n top: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_2_conv2d\"\n top: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_3_conv2d\"\n top: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_4_conv2d\"\n top: \"mixed_7/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_4_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_7/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_2_AVG_pool\"\n top: \"mixed_7/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_2_conv_conv2d\"\n top: \"mixed_7/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_2_conv_conv2d_bn\"\n top: \"mixed_7/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_7/chconcat\"\n bottom: \"mixed_7/conv_conv2d_bn\"\nbottom: \"mixed_7/tower_conv_2_conv2d_bn\"\nbottom: \"mixed_7/tower_1_conv_4_conv2d_bn\"\nbottom: \"mixed_7/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_7/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_8/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/chconcat\"\n top: \"mixed_8/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_conv_conv2d\"\n top: \"mixed_8/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_conv_conv2d_bn\"\n top: \"mixed_8/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_8/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_conv_conv2d_bn\"\n top: \"mixed_8/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 320\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_conv_1_conv2d\"\n top: \"mixed_8/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_conv_1_conv2d_bn\"\n top: \"mixed_8/tower_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/chconcat\"\n top: \"mixed_8/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_conv2d\"\n top: \"mixed_8/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_1_conv_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_1_conv2d\"\n top: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_2_conv2d\"\n top: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_3_conv2d\"\n top: \"mixed_8/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_3_conv2d_bn\"\n}\nlayer {\n name: \"mixed_8/max_pool\"\n type: \"Pooling\"\n bottom: \"mixed_7/chconcat\"\n top: \"mixed_8/max_pool\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"mixed_8/chconcat\"\n bottom: \"mixed_8/tower_conv_1_conv2d_bn\"\nbottom: \"mixed_8/tower_1_conv_3_conv2d_bn\"\nbottom: \"mixed_8/max_pool\"\n\n top: \"mixed_8/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_9/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 320\n pad: 0\n kernel_h: 1\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/conv_conv2d\"\n top: \"mixed_9/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/conv_conv2d_bn\"\n top: \"mixed_9/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_conv_conv2d\"\n top: \"mixed_9/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_conv_conv2d_bn\"\n top: \"mixed_9/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_conv_conv2d_bn\"\n top: \"mixed_9/tower_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_mixed_conv/conv2d\"\n top: \"mixed_9/tower_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_mixed_conv/conv2d_bn\"\n top: \"mixed_9/tower_mixed_conv/conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_conv_conv2d_bn\"\n top: \"mixed_9/tower_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_mixed_conv/1_conv2d\"\n top: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\n top: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 448\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_conv_conv2d\"\n top: \"mixed_9/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_conv_conv2d_bn\"\n top: \"mixed_9/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_1_conv_conv2d_bn\"\n top: \"mixed_9/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d\"\n top: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_9/tower_1_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_mixed_conv/conv2d\"\n top: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\n top: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_9/tower_1_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_mixed_conv/1_conv2d\"\n top: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\n top: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_9/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_2_AVG_pool\"\n top: \"mixed_9/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_2_conv_conv2d\"\n top: \"mixed_9/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_2_conv_conv2d_bn\"\n top: \"mixed_9/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_9/chconcat\"\n bottom: \"mixed_9/conv_conv2d_bn\"\nbottom: \"mixed_9/tower_mixed_conv/conv2d_bn\"\nbottom: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\nbottom: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\nbottom: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\nbottom: \"mixed_9/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_9/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_10/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 320\n pad: 0\n kernel_h: 1\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/conv_conv2d\"\n top: \"mixed_10/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/conv_conv2d_bn\"\n top: \"mixed_10/conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_conv_conv2d\"\n top: \"mixed_10/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_conv_conv2d_bn\"\n top: \"mixed_10/tower_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_conv_conv2d_bn\"\n top: \"mixed_10/tower_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_mixed_conv/conv2d\"\n top: \"mixed_10/tower_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_mixed_conv/conv2d_bn\"\n top: \"mixed_10/tower_mixed_conv/conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_conv_conv2d_bn\"\n top: \"mixed_10/tower_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_mixed_conv/1_conv2d\"\n top: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\n top: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 448\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_conv_conv2d\"\n top: \"mixed_10/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_conv_conv2d_bn\"\n top: \"mixed_10/tower_1_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_1_conv_conv2d_bn\"\n top: \"mixed_10/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d\"\n top: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_10/tower_1_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_mixed_conv/conv2d\"\n top: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\n top: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_10/tower_1_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_mixed_conv/1_conv2d\"\n top: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\n top: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/max_pool\"\n type: \"Pooling\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/max_pool\"\n pooling_param {\n pool: MAX\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_10/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/max_pool\"\n top: \"mixed_10/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_2_conv_conv2d\"\n top: \"mixed_10/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_2_conv_conv2d_bn\"\n top: \"mixed_10/tower_2_conv_conv2d_bn\"\n}\nlayer {\n name: \"mixed_10/chconcat\"\n bottom: \"mixed_10/conv_conv2d_bn\"\nbottom: \"mixed_10/tower_mixed_conv/conv2d_bn\"\nbottom: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\nbottom: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\nbottom: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\nbottom: \"mixed_10/tower_2_conv_conv2d_bn\"\n\n top: \"mixed_10/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"global_pool\"\n type: \"Pooling\"\n bottom: \"mixed_10/chconcat\"\n top: \"global_pool\"\n pooling_param {\n pool: AVE\n pad: 0\n kernel_size: 8\n stride: 1\n }\n}\nlayer {\n name: \"flatten\"\n type: \"Flatten\"\n bottom: \"global_pool\"\n top: \"flatten\"\n}\nlayer {\n name: \"fc1\"\n type: \"InnerProduct\"\n bottom: \"flatten\"\n top: \"fc1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 1000\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"fc1\"\n top: \"loss\"\n}\nlayer {\n name: \"acc/top-1\"\n type: \"Accuracy\"\n bottom: \"fc1\"\n top: \"acc/top-1\"\n include {\n phase: TEST\n }\n}\nlayer {\n name: \"acc/top-5\"\n type: \"Accuracy\"\n bottom: \"fc1\"\n top: \"acc/top-5\"\n include {\n phase: TEST\n }\n accuracy_param {\n top_k: 5\n }\n}" }, { "path": "presets/inceptionv3_orig.prototxt", "content": "name: \"Inception_v3\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n include {\n phase: TRAIN\n }\n transform_param {\n mirror: true\n crop_size: 299\n mean_value: 104\n mean_value: 117\n mean_value: 123\n }\n data_param {\n source: \"/mnt/disk/ILSVRC2012/300px_ilsvrc12_train_lmdb\"\n batch_size: 20\n backend: LMDB\n }\n}\nlayer {\n name: \"conv_conv2d\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_conv2d\"\n top: \"conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_conv2d_bn\"\n top: \"conv_conv2d_relu\"\n}\nlayer {\n name: \"conv_1_1/conv2d\"\n type: \"Convolution\"\n bottom: \"conv_conv2d_relu\"\n top: \"conv_1_1/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_1_1/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_1_1/conv2d\"\n top: \"conv_1_1/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_1_1/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_1_1/conv2d_bn\"\n top: \"conv_1_1/conv2d_relu\"\n}\nlayer {\n name: \"conv_2_2/conv2d\"\n type: \"Convolution\"\n bottom: \"conv_1_1/conv2d_relu\"\n top: \"conv_2_2/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_2_2/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_2_2/conv2d\"\n top: \"conv_2_2/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_2_2/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_2_2/conv2d_bn\"\n top: \"conv_2_2/conv2d_relu\"\n}\nlayer {\n name: \"pool\"\n type: \"Pooling\"\n bottom: \"conv_2_2/conv2d_relu\"\n top: \"pool\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"conv_3_3/conv2d\"\n type: \"Convolution\"\n bottom: \"pool\"\n top: \"conv_3_3/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 80\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_3_3/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_3_3/conv2d\"\n top: \"conv_3_3/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_3_3/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_3_3/conv2d_bn\"\n top: \"conv_3_3/conv2d_relu\"\n}\nlayer {\n name: \"conv_4_4/conv2d\"\n type: \"Convolution\"\n bottom: \"conv_3_3/conv2d_relu\"\n top: \"conv_4_4/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"conv_4_4/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"conv_4_4/conv2d\"\n top: \"conv_4_4/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"conv_4_4/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"conv_4_4/conv2d_bn\"\n top: \"conv_4_4/conv2d_relu\"\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv_4_4/conv2d_relu\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_conv/conv2d\"\n top: \"mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_conv/conv2d_bn\"\n top: \"mixed_conv/conv2d_relu\"\n}\nlayer {\n name: \"mixed_tower/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"mixed_tower/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 48\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/conv_conv2d\"\n top: \"mixed_tower/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/conv_conv2d_bn\"\n top: \"mixed_tower/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_tower/conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/conv_conv2d_relu\"\n top: \"mixed_tower/conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 2\n kernel_size: 5\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/conv_1_conv2d\"\n top: \"mixed_tower/conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/conv_1_conv2d_bn\"\n top: \"mixed_tower/conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_tower/1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"mixed_tower/1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/1_conv_conv2d\"\n top: \"mixed_tower/1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/1_conv_conv2d_bn\"\n top: \"mixed_tower/1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_tower/1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/1_conv_conv2d_relu\"\n top: \"mixed_tower/1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/1_conv_1_conv2d\"\n top: \"mixed_tower/1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/1_conv_1_conv2d_bn\"\n top: \"mixed_tower/1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_tower/1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/1_conv_1_conv2d_relu\"\n top: \"mixed_tower/1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/1_conv_2_conv2d\"\n top: \"mixed_tower/1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/1_conv_2_conv2d_bn\"\n top: \"mixed_tower/1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_tower/AVG_pool\"\n type: \"Pooling\"\n bottom: \"pool1\"\n top: \"mixed_tower/AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_tower/2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/AVG_pool\"\n top: \"mixed_tower/2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_tower/2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_tower/2_conv_conv2d\"\n top: \"mixed_tower/2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_tower/2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_tower/2_conv_conv2d_bn\"\n top: \"mixed_tower/2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_tower/chconcat\"\n bottom: \"mixed_conv/conv2d_relu\"\n bottom: \"mixed_tower/conv_1_conv2d_relu\"\n bottom: \"mixed_tower/1_conv_2_conv2d_relu\"\n bottom: \"mixed_tower/2_conv_conv2d_relu\"\n top: \"mixed_tower/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_1/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/conv_conv2d\"\n top: \"mixed_1/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/conv_conv2d_bn\"\n top: \"mixed_1/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_1/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 48\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_conv_conv2d\"\n top: \"mixed_1/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_conv_conv2d_bn\"\n top: \"mixed_1/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_1/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_conv_conv2d_relu\"\n top: \"mixed_1/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 2\n kernel_size: 5\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_conv_1_conv2d\"\n top: \"mixed_1/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_conv_1_conv2d_bn\"\n top: \"mixed_1/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_1/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_1_conv_conv2d\"\n top: \"mixed_1/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_1_conv_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_1/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_1_conv_conv2d_relu\"\n top: \"mixed_1/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_1_conv_1_conv2d\"\n top: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_1/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_1/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_1_conv_2_conv2d\"\n top: \"mixed_1/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_1/tower_1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_1/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_tower/chconcat\"\n top: \"mixed_1/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_1/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/tower_2_AVG_pool\"\n top: \"mixed_1/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_1/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_1/tower_2_conv_conv2d\"\n top: \"mixed_1/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_1/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_1/tower_2_conv_conv2d_bn\"\n top: \"mixed_1/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_1/chconcat\"\n bottom: \"mixed_1/conv_conv2d_relu\"\nbottom: \"mixed_1/tower_conv_1_conv2d_relu\"\nbottom: \"mixed_1/tower_1_conv_2_conv2d_relu\"\nbottom: \"mixed_1/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_1/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_2/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/conv_conv2d\"\n top: \"mixed_2/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/conv_conv2d_bn\"\n top: \"mixed_2/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_2/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 48\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_conv_conv2d\"\n top: \"mixed_2/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_conv_conv2d_bn\"\n top: \"mixed_2/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_2/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_conv_conv2d_relu\"\n top: \"mixed_2/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 2\n kernel_size: 5\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_conv_1_conv2d\"\n top: \"mixed_2/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_conv_1_conv2d_bn\"\n top: \"mixed_2/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_2/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_1_conv_conv2d\"\n top: \"mixed_2/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_1_conv_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_2/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_1_conv_conv2d_relu\"\n top: \"mixed_2/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_1_conv_1_conv2d\"\n top: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_2/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_2/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_1_conv_2_conv2d\"\n top: \"mixed_2/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_2/tower_1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_2/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_1/chconcat\"\n top: \"mixed_2/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_2/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/tower_2_AVG_pool\"\n top: \"mixed_2/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_2/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_2/tower_2_conv_conv2d\"\n top: \"mixed_2/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_2/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_2/tower_2_conv_conv2d_bn\"\n top: \"mixed_2/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_2/chconcat\"\n bottom: \"mixed_2/conv_conv2d_relu\"\nbottom: \"mixed_2/tower_conv_1_conv2d_relu\"\nbottom: \"mixed_2/tower_1_conv_2_conv2d_relu\"\nbottom: \"mixed_2/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_2/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_3/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/chconcat\"\n top: \"mixed_3/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/conv_conv2d\"\n top: \"mixed_3/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/conv_conv2d_bn\"\n top: \"mixed_3/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_3/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_2/chconcat\"\n top: \"mixed_3/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/tower_conv_conv2d\"\n top: \"mixed_3/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/tower_conv_conv2d_bn\"\n top: \"mixed_3/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_3/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/tower_conv_conv2d_relu\"\n top: \"mixed_3/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/tower_conv_1_conv2d\"\n top: \"mixed_3/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/tower_conv_1_conv2d_bn\"\n top: \"mixed_3/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_3/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/tower_conv_1_conv2d_relu\"\n top: \"mixed_3/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_3/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_3/tower_conv_2_conv2d\"\n top: \"mixed_3/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_3/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_3/tower_conv_2_conv2d_bn\"\n top: \"mixed_3/tower_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_3/max_pool\"\n type: \"Pooling\"\n bottom: \"mixed_2/chconcat\"\n top: \"mixed_3/max_pool\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"mixed_3/chconcat\"\n bottom: \"mixed_3/max_pool\"\nbottom: \"mixed_3/conv_conv2d_relu\"\nbottom: \"mixed_3/tower_conv_2_conv2d_relu\"\n\n top: \"mixed_3/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_4/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/conv_conv2d\"\n top: \"mixed_4/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/conv_conv2d_bn\"\n top: \"mixed_4/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_conv_conv2d\"\n top: \"mixed_4/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_conv_conv2d_bn\"\n top: \"mixed_4/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_conv_conv2d_relu\"\n top: \"mixed_4/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_conv_1_conv2d\"\n top: \"mixed_4/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_conv_1_conv2d_bn\"\n top: \"mixed_4/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_conv_1_conv2d_relu\"\n top: \"mixed_4/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_conv_2_conv2d\"\n top: \"mixed_4/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_conv_2_conv2d_bn\"\n top: \"mixed_4/tower_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_conv2d\"\n top: \"mixed_4/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_conv2d_relu\"\n top: \"mixed_4/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_1_conv2d\"\n top: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_4/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_2_conv2d\"\n top: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_2_conv2d_relu\"\n top: \"mixed_4/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 128\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_3_conv2d\"\n top: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_3_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_1_conv_3_conv2d_relu\"\n top: \"mixed_4/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_1_conv_4_conv2d\"\n top: \"mixed_4/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_4/tower_1_conv_4_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_3/chconcat\"\n top: \"mixed_4/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_4/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/tower_2_AVG_pool\"\n top: \"mixed_4/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_4/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_4/tower_2_conv_conv2d\"\n top: \"mixed_4/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_4/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_4/tower_2_conv_conv2d_bn\"\n top: \"mixed_4/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_4/chconcat\"\n bottom: \"mixed_4/conv_conv2d_relu\"\nbottom: \"mixed_4/tower_conv_2_conv2d_relu\"\nbottom: \"mixed_4/tower_1_conv_4_conv2d_relu\"\nbottom: \"mixed_4/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_4/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_5/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/conv_conv2d\"\n top: \"mixed_5/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/conv_conv2d_bn\"\n top: \"mixed_5/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_conv_conv2d\"\n top: \"mixed_5/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_conv_conv2d_bn\"\n top: \"mixed_5/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_conv_conv2d_relu\"\n top: \"mixed_5/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_conv_1_conv2d\"\n top: \"mixed_5/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_conv_1_conv2d_bn\"\n top: \"mixed_5/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_conv_1_conv2d_relu\"\n top: \"mixed_5/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_conv_2_conv2d\"\n top: \"mixed_5/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_conv_2_conv2d_bn\"\n top: \"mixed_5/tower_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_conv2d\"\n top: \"mixed_5/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_conv2d_relu\"\n top: \"mixed_5/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_1_conv2d\"\n top: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_5/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_2_conv2d\"\n top: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_2_conv2d_relu\"\n top: \"mixed_5/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_3_conv2d\"\n top: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_3_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_1_conv_3_conv2d_relu\"\n top: \"mixed_5/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_1_conv_4_conv2d\"\n top: \"mixed_5/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_5/tower_1_conv_4_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_4/chconcat\"\n top: \"mixed_5/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_5/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/tower_2_AVG_pool\"\n top: \"mixed_5/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_5/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_5/tower_2_conv_conv2d\"\n top: \"mixed_5/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_5/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_5/tower_2_conv_conv2d_bn\"\n top: \"mixed_5/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_5/chconcat\"\n bottom: \"mixed_5/conv_conv2d_relu\"\nbottom: \"mixed_5/tower_conv_2_conv2d_relu\"\nbottom: \"mixed_5/tower_1_conv_4_conv2d_relu\"\nbottom: \"mixed_5/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_5/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_6/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/conv_conv2d\"\n top: \"mixed_6/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/conv_conv2d_bn\"\n top: \"mixed_6/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_conv_conv2d\"\n top: \"mixed_6/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_conv_conv2d_bn\"\n top: \"mixed_6/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_conv_conv2d_relu\"\n top: \"mixed_6/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_conv_1_conv2d\"\n top: \"mixed_6/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_conv_1_conv2d_bn\"\n top: \"mixed_6/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_conv_1_conv2d_relu\"\n top: \"mixed_6/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_conv_2_conv2d\"\n top: \"mixed_6/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_conv_2_conv2d_bn\"\n top: \"mixed_6/tower_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_conv2d\"\n top: \"mixed_6/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_conv2d_relu\"\n top: \"mixed_6/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_1_conv2d\"\n top: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_6/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_2_conv2d\"\n top: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_2_conv2d_relu\"\n top: \"mixed_6/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 160\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_3_conv2d\"\n top: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_3_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_1_conv_3_conv2d_relu\"\n top: \"mixed_6/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_1_conv_4_conv2d\"\n top: \"mixed_6/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_6/tower_1_conv_4_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_5/chconcat\"\n top: \"mixed_6/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_6/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/tower_2_AVG_pool\"\n top: \"mixed_6/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_6/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_6/tower_2_conv_conv2d\"\n top: \"mixed_6/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_6/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_6/tower_2_conv_conv2d_bn\"\n top: \"mixed_6/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_6/chconcat\"\n bottom: \"mixed_6/conv_conv2d_relu\"\nbottom: \"mixed_6/tower_conv_2_conv2d_relu\"\nbottom: \"mixed_6/tower_1_conv_4_conv2d_relu\"\nbottom: \"mixed_6/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_6/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_7/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/conv_conv2d\"\n top: \"mixed_7/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/conv_conv2d_bn\"\n top: \"mixed_7/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_conv_conv2d\"\n top: \"mixed_7/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_conv_conv2d_bn\"\n top: \"mixed_7/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_conv_conv2d_relu\"\n top: \"mixed_7/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_conv_1_conv2d\"\n top: \"mixed_7/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_conv_1_conv2d_bn\"\n top: \"mixed_7/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_conv_1_conv2d_relu\"\n top: \"mixed_7/tower_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_conv_2_conv2d\"\n top: \"mixed_7/tower_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_conv_2_conv2d_bn\"\n top: \"mixed_7/tower_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_conv2d\"\n top: \"mixed_7/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_conv2d_relu\"\n top: \"mixed_7/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_1_conv2d\"\n top: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_7/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_2_conv2d\"\n top: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_2_conv2d_relu\"\n top: \"mixed_7/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_3_conv2d\"\n top: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_3_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_1_conv_4_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_1_conv_3_conv2d_relu\"\n top: \"mixed_7/tower_1_conv_4_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_1_conv_4_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_1_conv_4_conv2d\"\n top: \"mixed_7/tower_1_conv_4_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_1_conv_4_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_1_conv_4_conv2d_bn\"\n top: \"mixed_7/tower_1_conv_4_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_6/chconcat\"\n top: \"mixed_7/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_7/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/tower_2_AVG_pool\"\n top: \"mixed_7/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_7/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_7/tower_2_conv_conv2d\"\n top: \"mixed_7/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_7/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_7/tower_2_conv_conv2d_bn\"\n top: \"mixed_7/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_7/chconcat\"\n bottom: \"mixed_7/conv_conv2d_relu\"\nbottom: \"mixed_7/tower_conv_2_conv2d_relu\"\nbottom: \"mixed_7/tower_1_conv_4_conv2d_relu\"\nbottom: \"mixed_7/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_7/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_8/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/chconcat\"\n top: \"mixed_8/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_conv_conv2d\"\n top: \"mixed_8/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_conv_conv2d_bn\"\n top: \"mixed_8/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_8/tower_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_conv_conv2d_relu\"\n top: \"mixed_8/tower_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 320\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_conv_1_conv2d\"\n top: \"mixed_8/tower_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_conv_1_conv2d_bn\"\n top: \"mixed_8/tower_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_7/chconcat\"\n top: \"mixed_8/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_conv2d\"\n top: \"mixed_8/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_1_conv_conv2d_relu\"\n top: \"mixed_8/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 0\npad_w: 3\n kernel_h: 1\nkernel_w: 7\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_1_conv2d\"\n top: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_2_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_8/tower_1_conv_2_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad_h: 3\npad_w: 0\n kernel_h: 7\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_2_conv2d\"\n top: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_2_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_2_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_2_conv2d_relu\"\n}\nlayer {\n name: \"mixed_8/tower_1_conv_3_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/tower_1_conv_2_conv2d_relu\"\n top: \"mixed_8/tower_1_conv_3_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_8/tower_1_conv_3_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_8/tower_1_conv_3_conv2d\"\n top: \"mixed_8/tower_1_conv_3_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_8/tower_1_conv_3_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_8/tower_1_conv_3_conv2d_bn\"\n top: \"mixed_8/tower_1_conv_3_conv2d_relu\"\n}\nlayer {\n name: \"mixed_8/max_pool\"\n type: \"Pooling\"\n bottom: \"mixed_7/chconcat\"\n top: \"mixed_8/max_pool\"\n pooling_param {\n pool: MAX\n pad: 0\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"mixed_8/chconcat\"\n bottom: \"mixed_8/tower_conv_1_conv2d_relu\"\nbottom: \"mixed_8/tower_1_conv_3_conv2d_relu\"\nbottom: \"mixed_8/max_pool\"\n\n top: \"mixed_8/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_9/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 320\n pad: 0\n kernel_h: 1\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/conv_conv2d\"\n top: \"mixed_9/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/conv_conv2d_bn\"\n top: \"mixed_9/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_conv_conv2d\"\n top: \"mixed_9/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_conv_conv2d_bn\"\n top: \"mixed_9/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_conv_conv2d_relu\"\n top: \"mixed_9/tower_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_mixed_conv/conv2d\"\n top: \"mixed_9/tower_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_mixed_conv/conv2d_bn\"\n top: \"mixed_9/tower_mixed_conv/conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_conv_conv2d_relu\"\n top: \"mixed_9/tower_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_mixed_conv/1_conv2d\"\n top: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_mixed_conv/1_conv2d_bn\"\n top: \"mixed_9/tower_mixed_conv/1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 448\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_conv_conv2d\"\n top: \"mixed_9/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_conv_conv2d_bn\"\n top: \"mixed_9/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_1_conv_conv2d_relu\"\n top: \"mixed_9/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d\"\n top: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_9/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_9/tower_1_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_mixed_conv/conv2d\"\n top: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_mixed_conv/conv2d_bn\"\n top: \"mixed_9/tower_1_mixed_conv/conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_9/tower_1_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_1_mixed_conv/1_conv2d\"\n top: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_1_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_1_mixed_conv/1_conv2d_bn\"\n top: \"mixed_9/tower_1_mixed_conv/1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/tower_2_AVG_pool\"\n type: \"Pooling\"\n bottom: \"mixed_8/chconcat\"\n top: \"mixed_9/tower_2_AVG_pool\"\n pooling_param {\n pool: AVE\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_9/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/tower_2_AVG_pool\"\n top: \"mixed_9/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_9/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_9/tower_2_conv_conv2d\"\n top: \"mixed_9/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_9/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_9/tower_2_conv_conv2d_bn\"\n top: \"mixed_9/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_9/chconcat\"\n bottom: \"mixed_9/conv_conv2d_relu\"\nbottom: \"mixed_9/tower_mixed_conv/conv2d_relu\"\nbottom: \"mixed_9/tower_mixed_conv/1_conv2d_relu\"\nbottom: \"mixed_9/tower_1_mixed_conv/conv2d_relu\"\nbottom: \"mixed_9/tower_1_mixed_conv/1_conv2d_relu\"\nbottom: \"mixed_9/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_9/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"mixed_10/conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 320\n pad: 0\n kernel_h: 1\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/conv_conv2d\"\n top: \"mixed_10/conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/conv_conv2d_bn\"\n top: \"mixed_10/conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/tower_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/tower_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_conv_conv2d\"\n top: \"mixed_10/tower_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_conv_conv2d_bn\"\n top: \"mixed_10/tower_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_conv_conv2d_relu\"\n top: \"mixed_10/tower_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_mixed_conv/conv2d\"\n top: \"mixed_10/tower_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_mixed_conv/conv2d_bn\"\n top: \"mixed_10/tower_mixed_conv/conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_conv_conv2d_relu\"\n top: \"mixed_10/tower_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_mixed_conv/1_conv2d\"\n top: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_mixed_conv/1_conv2d_bn\"\n top: \"mixed_10/tower_mixed_conv/1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/tower_1_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/tower_1_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 448\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_conv_conv2d\"\n top: \"mixed_10/tower_1_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_conv_conv2d_bn\"\n top: \"mixed_10/tower_1_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/tower_1_conv_1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_1_conv_conv2d_relu\"\n top: \"mixed_10/tower_1_conv_1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d\"\n top: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_conv_1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d_bn\"\n top: \"mixed_10/tower_1_conv_1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_10/tower_1_mixed_conv/conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 0\npad_w: 1\n kernel_h: 1\nkernel_w: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_mixed_conv/conv2d\"\n top: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_mixed_conv/conv2d_bn\"\n top: \"mixed_10/tower_1_mixed_conv/conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/1_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/tower_1_conv_1_conv2d_relu\"\n top: \"mixed_10/tower_1_mixed_conv/1_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 384\n pad_h: 1\npad_w: 0\n kernel_h: 3\nkernel_w: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_1_mixed_conv/1_conv2d\"\n top: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_1_mixed_conv/1_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_1_mixed_conv/1_conv2d_bn\"\n top: \"mixed_10/tower_1_mixed_conv/1_conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/max_pool\"\n type: \"Pooling\"\n bottom: \"mixed_9/chconcat\"\n top: \"mixed_10/max_pool\"\n pooling_param {\n pool: MAX\n pad: 1\n kernel_size: 3\n stride: 1\n }\n}\nlayer {\n name: \"mixed_10/tower_2_conv_conv2d\"\n type: \"Convolution\"\n bottom: \"mixed_10/max_pool\"\n top: \"mixed_10/tower_2_conv_conv2d\"\n convolution_param {\n bias_term: false\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n\n}\nlayer {\n name: \"mixed_10/tower_2_conv_conv2d_bn\"\n type: \"BatchNorm\"\n bottom: \"mixed_10/tower_2_conv_conv2d\"\n top: \"mixed_10/tower_2_conv_conv2d_bn\"\n batch_norm_param {\n use_global_stats: false\n eps: 0.001\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n param {\n lr_mult: 0\n }\n}\nlayer {\n name: \"mixed_10/tower_2_conv_conv2d_relu\"\n type: \"ReLU\"\n bottom: \"mixed_10/tower_2_conv_conv2d_bn\"\n top: \"mixed_10/tower_2_conv_conv2d_relu\"\n}\nlayer {\n name: \"mixed_10/chconcat\"\n bottom: \"mixed_10/conv_conv2d_relu\"\nbottom: \"mixed_10/tower_mixed_conv/conv2d_relu\"\nbottom: \"mixed_10/tower_mixed_conv/1_conv2d_relu\"\nbottom: \"mixed_10/tower_1_mixed_conv/conv2d_relu\"\nbottom: \"mixed_10/tower_1_mixed_conv/1_conv2d_relu\"\nbottom: \"mixed_10/tower_2_conv_conv2d_relu\"\n\n top: \"mixed_10/chconcat\"\n type: \"Concat\"\n concat_param {\n axis: 1\n }\n}\nlayer {\n name: \"global_pool\"\n type: \"Pooling\"\n bottom: \"mixed_10/chconcat\"\n top: \"global_pool\"\n pooling_param {\n pool: AVE\n pad: 0\n kernel_size: 8\n stride: 1\n }\n}\nlayer {\n name: \"flatten\"\n type: \"Flatten\"\n bottom: \"global_pool\"\n top: \"flatten\"\n}\nlayer {\n name: \"fc1\"\n type: \"InnerProduct\"\n bottom: \"flatten\"\n top: \"fc1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 1000\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"fc1\"\n top: \"loss\"\n}\nlayer {\n name: \"acc/top-1\"\n type: \"Accuracy\"\n bottom: \"fc1\"\n top: \"acc/top-1\"\n include {\n phase: TEST\n }\n}\nlayer {\n name: \"acc/top-5\"\n type: \"Accuracy\"\n bottom: \"fc1\"\n top: \"acc/top-5\"\n include {\n phase: TEST\n }\n accuracy_param {\n top_k: 5\n }\n}" }, { "path": "presets/inceptionv4.prototxt", "content": "#downloaded from http://github.com/soeaver/caffe-model\nname: \"Inception v4\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n input_param {\n shape: {\n dim: 1\n dim: 3\n dim: 299\n dim: 299\n }\n }\n}\nlayer {\n name: \"conv1_3x3_s2\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1_3x3_s2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv1_3x3_s2_bn\"\n type: \"BatchNorm\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv1_3x3_s2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"conv1_3x3_s2_scale\"\n type: \"Scale\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv1_3x3_s2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"conv1_3x3_s2_relu\"\n type: \"ReLU\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv1_3x3_s2\"\n}\nlayer {\n name: \"conv2_3x3_s1\"\n type: \"Convolution\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv2_3x3_s1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv2_3x3_s1_bn\"\n type: \"BatchNorm\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv2_3x3_s1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"conv2_3x3_s1_scale\"\n type: \"Scale\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv2_3x3_s1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"conv2_3x3_s1_relu\"\n type: \"ReLU\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv2_3x3_s1\"\n}\nlayer {\n name: \"conv3_3x3_s1\"\n type: \"Convolution\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv3_3x3_s1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv3_3x3_s1_bn\"\n type: \"BatchNorm\"\n bottom: \"conv3_3x3_s1\"\n top: \"conv3_3x3_s1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"conv3_3x3_s1_scale\"\n type: \"Scale\"\n bottom: \"conv3_3x3_s1\"\n top: \"conv3_3x3_s1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"conv3_3x3_s1_relu\"\n type: \"ReLU\"\n bottom: \"conv3_3x3_s1\"\n top: \"conv3_3x3_s1\"\n}\nlayer {\n name: \"inception_stem1_3x3_s2\"\n type: \"Convolution\"\n bottom: \"conv3_3x3_s1\"\n top: \"inception_stem1_3x3_s2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem1_3x3_s2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem1_3x3_s2\"\n top: \"inception_stem1_3x3_s2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem1_3x3_s2_scale\"\n type: \"Scale\"\n bottom: \"inception_stem1_3x3_s2\"\n top: \"inception_stem1_3x3_s2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem1_3x3_s2_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem1_3x3_s2\"\n top: \"inception_stem1_3x3_s2\"\n}\nlayer {\n name: \"inception_stem1_pool\"\n type: \"Pooling\"\n bottom: \"conv3_3x3_s1\"\n top: \"inception_stem1_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"inception_stem1\"\n type: \"Concat\"\n bottom: \"inception_stem1_3x3_s2\"\n bottom: \"inception_stem1_pool\"\n top: \"inception_stem1\"\n}\nlayer {\n name: \"inception_stem2_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem1\"\n top: \"inception_stem2_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3_reduce\"\n}\nlayer {\n name: \"inception_stem2_3x3\"\n type: \"Convolution\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_3x3\"\n top: \"inception_stem2_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_3x3\"\n top: \"inception_stem2_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_3x3\"\n top: \"inception_stem2_3x3\"\n}\nlayer {\n name: \"inception_stem2_7x1_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem1\"\n top: \"inception_stem2_7x1_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_7x1_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_7x1_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_7x1_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1_reduce\"\n}\nlayer {\n name: \"inception_stem2_7x1\"\n type: \"Convolution\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_stem2_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_7x1\"\n}\nlayer {\n name: \"inception_stem2_1x7\"\n type: \"Convolution\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_stem2_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_1x7\"\n}\nlayer {\n name: \"inception_stem2_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2_3x3_2\"\n}\nlayer {\n name: \"inception_stem2\"\n type: \"Concat\"\n bottom: \"inception_stem2_3x3\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2\"\n}\nlayer {\n name: \"inception_stem3_3x3_s2\"\n type: \"Convolution\"\n bottom: \"inception_stem2\"\n top: \"inception_stem3_3x3_s2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem3_3x3_s2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem3_3x3_s2\"\n top: \"inception_stem3_3x3_s2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem3_3x3_s2_scale\"\n type: \"Scale\"\n bottom: \"inception_stem3_3x3_s2\"\n top: \"inception_stem3_3x3_s2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem3_3x3_s2_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem3_3x3_s2\"\n top: \"inception_stem3_3x3_s2\"\n}\nlayer {\n name: \"inception_stem3_pool\"\n type: \"Pooling\"\n bottom: \"inception_stem2\"\n top: \"inception_stem3_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"inception_stem3\"\n type: \"Concat\"\n bottom: \"inception_stem3_3x3_s2\"\n bottom: \"inception_stem3_pool\"\n top: \"inception_stem3\"\n}\nlayer {\n name: \"inception_a1_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_stem3\"\n top: \"inception_a1_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_a1_1x1\"\n type: \"Convolution\"\n bottom: \"inception_a1_pool_ave\"\n top: \"inception_a1_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a1_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a1_1x1\"\n top: \"inception_a1_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a1_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_a1_1x1\"\n top: \"inception_a1_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a1_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_a1_1x1\"\n top: \"inception_a1_1x1\"\n}\nlayer {\n name: \"inception_a1_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_stem3\"\n top: \"inception_a1_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a1_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a1_1x1_2\"\n top: \"inception_a1_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a1_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a1_1x1_2\"\n top: \"inception_a1_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a1_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a1_1x1_2\"\n top: \"inception_a1_1x1_2\"\n}\nlayer {\n name: \"inception_a1_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem3\"\n top: \"inception_a1_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a1_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a1_3x3_reduce\"\n top: \"inception_a1_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a1_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a1_3x3_reduce\"\n top: \"inception_a1_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a1_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a1_3x3_reduce\"\n top: \"inception_a1_3x3_reduce\"\n}\nlayer {\n name: \"inception_a1_3x3\"\n type: \"Convolution\"\n bottom: \"inception_a1_3x3_reduce\"\n top: \"inception_a1_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a1_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a1_3x3\"\n top: \"inception_a1_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a1_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_a1_3x3\"\n top: \"inception_a1_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a1_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a1_3x3\"\n top: \"inception_a1_3x3\"\n}\nlayer {\n name: \"inception_a1_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem3\"\n top: \"inception_a1_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a1_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a1_3x3_2_reduce\"\n top: \"inception_a1_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a1_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a1_3x3_2_reduce\"\n top: \"inception_a1_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a1_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a1_3x3_2_reduce\"\n top: \"inception_a1_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_a1_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_a1_3x3_2_reduce\"\n top: \"inception_a1_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a1_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a1_3x3_2\"\n top: \"inception_a1_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a1_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a1_3x3_2\"\n top: \"inception_a1_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a1_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a1_3x3_2\"\n top: \"inception_a1_3x3_2\"\n}\nlayer {\n name: \"inception_a1_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_a1_3x3_2\"\n top: \"inception_a1_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a1_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a1_3x3_3\"\n top: \"inception_a1_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a1_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_a1_3x3_3\"\n top: \"inception_a1_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a1_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a1_3x3_3\"\n top: \"inception_a1_3x3_3\"\n}\nlayer {\n name: \"inception_a1_concat\"\n type: \"Concat\"\n bottom: \"inception_a1_1x1\"\n bottom: \"inception_a1_1x1_2\"\n bottom: \"inception_a1_3x3\"\n bottom: \"inception_a1_3x3_3\"\n top: \"inception_a1_concat\"\n}\nlayer {\n name: \"inception_a2_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_a1_concat\"\n top: \"inception_a2_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_a2_1x1\"\n type: \"Convolution\"\n bottom: \"inception_a2_pool_ave\"\n top: \"inception_a2_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a2_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a2_1x1\"\n top: \"inception_a2_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a2_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_a2_1x1\"\n top: \"inception_a2_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a2_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_a2_1x1\"\n top: \"inception_a2_1x1\"\n}\nlayer {\n name: \"inception_a2_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_a1_concat\"\n top: \"inception_a2_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a2_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a2_1x1_2\"\n top: \"inception_a2_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a2_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a2_1x1_2\"\n top: \"inception_a2_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a2_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a2_1x1_2\"\n top: \"inception_a2_1x1_2\"\n}\nlayer {\n name: \"inception_a2_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_a1_concat\"\n top: \"inception_a2_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a2_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a2_3x3_reduce\"\n top: \"inception_a2_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a2_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a2_3x3_reduce\"\n top: \"inception_a2_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a2_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a2_3x3_reduce\"\n top: \"inception_a2_3x3_reduce\"\n}\nlayer {\n name: \"inception_a2_3x3\"\n type: \"Convolution\"\n bottom: \"inception_a2_3x3_reduce\"\n top: \"inception_a2_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a2_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a2_3x3\"\n top: \"inception_a2_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a2_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_a2_3x3\"\n top: \"inception_a2_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a2_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a2_3x3\"\n top: \"inception_a2_3x3\"\n}\nlayer {\n name: \"inception_a2_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_a1_concat\"\n top: \"inception_a2_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a2_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a2_3x3_2_reduce\"\n top: \"inception_a2_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a2_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a2_3x3_2_reduce\"\n top: \"inception_a2_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a2_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a2_3x3_2_reduce\"\n top: \"inception_a2_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_a2_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_a2_3x3_2_reduce\"\n top: \"inception_a2_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a2_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a2_3x3_2\"\n top: \"inception_a2_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a2_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a2_3x3_2\"\n top: \"inception_a2_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a2_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a2_3x3_2\"\n top: \"inception_a2_3x3_2\"\n}\nlayer {\n name: \"inception_a2_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_a2_3x3_2\"\n top: \"inception_a2_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a2_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a2_3x3_3\"\n top: \"inception_a2_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a2_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_a2_3x3_3\"\n top: \"inception_a2_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a2_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a2_3x3_3\"\n top: \"inception_a2_3x3_3\"\n}\nlayer {\n name: \"inception_a2_concat\"\n type: \"Concat\"\n bottom: \"inception_a2_1x1\"\n bottom: \"inception_a2_1x1_2\"\n bottom: \"inception_a2_3x3\"\n bottom: \"inception_a2_3x3_3\"\n top: \"inception_a2_concat\"\n}\nlayer {\n name: \"inception_a3_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_a2_concat\"\n top: \"inception_a3_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_a3_1x1\"\n type: \"Convolution\"\n bottom: \"inception_a3_pool_ave\"\n top: \"inception_a3_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a3_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a3_1x1\"\n top: \"inception_a3_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a3_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_a3_1x1\"\n top: \"inception_a3_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a3_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_a3_1x1\"\n top: \"inception_a3_1x1\"\n}\nlayer {\n name: \"inception_a3_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_a2_concat\"\n top: \"inception_a3_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a3_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a3_1x1_2\"\n top: \"inception_a3_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a3_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a3_1x1_2\"\n top: \"inception_a3_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a3_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a3_1x1_2\"\n top: \"inception_a3_1x1_2\"\n}\nlayer {\n name: \"inception_a3_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_a2_concat\"\n top: \"inception_a3_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a3_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a3_3x3_reduce\"\n top: \"inception_a3_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a3_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a3_3x3_reduce\"\n top: \"inception_a3_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a3_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a3_3x3_reduce\"\n top: \"inception_a3_3x3_reduce\"\n}\nlayer {\n name: \"inception_a3_3x3\"\n type: \"Convolution\"\n bottom: \"inception_a3_3x3_reduce\"\n top: \"inception_a3_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a3_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a3_3x3\"\n top: \"inception_a3_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a3_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_a3_3x3\"\n top: \"inception_a3_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a3_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a3_3x3\"\n top: \"inception_a3_3x3\"\n}\nlayer {\n name: \"inception_a3_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_a2_concat\"\n top: \"inception_a3_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a3_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a3_3x3_2_reduce\"\n top: \"inception_a3_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a3_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a3_3x3_2_reduce\"\n top: \"inception_a3_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a3_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a3_3x3_2_reduce\"\n top: \"inception_a3_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_a3_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_a3_3x3_2_reduce\"\n top: \"inception_a3_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a3_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a3_3x3_2\"\n top: \"inception_a3_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a3_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a3_3x3_2\"\n top: \"inception_a3_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a3_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a3_3x3_2\"\n top: \"inception_a3_3x3_2\"\n}\nlayer {\n name: \"inception_a3_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_a3_3x3_2\"\n top: \"inception_a3_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a3_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a3_3x3_3\"\n top: \"inception_a3_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a3_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_a3_3x3_3\"\n top: \"inception_a3_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a3_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a3_3x3_3\"\n top: \"inception_a3_3x3_3\"\n}\nlayer {\n name: \"inception_a3_concat\"\n type: \"Concat\"\n bottom: \"inception_a3_1x1\"\n bottom: \"inception_a3_1x1_2\"\n bottom: \"inception_a3_3x3\"\n bottom: \"inception_a3_3x3_3\"\n top: \"inception_a3_concat\"\n}\nlayer {\n name: \"inception_a4_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_a3_concat\"\n top: \"inception_a4_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_a4_1x1\"\n type: \"Convolution\"\n bottom: \"inception_a4_pool_ave\"\n top: \"inception_a4_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a4_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a4_1x1\"\n top: \"inception_a4_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a4_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_a4_1x1\"\n top: \"inception_a4_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a4_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_a4_1x1\"\n top: \"inception_a4_1x1\"\n}\nlayer {\n name: \"inception_a4_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_a3_concat\"\n top: \"inception_a4_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a4_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a4_1x1_2\"\n top: \"inception_a4_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a4_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a4_1x1_2\"\n top: \"inception_a4_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a4_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a4_1x1_2\"\n top: \"inception_a4_1x1_2\"\n}\nlayer {\n name: \"inception_a4_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_a3_concat\"\n top: \"inception_a4_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a4_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a4_3x3_reduce\"\n top: \"inception_a4_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a4_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a4_3x3_reduce\"\n top: \"inception_a4_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a4_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a4_3x3_reduce\"\n top: \"inception_a4_3x3_reduce\"\n}\nlayer {\n name: \"inception_a4_3x3\"\n type: \"Convolution\"\n bottom: \"inception_a4_3x3_reduce\"\n top: \"inception_a4_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a4_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a4_3x3\"\n top: \"inception_a4_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a4_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_a4_3x3\"\n top: \"inception_a4_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a4_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a4_3x3\"\n top: \"inception_a4_3x3\"\n}\nlayer {\n name: \"inception_a4_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_a3_concat\"\n top: \"inception_a4_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a4_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a4_3x3_2_reduce\"\n top: \"inception_a4_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a4_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_a4_3x3_2_reduce\"\n top: \"inception_a4_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a4_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_a4_3x3_2_reduce\"\n top: \"inception_a4_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_a4_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_a4_3x3_2_reduce\"\n top: \"inception_a4_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a4_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a4_3x3_2\"\n top: \"inception_a4_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a4_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_a4_3x3_2\"\n top: \"inception_a4_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a4_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_a4_3x3_2\"\n top: \"inception_a4_3x3_2\"\n}\nlayer {\n name: \"inception_a4_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_a4_3x3_2\"\n top: \"inception_a4_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_a4_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_a4_3x3_3\"\n top: \"inception_a4_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_a4_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_a4_3x3_3\"\n top: \"inception_a4_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_a4_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_a4_3x3_3\"\n top: \"inception_a4_3x3_3\"\n}\nlayer {\n name: \"inception_a4_concat\"\n type: \"Concat\"\n bottom: \"inception_a4_1x1\"\n bottom: \"inception_a4_1x1_2\"\n bottom: \"inception_a4_3x3\"\n bottom: \"inception_a4_3x3_3\"\n top: \"inception_a4_concat\"\n}\nlayer {\n name: \"reduction_a_pool\"\n type: \"Pooling\"\n bottom: \"inception_a4_concat\"\n top: \"reduction_a_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"reduction_a_3x3\"\n type: \"Convolution\"\n bottom: \"inception_a4_concat\"\n top: \"reduction_a_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3\"\n top: \"reduction_a_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3\"\n top: \"reduction_a_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3\"\n top: \"reduction_a_3x3\"\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_a4_concat\"\n top: \"reduction_a_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2_reduce\"\n}\nlayer {\n name: \"reduction_a_3x3_2\"\n type: \"Convolution\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_2\"\n}\nlayer {\n name: \"reduction_a_3x3_3\"\n type: \"Convolution\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_3x3_3\"\n}\nlayer {\n name: \"reduction_a_concat\"\n type: \"Concat\"\n bottom: \"reduction_a_pool\"\n bottom: \"reduction_a_3x3\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_concat\"\n}\nlayer {\n name: \"inception_b1_pool_ave\"\n type: \"Pooling\"\n bottom: \"reduction_a_concat\"\n top: \"inception_b1_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_b1_1x1\"\n type: \"Convolution\"\n bottom: \"inception_b1_pool_ave\"\n top: \"inception_b1_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b1_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_1x1\"\n top: \"inception_b1_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_1x1\"\n top: \"inception_b1_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_1x1\"\n top: \"inception_b1_1x1\"\n}\nlayer {\n name: \"inception_b1_1x1_2\"\n type: \"Convolution\"\n bottom: \"reduction_a_concat\"\n top: \"inception_b1_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b1_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_1x1_2\"\n top: \"inception_b1_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_1x1_2\"\n top: \"inception_b1_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_1x1_2\"\n top: \"inception_b1_1x1_2\"\n}\nlayer {\n name: \"inception_b1_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"reduction_a_concat\"\n top: \"inception_b1_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b1_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_1x7_reduce\"\n top: \"inception_b1_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_1x7_reduce\"\n top: \"inception_b1_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_1x7_reduce\"\n top: \"inception_b1_1x7_reduce\"\n}\nlayer {\n name: \"inception_b1_1x7\"\n type: \"Convolution\"\n bottom: \"inception_b1_1x7_reduce\"\n top: \"inception_b1_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b1_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_1x7\"\n top: \"inception_b1_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_1x7\"\n top: \"inception_b1_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_1x7\"\n top: \"inception_b1_1x7\"\n}\nlayer {\n name: \"inception_b1_7x1\"\n type: \"Convolution\"\n bottom: \"inception_b1_1x7\"\n top: \"inception_b1_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b1_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_7x1\"\n top: \"inception_b1_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_7x1\"\n top: \"inception_b1_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_7x1\"\n top: \"inception_b1_7x1\"\n}\nlayer {\n name: \"inception_b1_1x7_2_reduce\"\n type: \"Convolution\"\n bottom: \"reduction_a_concat\"\n top: \"inception_b1_1x7_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b1_1x7_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_1x7_2_reduce\"\n top: \"inception_b1_1x7_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_1x7_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_1x7_2_reduce\"\n top: \"inception_b1_1x7_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_1x7_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_1x7_2_reduce\"\n top: \"inception_b1_1x7_2_reduce\"\n}\nlayer {\n name: \"inception_b1_1x7_2\"\n type: \"Convolution\"\n bottom: \"inception_b1_1x7_2_reduce\"\n top: \"inception_b1_1x7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b1_1x7_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_1x7_2\"\n top: \"inception_b1_1x7_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_1x7_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_1x7_2\"\n top: \"inception_b1_1x7_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_1x7_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_1x7_2\"\n top: \"inception_b1_1x7_2\"\n}\nlayer {\n name: \"inception_b1_7x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b1_1x7_2\"\n top: \"inception_b1_7x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b1_7x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_7x1_2\"\n top: \"inception_b1_7x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_7x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_7x1_2\"\n top: \"inception_b1_7x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_7x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_7x1_2\"\n top: \"inception_b1_7x1_2\"\n}\nlayer {\n name: \"inception_b1_1x7_3\"\n type: \"Convolution\"\n bottom: \"inception_b1_7x1_2\"\n top: \"inception_b1_1x7_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b1_1x7_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_1x7_3\"\n top: \"inception_b1_1x7_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_1x7_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_1x7_3\"\n top: \"inception_b1_1x7_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_1x7_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_1x7_3\"\n top: \"inception_b1_1x7_3\"\n}\nlayer {\n name: \"inception_b1_7x1_3\"\n type: \"Convolution\"\n bottom: \"inception_b1_1x7_3\"\n top: \"inception_b1_7x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b1_7x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b1_7x1_3\"\n top: \"inception_b1_7x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b1_7x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b1_7x1_3\"\n top: \"inception_b1_7x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b1_7x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b1_7x1_3\"\n top: \"inception_b1_7x1_3\"\n}\nlayer {\n name: \"inception_b1_concat\"\n type: \"Concat\"\n bottom: \"inception_b1_1x1\"\n bottom: \"inception_b1_1x1_2\"\n bottom: \"inception_b1_7x1\"\n bottom: \"inception_b1_7x1_3\"\n top: \"inception_b1_concat\"\n}\nlayer {\n name: \"inception_b2_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_b1_concat\"\n top: \"inception_b2_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_b2_1x1\"\n type: \"Convolution\"\n bottom: \"inception_b2_pool_ave\"\n top: \"inception_b2_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b2_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_1x1\"\n top: \"inception_b2_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_1x1\"\n top: \"inception_b2_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_1x1\"\n top: \"inception_b2_1x1\"\n}\nlayer {\n name: \"inception_b2_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b1_concat\"\n top: \"inception_b2_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b2_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_1x1_2\"\n top: \"inception_b2_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_1x1_2\"\n top: \"inception_b2_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_1x1_2\"\n top: \"inception_b2_1x1_2\"\n}\nlayer {\n name: \"inception_b2_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b1_concat\"\n top: \"inception_b2_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b2_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_1x7_reduce\"\n top: \"inception_b2_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_1x7_reduce\"\n top: \"inception_b2_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_1x7_reduce\"\n top: \"inception_b2_1x7_reduce\"\n}\nlayer {\n name: \"inception_b2_1x7\"\n type: \"Convolution\"\n bottom: \"inception_b2_1x7_reduce\"\n top: \"inception_b2_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b2_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_1x7\"\n top: \"inception_b2_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_1x7\"\n top: \"inception_b2_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_1x7\"\n top: \"inception_b2_1x7\"\n}\nlayer {\n name: \"inception_b2_7x1\"\n type: \"Convolution\"\n bottom: \"inception_b2_1x7\"\n top: \"inception_b2_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b2_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_7x1\"\n top: \"inception_b2_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_7x1\"\n top: \"inception_b2_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_7x1\"\n top: \"inception_b2_7x1\"\n}\nlayer {\n name: \"inception_b2_1x7_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b1_concat\"\n top: \"inception_b2_1x7_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b2_1x7_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_1x7_2_reduce\"\n top: \"inception_b2_1x7_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_1x7_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_1x7_2_reduce\"\n top: \"inception_b2_1x7_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_1x7_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_1x7_2_reduce\"\n top: \"inception_b2_1x7_2_reduce\"\n}\nlayer {\n name: \"inception_b2_1x7_2\"\n type: \"Convolution\"\n bottom: \"inception_b2_1x7_2_reduce\"\n top: \"inception_b2_1x7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b2_1x7_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_1x7_2\"\n top: \"inception_b2_1x7_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_1x7_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_1x7_2\"\n top: \"inception_b2_1x7_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_1x7_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_1x7_2\"\n top: \"inception_b2_1x7_2\"\n}\nlayer {\n name: \"inception_b2_7x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b2_1x7_2\"\n top: \"inception_b2_7x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b2_7x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_7x1_2\"\n top: \"inception_b2_7x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_7x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_7x1_2\"\n top: \"inception_b2_7x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_7x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_7x1_2\"\n top: \"inception_b2_7x1_2\"\n}\nlayer {\n name: \"inception_b2_1x7_3\"\n type: \"Convolution\"\n bottom: \"inception_b2_7x1_2\"\n top: \"inception_b2_1x7_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b2_1x7_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_1x7_3\"\n top: \"inception_b2_1x7_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_1x7_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_1x7_3\"\n top: \"inception_b2_1x7_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_1x7_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_1x7_3\"\n top: \"inception_b2_1x7_3\"\n}\nlayer {\n name: \"inception_b2_7x1_3\"\n type: \"Convolution\"\n bottom: \"inception_b2_1x7_3\"\n top: \"inception_b2_7x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b2_7x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b2_7x1_3\"\n top: \"inception_b2_7x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b2_7x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b2_7x1_3\"\n top: \"inception_b2_7x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b2_7x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b2_7x1_3\"\n top: \"inception_b2_7x1_3\"\n}\nlayer {\n name: \"inception_b2_concat\"\n type: \"Concat\"\n bottom: \"inception_b2_1x1\"\n bottom: \"inception_b2_1x1_2\"\n bottom: \"inception_b2_7x1\"\n bottom: \"inception_b2_7x1_3\"\n top: \"inception_b2_concat\"\n}\nlayer {\n name: \"inception_b3_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_b2_concat\"\n top: \"inception_b3_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_b3_1x1\"\n type: \"Convolution\"\n bottom: \"inception_b3_pool_ave\"\n top: \"inception_b3_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b3_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_1x1\"\n top: \"inception_b3_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_1x1\"\n top: \"inception_b3_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_1x1\"\n top: \"inception_b3_1x1\"\n}\nlayer {\n name: \"inception_b3_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b2_concat\"\n top: \"inception_b3_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b3_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_1x1_2\"\n top: \"inception_b3_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_1x1_2\"\n top: \"inception_b3_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_1x1_2\"\n top: \"inception_b3_1x1_2\"\n}\nlayer {\n name: \"inception_b3_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b2_concat\"\n top: \"inception_b3_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b3_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_1x7_reduce\"\n top: \"inception_b3_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_1x7_reduce\"\n top: \"inception_b3_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_1x7_reduce\"\n top: \"inception_b3_1x7_reduce\"\n}\nlayer {\n name: \"inception_b3_1x7\"\n type: \"Convolution\"\n bottom: \"inception_b3_1x7_reduce\"\n top: \"inception_b3_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b3_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_1x7\"\n top: \"inception_b3_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_1x7\"\n top: \"inception_b3_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_1x7\"\n top: \"inception_b3_1x7\"\n}\nlayer {\n name: \"inception_b3_7x1\"\n type: \"Convolution\"\n bottom: \"inception_b3_1x7\"\n top: \"inception_b3_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b3_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_7x1\"\n top: \"inception_b3_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_7x1\"\n top: \"inception_b3_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_7x1\"\n top: \"inception_b3_7x1\"\n}\nlayer {\n name: \"inception_b3_1x7_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b2_concat\"\n top: \"inception_b3_1x7_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b3_1x7_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_1x7_2_reduce\"\n top: \"inception_b3_1x7_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_1x7_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_1x7_2_reduce\"\n top: \"inception_b3_1x7_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_1x7_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_1x7_2_reduce\"\n top: \"inception_b3_1x7_2_reduce\"\n}\nlayer {\n name: \"inception_b3_1x7_2\"\n type: \"Convolution\"\n bottom: \"inception_b3_1x7_2_reduce\"\n top: \"inception_b3_1x7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b3_1x7_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_1x7_2\"\n top: \"inception_b3_1x7_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_1x7_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_1x7_2\"\n top: \"inception_b3_1x7_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_1x7_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_1x7_2\"\n top: \"inception_b3_1x7_2\"\n}\nlayer {\n name: \"inception_b3_7x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b3_1x7_2\"\n top: \"inception_b3_7x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b3_7x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_7x1_2\"\n top: \"inception_b3_7x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_7x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_7x1_2\"\n top: \"inception_b3_7x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_7x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_7x1_2\"\n top: \"inception_b3_7x1_2\"\n}\nlayer {\n name: \"inception_b3_1x7_3\"\n type: \"Convolution\"\n bottom: \"inception_b3_7x1_2\"\n top: \"inception_b3_1x7_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b3_1x7_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_1x7_3\"\n top: \"inception_b3_1x7_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_1x7_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_1x7_3\"\n top: \"inception_b3_1x7_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_1x7_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_1x7_3\"\n top: \"inception_b3_1x7_3\"\n}\nlayer {\n name: \"inception_b3_7x1_3\"\n type: \"Convolution\"\n bottom: \"inception_b3_1x7_3\"\n top: \"inception_b3_7x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b3_7x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b3_7x1_3\"\n top: \"inception_b3_7x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b3_7x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b3_7x1_3\"\n top: \"inception_b3_7x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b3_7x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b3_7x1_3\"\n top: \"inception_b3_7x1_3\"\n}\nlayer {\n name: \"inception_b3_concat\"\n type: \"Concat\"\n bottom: \"inception_b3_1x1\"\n bottom: \"inception_b3_1x1_2\"\n bottom: \"inception_b3_7x1\"\n bottom: \"inception_b3_7x1_3\"\n top: \"inception_b3_concat\"\n}\nlayer {\n name: \"inception_b4_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_b3_concat\"\n top: \"inception_b4_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_b4_1x1\"\n type: \"Convolution\"\n bottom: \"inception_b4_pool_ave\"\n top: \"inception_b4_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b4_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_1x1\"\n top: \"inception_b4_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_1x1\"\n top: \"inception_b4_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_1x1\"\n top: \"inception_b4_1x1\"\n}\nlayer {\n name: \"inception_b4_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b3_concat\"\n top: \"inception_b4_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b4_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_1x1_2\"\n top: \"inception_b4_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_1x1_2\"\n top: \"inception_b4_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_1x1_2\"\n top: \"inception_b4_1x1_2\"\n}\nlayer {\n name: \"inception_b4_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b3_concat\"\n top: \"inception_b4_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b4_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_1x7_reduce\"\n top: \"inception_b4_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_1x7_reduce\"\n top: \"inception_b4_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_1x7_reduce\"\n top: \"inception_b4_1x7_reduce\"\n}\nlayer {\n name: \"inception_b4_1x7\"\n type: \"Convolution\"\n bottom: \"inception_b4_1x7_reduce\"\n top: \"inception_b4_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b4_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_1x7\"\n top: \"inception_b4_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_1x7\"\n top: \"inception_b4_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_1x7\"\n top: \"inception_b4_1x7\"\n}\nlayer {\n name: \"inception_b4_7x1\"\n type: \"Convolution\"\n bottom: \"inception_b4_1x7\"\n top: \"inception_b4_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b4_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_7x1\"\n top: \"inception_b4_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_7x1\"\n top: \"inception_b4_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_7x1\"\n top: \"inception_b4_7x1\"\n}\nlayer {\n name: \"inception_b4_1x7_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b3_concat\"\n top: \"inception_b4_1x7_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b4_1x7_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_1x7_2_reduce\"\n top: \"inception_b4_1x7_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_1x7_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_1x7_2_reduce\"\n top: \"inception_b4_1x7_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_1x7_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_1x7_2_reduce\"\n top: \"inception_b4_1x7_2_reduce\"\n}\nlayer {\n name: \"inception_b4_1x7_2\"\n type: \"Convolution\"\n bottom: \"inception_b4_1x7_2_reduce\"\n top: \"inception_b4_1x7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b4_1x7_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_1x7_2\"\n top: \"inception_b4_1x7_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_1x7_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_1x7_2\"\n top: \"inception_b4_1x7_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_1x7_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_1x7_2\"\n top: \"inception_b4_1x7_2\"\n}\nlayer {\n name: \"inception_b4_7x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b4_1x7_2\"\n top: \"inception_b4_7x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b4_7x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_7x1_2\"\n top: \"inception_b4_7x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_7x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_7x1_2\"\n top: \"inception_b4_7x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_7x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_7x1_2\"\n top: \"inception_b4_7x1_2\"\n}\nlayer {\n name: \"inception_b4_1x7_3\"\n type: \"Convolution\"\n bottom: \"inception_b4_7x1_2\"\n top: \"inception_b4_1x7_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b4_1x7_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_1x7_3\"\n top: \"inception_b4_1x7_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_1x7_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_1x7_3\"\n top: \"inception_b4_1x7_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_1x7_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_1x7_3\"\n top: \"inception_b4_1x7_3\"\n}\nlayer {\n name: \"inception_b4_7x1_3\"\n type: \"Convolution\"\n bottom: \"inception_b4_1x7_3\"\n top: \"inception_b4_7x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b4_7x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b4_7x1_3\"\n top: \"inception_b4_7x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b4_7x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b4_7x1_3\"\n top: \"inception_b4_7x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b4_7x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b4_7x1_3\"\n top: \"inception_b4_7x1_3\"\n}\nlayer {\n name: \"inception_b4_concat\"\n type: \"Concat\"\n bottom: \"inception_b4_1x1\"\n bottom: \"inception_b4_1x1_2\"\n bottom: \"inception_b4_7x1\"\n bottom: \"inception_b4_7x1_3\"\n top: \"inception_b4_concat\"\n}\nlayer {\n name: \"inception_b5_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_b4_concat\"\n top: \"inception_b5_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_b5_1x1\"\n type: \"Convolution\"\n bottom: \"inception_b5_pool_ave\"\n top: \"inception_b5_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b5_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_1x1\"\n top: \"inception_b5_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_1x1\"\n top: \"inception_b5_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_1x1\"\n top: \"inception_b5_1x1\"\n}\nlayer {\n name: \"inception_b5_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b4_concat\"\n top: \"inception_b5_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b5_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_1x1_2\"\n top: \"inception_b5_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_1x1_2\"\n top: \"inception_b5_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_1x1_2\"\n top: \"inception_b5_1x1_2\"\n}\nlayer {\n name: \"inception_b5_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b4_concat\"\n top: \"inception_b5_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b5_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_1x7_reduce\"\n top: \"inception_b5_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_1x7_reduce\"\n top: \"inception_b5_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_1x7_reduce\"\n top: \"inception_b5_1x7_reduce\"\n}\nlayer {\n name: \"inception_b5_1x7\"\n type: \"Convolution\"\n bottom: \"inception_b5_1x7_reduce\"\n top: \"inception_b5_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b5_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_1x7\"\n top: \"inception_b5_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_1x7\"\n top: \"inception_b5_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_1x7\"\n top: \"inception_b5_1x7\"\n}\nlayer {\n name: \"inception_b5_7x1\"\n type: \"Convolution\"\n bottom: \"inception_b5_1x7\"\n top: \"inception_b5_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b5_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_7x1\"\n top: \"inception_b5_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_7x1\"\n top: \"inception_b5_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_7x1\"\n top: \"inception_b5_7x1\"\n}\nlayer {\n name: \"inception_b5_1x7_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b4_concat\"\n top: \"inception_b5_1x7_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b5_1x7_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_1x7_2_reduce\"\n top: \"inception_b5_1x7_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_1x7_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_1x7_2_reduce\"\n top: \"inception_b5_1x7_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_1x7_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_1x7_2_reduce\"\n top: \"inception_b5_1x7_2_reduce\"\n}\nlayer {\n name: \"inception_b5_1x7_2\"\n type: \"Convolution\"\n bottom: \"inception_b5_1x7_2_reduce\"\n top: \"inception_b5_1x7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b5_1x7_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_1x7_2\"\n top: \"inception_b5_1x7_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_1x7_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_1x7_2\"\n top: \"inception_b5_1x7_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_1x7_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_1x7_2\"\n top: \"inception_b5_1x7_2\"\n}\nlayer {\n name: \"inception_b5_7x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b5_1x7_2\"\n top: \"inception_b5_7x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b5_7x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_7x1_2\"\n top: \"inception_b5_7x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_7x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_7x1_2\"\n top: \"inception_b5_7x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_7x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_7x1_2\"\n top: \"inception_b5_7x1_2\"\n}\nlayer {\n name: \"inception_b5_1x7_3\"\n type: \"Convolution\"\n bottom: \"inception_b5_7x1_2\"\n top: \"inception_b5_1x7_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b5_1x7_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_1x7_3\"\n top: \"inception_b5_1x7_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_1x7_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_1x7_3\"\n top: \"inception_b5_1x7_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_1x7_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_1x7_3\"\n top: \"inception_b5_1x7_3\"\n}\nlayer {\n name: \"inception_b5_7x1_3\"\n type: \"Convolution\"\n bottom: \"inception_b5_1x7_3\"\n top: \"inception_b5_7x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b5_7x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b5_7x1_3\"\n top: \"inception_b5_7x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b5_7x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b5_7x1_3\"\n top: \"inception_b5_7x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b5_7x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b5_7x1_3\"\n top: \"inception_b5_7x1_3\"\n}\nlayer {\n name: \"inception_b5_concat\"\n type: \"Concat\"\n bottom: \"inception_b5_1x1\"\n bottom: \"inception_b5_1x1_2\"\n bottom: \"inception_b5_7x1\"\n bottom: \"inception_b5_7x1_3\"\n top: \"inception_b5_concat\"\n}\nlayer {\n name: \"inception_b6_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_b5_concat\"\n top: \"inception_b6_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_b6_1x1\"\n type: \"Convolution\"\n bottom: \"inception_b6_pool_ave\"\n top: \"inception_b6_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b6_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_1x1\"\n top: \"inception_b6_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_1x1\"\n top: \"inception_b6_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_1x1\"\n top: \"inception_b6_1x1\"\n}\nlayer {\n name: \"inception_b6_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b5_concat\"\n top: \"inception_b6_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b6_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_1x1_2\"\n top: \"inception_b6_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_1x1_2\"\n top: \"inception_b6_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_1x1_2\"\n top: \"inception_b6_1x1_2\"\n}\nlayer {\n name: \"inception_b6_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b5_concat\"\n top: \"inception_b6_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b6_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_1x7_reduce\"\n top: \"inception_b6_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_1x7_reduce\"\n top: \"inception_b6_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_1x7_reduce\"\n top: \"inception_b6_1x7_reduce\"\n}\nlayer {\n name: \"inception_b6_1x7\"\n type: \"Convolution\"\n bottom: \"inception_b6_1x7_reduce\"\n top: \"inception_b6_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b6_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_1x7\"\n top: \"inception_b6_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_1x7\"\n top: \"inception_b6_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_1x7\"\n top: \"inception_b6_1x7\"\n}\nlayer {\n name: \"inception_b6_7x1\"\n type: \"Convolution\"\n bottom: \"inception_b6_1x7\"\n top: \"inception_b6_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b6_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_7x1\"\n top: \"inception_b6_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_7x1\"\n top: \"inception_b6_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_7x1\"\n top: \"inception_b6_7x1\"\n}\nlayer {\n name: \"inception_b6_1x7_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b5_concat\"\n top: \"inception_b6_1x7_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b6_1x7_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_1x7_2_reduce\"\n top: \"inception_b6_1x7_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_1x7_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_1x7_2_reduce\"\n top: \"inception_b6_1x7_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_1x7_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_1x7_2_reduce\"\n top: \"inception_b6_1x7_2_reduce\"\n}\nlayer {\n name: \"inception_b6_1x7_2\"\n type: \"Convolution\"\n bottom: \"inception_b6_1x7_2_reduce\"\n top: \"inception_b6_1x7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b6_1x7_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_1x7_2\"\n top: \"inception_b6_1x7_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_1x7_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_1x7_2\"\n top: \"inception_b6_1x7_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_1x7_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_1x7_2\"\n top: \"inception_b6_1x7_2\"\n}\nlayer {\n name: \"inception_b6_7x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b6_1x7_2\"\n top: \"inception_b6_7x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b6_7x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_7x1_2\"\n top: \"inception_b6_7x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_7x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_7x1_2\"\n top: \"inception_b6_7x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_7x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_7x1_2\"\n top: \"inception_b6_7x1_2\"\n}\nlayer {\n name: \"inception_b6_1x7_3\"\n type: \"Convolution\"\n bottom: \"inception_b6_7x1_2\"\n top: \"inception_b6_1x7_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b6_1x7_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_1x7_3\"\n top: \"inception_b6_1x7_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_1x7_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_1x7_3\"\n top: \"inception_b6_1x7_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_1x7_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_1x7_3\"\n top: \"inception_b6_1x7_3\"\n}\nlayer {\n name: \"inception_b6_7x1_3\"\n type: \"Convolution\"\n bottom: \"inception_b6_1x7_3\"\n top: \"inception_b6_7x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b6_7x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b6_7x1_3\"\n top: \"inception_b6_7x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b6_7x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b6_7x1_3\"\n top: \"inception_b6_7x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b6_7x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b6_7x1_3\"\n top: \"inception_b6_7x1_3\"\n}\nlayer {\n name: \"inception_b6_concat\"\n type: \"Concat\"\n bottom: \"inception_b6_1x1\"\n bottom: \"inception_b6_1x1_2\"\n bottom: \"inception_b6_7x1\"\n bottom: \"inception_b6_7x1_3\"\n top: \"inception_b6_concat\"\n}\nlayer {\n name: \"inception_b7_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_b6_concat\"\n top: \"inception_b7_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_b7_1x1\"\n type: \"Convolution\"\n bottom: \"inception_b7_pool_ave\"\n top: \"inception_b7_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b7_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_1x1\"\n top: \"inception_b7_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_1x1\"\n top: \"inception_b7_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_1x1\"\n top: \"inception_b7_1x1\"\n}\nlayer {\n name: \"inception_b7_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b6_concat\"\n top: \"inception_b7_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b7_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_1x1_2\"\n top: \"inception_b7_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_1x1_2\"\n top: \"inception_b7_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_1x1_2\"\n top: \"inception_b7_1x1_2\"\n}\nlayer {\n name: \"inception_b7_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b6_concat\"\n top: \"inception_b7_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b7_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_1x7_reduce\"\n top: \"inception_b7_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_1x7_reduce\"\n top: \"inception_b7_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_1x7_reduce\"\n top: \"inception_b7_1x7_reduce\"\n}\nlayer {\n name: \"inception_b7_1x7\"\n type: \"Convolution\"\n bottom: \"inception_b7_1x7_reduce\"\n top: \"inception_b7_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b7_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_1x7\"\n top: \"inception_b7_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_1x7\"\n top: \"inception_b7_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_1x7\"\n top: \"inception_b7_1x7\"\n}\nlayer {\n name: \"inception_b7_7x1\"\n type: \"Convolution\"\n bottom: \"inception_b7_1x7\"\n top: \"inception_b7_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b7_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_7x1\"\n top: \"inception_b7_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_7x1\"\n top: \"inception_b7_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_7x1\"\n top: \"inception_b7_7x1\"\n}\nlayer {\n name: \"inception_b7_1x7_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b6_concat\"\n top: \"inception_b7_1x7_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_b7_1x7_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_1x7_2_reduce\"\n top: \"inception_b7_1x7_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_1x7_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_1x7_2_reduce\"\n top: \"inception_b7_1x7_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_1x7_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_1x7_2_reduce\"\n top: \"inception_b7_1x7_2_reduce\"\n}\nlayer {\n name: \"inception_b7_1x7_2\"\n type: \"Convolution\"\n bottom: \"inception_b7_1x7_2_reduce\"\n top: \"inception_b7_1x7_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b7_1x7_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_1x7_2\"\n top: \"inception_b7_1x7_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_1x7_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_1x7_2\"\n top: \"inception_b7_1x7_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_1x7_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_1x7_2\"\n top: \"inception_b7_1x7_2\"\n}\nlayer {\n name: \"inception_b7_7x1_2\"\n type: \"Convolution\"\n bottom: \"inception_b7_1x7_2\"\n top: \"inception_b7_7x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b7_7x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_7x1_2\"\n top: \"inception_b7_7x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_7x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_7x1_2\"\n top: \"inception_b7_7x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_7x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_7x1_2\"\n top: \"inception_b7_7x1_2\"\n}\nlayer {\n name: \"inception_b7_1x7_3\"\n type: \"Convolution\"\n bottom: \"inception_b7_7x1_2\"\n top: \"inception_b7_1x7_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_b7_1x7_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_1x7_3\"\n top: \"inception_b7_1x7_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_1x7_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_1x7_3\"\n top: \"inception_b7_1x7_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_1x7_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_1x7_3\"\n top: \"inception_b7_1x7_3\"\n}\nlayer {\n name: \"inception_b7_7x1_3\"\n type: \"Convolution\"\n bottom: \"inception_b7_1x7_3\"\n top: \"inception_b7_7x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_b7_7x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_b7_7x1_3\"\n top: \"inception_b7_7x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_b7_7x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_b7_7x1_3\"\n top: \"inception_b7_7x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_b7_7x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_b7_7x1_3\"\n top: \"inception_b7_7x1_3\"\n}\nlayer {\n name: \"inception_b7_concat\"\n type: \"Concat\"\n bottom: \"inception_b7_1x1\"\n bottom: \"inception_b7_1x1_2\"\n bottom: \"inception_b7_7x1\"\n bottom: \"inception_b7_7x1_3\"\n top: \"inception_b7_concat\"\n}\nlayer {\n name: \"reduction_b_pool\"\n type: \"Pooling\"\n bottom: \"inception_b7_concat\"\n top: \"reduction_b_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b7_concat\"\n top: \"reduction_b_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3_reduce\"\n}\nlayer {\n name: \"reduction_b_3x3\"\n type: \"Convolution\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3\"\n top: \"reduction_b_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3\"\n top: \"reduction_b_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3\"\n top: \"reduction_b_3x3\"\n}\nlayer {\n name: \"reduction_b_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_b7_concat\"\n top: \"reduction_b_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_1x7_reduce\"\n top: \"reduction_b_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_1x7_reduce\"\n top: \"reduction_b_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_1x7_reduce\"\n top: \"reduction_b_1x7_reduce\"\n}\nlayer {\n name: \"reduction_b_1x7\"\n type: \"Convolution\"\n bottom: \"reduction_b_1x7_reduce\"\n top: \"reduction_b_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"reduction_b_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_1x7\"\n top: \"reduction_b_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_1x7_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_1x7\"\n top: \"reduction_b_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_1x7_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_1x7\"\n top: \"reduction_b_1x7\"\n}\nlayer {\n name: \"reduction_b_7x1\"\n type: \"Convolution\"\n bottom: \"reduction_b_1x7\"\n top: \"reduction_b_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 320\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"reduction_b_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_7x1\"\n top: \"reduction_b_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_7x1_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_7x1\"\n top: \"reduction_b_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_7x1_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_7x1\"\n top: \"reduction_b_7x1\"\n}\nlayer {\n name: \"reduction_b_3x3_2\"\n type: \"Convolution\"\n bottom: \"reduction_b_7x1\"\n top: \"reduction_b_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 320\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_2\"\n top: \"reduction_b_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_2\"\n top: \"reduction_b_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_2\"\n top: \"reduction_b_3x3_2\"\n}\nlayer {\n name: \"reduction_b_concat\"\n type: \"Concat\"\n bottom: \"reduction_b_pool\"\n bottom: \"reduction_b_3x3\"\n bottom: \"reduction_b_3x3_2\"\n top: \"reduction_b_concat\"\n}\nlayer {\n name: \"inception_c1_pool_ave\"\n type: \"Pooling\"\n bottom: \"reduction_b_concat\"\n top: \"inception_c1_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_c1_1x1\"\n type: \"Convolution\"\n bottom: \"inception_c1_pool_ave\"\n top: \"inception_c1_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c1_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_1x1\"\n top: \"inception_c1_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_1x1\"\n top: \"inception_c1_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_1x1\"\n top: \"inception_c1_1x1\"\n}\nlayer {\n name: \"inception_c1_1x1_2\"\n type: \"Convolution\"\n bottom: \"reduction_b_concat\"\n top: \"inception_c1_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c1_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_1x1_2\"\n top: \"inception_c1_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_1x1_2\"\n top: \"inception_c1_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_1x1_2\"\n top: \"inception_c1_1x1_2\"\n}\nlayer {\n name: \"inception_c1_1x1_3\"\n type: \"Convolution\"\n bottom: \"reduction_b_concat\"\n top: \"inception_c1_1x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c1_1x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_1x1_3\"\n top: \"inception_c1_1x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_1x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_1x1_3\"\n top: \"inception_c1_1x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_1x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_1x1_3\"\n top: \"inception_c1_1x1_3\"\n}\nlayer {\n name: \"inception_c1_1x3\"\n type: \"Convolution\"\n bottom: \"inception_c1_1x1_3\"\n top: \"inception_c1_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c1_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_1x3\"\n top: \"inception_c1_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_1x3\"\n top: \"inception_c1_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_1x3\"\n top: \"inception_c1_1x3\"\n}\nlayer {\n name: \"inception_c1_3x1\"\n type: \"Convolution\"\n bottom: \"inception_c1_1x1_3\"\n top: \"inception_c1_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c1_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_3x1\"\n top: \"inception_c1_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_3x1\"\n top: \"inception_c1_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_3x1\"\n top: \"inception_c1_3x1\"\n}\nlayer {\n name: \"inception_c1_1x1_4\"\n type: \"Convolution\"\n bottom: \"reduction_b_concat\"\n top: \"inception_c1_1x1_4\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c1_1x1_4_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_1x1_4\"\n top: \"inception_c1_1x1_4\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_1x1_4_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_1x1_4\"\n top: \"inception_c1_1x1_4\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_1x1_4_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_1x1_4\"\n top: \"inception_c1_1x1_4\"\n}\nlayer {\n name: \"inception_c1_1x3_2\"\n type: \"Convolution\"\n bottom: \"inception_c1_1x1_4\"\n top: \"inception_c1_1x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 448\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c1_1x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_1x3_2\"\n top: \"inception_c1_1x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_1x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_1x3_2\"\n top: \"inception_c1_1x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_1x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_1x3_2\"\n top: \"inception_c1_1x3_2\"\n}\nlayer {\n name: \"inception_c1_3x1_2\"\n type: \"Convolution\"\n bottom: \"inception_c1_1x3_2\"\n top: \"inception_c1_3x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c1_3x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_3x1_2\"\n top: \"inception_c1_3x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_3x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_3x1_2\"\n top: \"inception_c1_3x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_3x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_3x1_2\"\n top: \"inception_c1_3x1_2\"\n}\nlayer {\n name: \"inception_c1_1x3_3\"\n type: \"Convolution\"\n bottom: \"inception_c1_3x1_2\"\n top: \"inception_c1_1x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c1_1x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_1x3_3\"\n top: \"inception_c1_1x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_1x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_1x3_3\"\n top: \"inception_c1_1x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_1x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_1x3_3\"\n top: \"inception_c1_1x3_3\"\n}\nlayer {\n name: \"inception_c1_3x1_3\"\n type: \"Convolution\"\n bottom: \"inception_c1_3x1_2\"\n top: \"inception_c1_3x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c1_3x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c1_3x1_3\"\n top: \"inception_c1_3x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c1_3x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c1_3x1_3\"\n top: \"inception_c1_3x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c1_3x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c1_3x1_3\"\n top: \"inception_c1_3x1_3\"\n}\nlayer {\n name: \"inception_c1_concat\"\n type: \"Concat\"\n bottom: \"inception_c1_1x1\"\n bottom: \"inception_c1_1x1_2\"\n bottom: \"inception_c1_1x3\"\n bottom: \"inception_c1_3x1\"\n bottom: \"inception_c1_1x3_3\"\n bottom: \"inception_c1_3x1_3\"\n top: \"inception_c1_concat\"\n}\nlayer {\n name: \"inception_c2_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_c1_concat\"\n top: \"inception_c2_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_c2_1x1\"\n type: \"Convolution\"\n bottom: \"inception_c2_pool_ave\"\n top: \"inception_c2_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c2_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_1x1\"\n top: \"inception_c2_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_1x1\"\n top: \"inception_c2_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_1x1\"\n top: \"inception_c2_1x1\"\n}\nlayer {\n name: \"inception_c2_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_c1_concat\"\n top: \"inception_c2_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c2_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_1x1_2\"\n top: \"inception_c2_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_1x1_2\"\n top: \"inception_c2_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_1x1_2\"\n top: \"inception_c2_1x1_2\"\n}\nlayer {\n name: \"inception_c2_1x1_3\"\n type: \"Convolution\"\n bottom: \"inception_c1_concat\"\n top: \"inception_c2_1x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c2_1x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_1x1_3\"\n top: \"inception_c2_1x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_1x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_1x1_3\"\n top: \"inception_c2_1x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_1x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_1x1_3\"\n top: \"inception_c2_1x1_3\"\n}\nlayer {\n name: \"inception_c2_1x3\"\n type: \"Convolution\"\n bottom: \"inception_c2_1x1_3\"\n top: \"inception_c2_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c2_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_1x3\"\n top: \"inception_c2_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_1x3\"\n top: \"inception_c2_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_1x3\"\n top: \"inception_c2_1x3\"\n}\nlayer {\n name: \"inception_c2_3x1\"\n type: \"Convolution\"\n bottom: \"inception_c2_1x1_3\"\n top: \"inception_c2_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c2_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_3x1\"\n top: \"inception_c2_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_3x1\"\n top: \"inception_c2_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_3x1\"\n top: \"inception_c2_3x1\"\n}\nlayer {\n name: \"inception_c2_1x1_4\"\n type: \"Convolution\"\n bottom: \"inception_c1_concat\"\n top: \"inception_c2_1x1_4\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c2_1x1_4_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_1x1_4\"\n top: \"inception_c2_1x1_4\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_1x1_4_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_1x1_4\"\n top: \"inception_c2_1x1_4\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_1x1_4_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_1x1_4\"\n top: \"inception_c2_1x1_4\"\n}\nlayer {\n name: \"inception_c2_1x3_2\"\n type: \"Convolution\"\n bottom: \"inception_c2_1x1_4\"\n top: \"inception_c2_1x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 448\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c2_1x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_1x3_2\"\n top: \"inception_c2_1x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_1x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_1x3_2\"\n top: \"inception_c2_1x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_1x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_1x3_2\"\n top: \"inception_c2_1x3_2\"\n}\nlayer {\n name: \"inception_c2_3x1_2\"\n type: \"Convolution\"\n bottom: \"inception_c2_1x3_2\"\n top: \"inception_c2_3x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c2_3x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_3x1_2\"\n top: \"inception_c2_3x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_3x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_3x1_2\"\n top: \"inception_c2_3x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_3x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_3x1_2\"\n top: \"inception_c2_3x1_2\"\n}\nlayer {\n name: \"inception_c2_1x3_3\"\n type: \"Convolution\"\n bottom: \"inception_c2_3x1_2\"\n top: \"inception_c2_1x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c2_1x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_1x3_3\"\n top: \"inception_c2_1x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_1x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_1x3_3\"\n top: \"inception_c2_1x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_1x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_1x3_3\"\n top: \"inception_c2_1x3_3\"\n}\nlayer {\n name: \"inception_c2_3x1_3\"\n type: \"Convolution\"\n bottom: \"inception_c2_3x1_2\"\n top: \"inception_c2_3x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c2_3x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c2_3x1_3\"\n top: \"inception_c2_3x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c2_3x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c2_3x1_3\"\n top: \"inception_c2_3x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c2_3x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c2_3x1_3\"\n top: \"inception_c2_3x1_3\"\n}\nlayer {\n name: \"inception_c2_concat\"\n type: \"Concat\"\n bottom: \"inception_c2_1x1\"\n bottom: \"inception_c2_1x1_2\"\n bottom: \"inception_c2_1x3\"\n bottom: \"inception_c2_3x1\"\n bottom: \"inception_c2_1x3_3\"\n bottom: \"inception_c2_3x1_3\"\n top: \"inception_c2_concat\"\n}\nlayer {\n name: \"inception_c3_pool_ave\"\n type: \"Pooling\"\n bottom: \"inception_c2_concat\"\n top: \"inception_c3_pool_ave\"\n pooling_param {\n pool: AVE\n kernel_size: 3\n stride: 1\n pad: 1\n }\n}\nlayer {\n name: \"inception_c3_1x1\"\n type: \"Convolution\"\n bottom: \"inception_c3_pool_ave\"\n top: \"inception_c3_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c3_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_1x1\"\n top: \"inception_c3_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_1x1\"\n top: \"inception_c3_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_1x1\"\n top: \"inception_c3_1x1\"\n}\nlayer {\n name: \"inception_c3_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_c2_concat\"\n top: \"inception_c3_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c3_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_1x1_2\"\n top: \"inception_c3_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_1x1_2\"\n top: \"inception_c3_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_1x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_1x1_2\"\n top: \"inception_c3_1x1_2\"\n}\nlayer {\n name: \"inception_c3_1x1_3\"\n type: \"Convolution\"\n bottom: \"inception_c2_concat\"\n top: \"inception_c3_1x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c3_1x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_1x1_3\"\n top: \"inception_c3_1x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_1x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_1x1_3\"\n top: \"inception_c3_1x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_1x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_1x1_3\"\n top: \"inception_c3_1x1_3\"\n}\nlayer {\n name: \"inception_c3_1x3\"\n type: \"Convolution\"\n bottom: \"inception_c3_1x1_3\"\n top: \"inception_c3_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c3_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_1x3\"\n top: \"inception_c3_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_1x3\"\n top: \"inception_c3_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_1x3\"\n top: \"inception_c3_1x3\"\n}\nlayer {\n name: \"inception_c3_3x1\"\n type: \"Convolution\"\n bottom: \"inception_c3_1x1_3\"\n top: \"inception_c3_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c3_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_3x1\"\n top: \"inception_c3_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_3x1\"\n top: \"inception_c3_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_3x1\"\n top: \"inception_c3_3x1\"\n}\nlayer {\n name: \"inception_c3_1x1_4\"\n type: \"Convolution\"\n bottom: \"inception_c2_concat\"\n top: \"inception_c3_1x1_4\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_c3_1x1_4_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_1x1_4\"\n top: \"inception_c3_1x1_4\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_1x1_4_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_1x1_4\"\n top: \"inception_c3_1x1_4\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_1x1_4_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_1x1_4\"\n top: \"inception_c3_1x1_4\"\n}\nlayer {\n name: \"inception_c3_1x3_2\"\n type: \"Convolution\"\n bottom: \"inception_c3_1x1_4\"\n top: \"inception_c3_1x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 448\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c3_1x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_1x3_2\"\n top: \"inception_c3_1x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_1x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_1x3_2\"\n top: \"inception_c3_1x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_1x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_1x3_2\"\n top: \"inception_c3_1x3_2\"\n}\nlayer {\n name: \"inception_c3_3x1_2\"\n type: \"Convolution\"\n bottom: \"inception_c3_1x3_2\"\n top: \"inception_c3_3x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 512\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c3_3x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_3x1_2\"\n top: \"inception_c3_3x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_3x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_3x1_2\"\n top: \"inception_c3_3x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_3x1_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_3x1_2\"\n top: \"inception_c3_3x1_2\"\n}\nlayer {\n name: \"inception_c3_1x3_3\"\n type: \"Convolution\"\n bottom: \"inception_c3_3x1_2\"\n top: \"inception_c3_1x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_c3_1x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_1x3_3\"\n top: \"inception_c3_1x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_1x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_1x3_3\"\n top: \"inception_c3_1x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_1x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_1x3_3\"\n top: \"inception_c3_1x3_3\"\n}\nlayer {\n name: \"inception_c3_3x1_3\"\n type: \"Convolution\"\n bottom: \"inception_c3_3x1_2\"\n top: \"inception_c3_3x1_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_c3_3x1_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_c3_3x1_3\"\n top: \"inception_c3_3x1_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_c3_3x1_3_scale\"\n type: \"Scale\"\n bottom: \"inception_c3_3x1_3\"\n top: \"inception_c3_3x1_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_c3_3x1_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_c3_3x1_3\"\n top: \"inception_c3_3x1_3\"\n}\nlayer {\n name: \"inception_c3_concat\"\n type: \"Concat\"\n bottom: \"inception_c3_1x1\"\n bottom: \"inception_c3_1x1_2\"\n bottom: \"inception_c3_1x3\"\n bottom: \"inception_c3_3x1\"\n bottom: \"inception_c3_1x3_3\"\n bottom: \"inception_c3_3x1_3\"\n top: \"inception_c3_concat\"\n}\nlayer {\n name: \"pool_8x8_s1\"\n type: \"Pooling\"\n bottom: \"inception_c3_concat\"\n top: \"pool_8x8_s1\"\n pooling_param {\n pool: AVE\n global_pooling: true\n }\n}\nlayer {\n name: \"pool_8x8_s1_drop\"\n type: \"Dropout\"\n bottom: \"pool_8x8_s1\"\n top: \"pool_8x8_s1_drop\"\n dropout_param {\n dropout_ratio: 0.2\n }\n}\nlayer {\n name: \"classifier\"\n type: \"InnerProduct\"\n bottom: \"pool_8x8_s1_drop\"\n top: \"classifier\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 1000\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"classifier\"\n top: \"loss\"\n}" }, { "path": "presets/inceptionv4_resnet.prototxt", "content": "#downloaded from http://github.com/soeaver/caffe-model\nname: \"inception_resnet_v2\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n include {\n phase: TRAIN\n }\n transform_param {\n mirror: true\n crop_size: 299\n mean_value: 104\n mean_value: 117\n mean_value: 123\n }\n data_param {\n source: \"examples/imagenet/ilsvrc12_train_lmdb\"\n batch_size: 64\n backend: LMDB\n }\n}\n\nlayer {\n name: \"conv1_3x3_s2\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1_3x3_s2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv1_3x3_s2_bn\"\n type: \"BatchNorm\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv1_3x3_s2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"conv1_3x3_s2_scale\"\n type: \"Scale\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv1_3x3_s2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"conv1_3x3_s2_relu\"\n type: \"ReLU\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv1_3x3_s2\"\n}\nlayer {\n name: \"conv2_3x3_s1\"\n type: \"Convolution\"\n bottom: \"conv1_3x3_s2\"\n top: \"conv2_3x3_s1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv2_3x3_s1_bn\"\n type: \"BatchNorm\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv2_3x3_s1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"conv2_3x3_s1_scale\"\n type: \"Scale\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv2_3x3_s1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"conv2_3x3_s1_relu\"\n type: \"ReLU\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv2_3x3_s1\"\n}\nlayer {\n name: \"conv3_3x3_s1\"\n type: \"Convolution\"\n bottom: \"conv2_3x3_s1\"\n top: \"conv3_3x3_s1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"conv3_3x3_s1_bn\"\n type: \"BatchNorm\"\n bottom: \"conv3_3x3_s1\"\n top: \"conv3_3x3_s1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"conv3_3x3_s1_scale\"\n type: \"Scale\"\n bottom: \"conv3_3x3_s1\"\n top: \"conv3_3x3_s1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"conv3_3x3_s1_relu\"\n type: \"ReLU\"\n bottom: \"conv3_3x3_s1\"\n top: \"conv3_3x3_s1\"\n}\nlayer {\n name: \"inception_stem1_3x3_s2\"\n type: \"Convolution\"\n bottom: \"conv3_3x3_s1\"\n top: \"inception_stem1_3x3_s2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem1_3x3_s2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem1_3x3_s2\"\n top: \"inception_stem1_3x3_s2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem1_3x3_s2_scale\"\n type: \"Scale\"\n bottom: \"inception_stem1_3x3_s2\"\n top: \"inception_stem1_3x3_s2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem1_3x3_s2_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem1_3x3_s2\"\n top: \"inception_stem1_3x3_s2\"\n}\nlayer {\n name: \"inception_stem1_pool\"\n type: \"Pooling\"\n bottom: \"conv3_3x3_s1\"\n top: \"inception_stem1_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"inception_stem1\"\n type: \"Concat\"\n bottom: \"inception_stem1_3x3_s2\"\n bottom: \"inception_stem1_pool\"\n top: \"inception_stem1\"\n}\nlayer {\n name: \"inception_stem2_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem1\"\n top: \"inception_stem2_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3_reduce\"\n}\nlayer {\n name: \"inception_stem2_3x3\"\n type: \"Convolution\"\n bottom: \"inception_stem2_3x3_reduce\"\n top: \"inception_stem2_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_3x3\"\n top: \"inception_stem2_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_3x3\"\n top: \"inception_stem2_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_3x3\"\n top: \"inception_stem2_3x3\"\n}\nlayer {\n name: \"inception_stem2_7x1_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem1\"\n top: \"inception_stem2_7x1_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_7x1_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_7x1_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_7x1_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1_reduce\"\n}\nlayer {\n name: \"inception_stem2_7x1\"\n type: \"Convolution\"\n bottom: \"inception_stem2_7x1_reduce\"\n top: \"inception_stem2_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_stem2_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_7x1\"\n}\nlayer {\n name: \"inception_stem2_1x7\"\n type: \"Convolution\"\n bottom: \"inception_stem2_7x1\"\n top: \"inception_stem2_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_stem2_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_1x7\"\n}\nlayer {\n name: \"inception_stem2_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_stem2_1x7\"\n top: \"inception_stem2_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 96\n pad: 0\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem2_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem2_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem2_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2_3x3_2\"\n}\nlayer {\n name: \"inception_stem2\"\n type: \"Concat\"\n bottom: \"inception_stem2_3x3\"\n bottom: \"inception_stem2_3x3_2\"\n top: \"inception_stem2\"\n}\nlayer {\n name: \"inception_stem3_3x3_s2\"\n type: \"Convolution\"\n bottom: \"inception_stem2\"\n top: \"inception_stem3_3x3_s2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_stem3_3x3_s2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_stem3_3x3_s2\"\n top: \"inception_stem3_3x3_s2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_stem3_3x3_s2_scale\"\n type: \"Scale\"\n bottom: \"inception_stem3_3x3_s2\"\n top: \"inception_stem3_3x3_s2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_stem3_3x3_s2_relu\"\n type: \"ReLU\"\n bottom: \"inception_stem3_3x3_s2\"\n top: \"inception_stem3_3x3_s2\"\n}\nlayer {\n name: \"inception_stem3_pool\"\n type: \"Pooling\"\n bottom: \"inception_stem2\"\n top: \"inception_stem3_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"inception_stem3\"\n type: \"Concat\"\n bottom: \"inception_stem3_3x3_s2\"\n bottom: \"inception_stem3_pool\"\n top: \"inception_stem3\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_1x1\"\n type: \"Convolution\"\n bottom: \"inception_stem3\"\n top: \"inception_resnet_v2_a1_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a1_1x1\"\n top: \"inception_resnet_v2_a1_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a1_1x1\"\n top: \"inception_resnet_v2_a1_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a1_1x1\"\n top: \"inception_resnet_v2_a1_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem3\"\n top: \"inception_resnet_v2_a1_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a1_3x3_reduce\"\n top: \"inception_resnet_v2_a1_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a1_3x3_reduce\"\n top: \"inception_resnet_v2_a1_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a1_3x3_reduce\"\n top: \"inception_resnet_v2_a1_3x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a1_3x3_reduce\"\n top: \"inception_resnet_v2_a1_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a1_3x3\"\n top: \"inception_resnet_v2_a1_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a1_3x3\"\n top: \"inception_resnet_v2_a1_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a1_3x3\"\n top: \"inception_resnet_v2_a1_3x3\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_stem3\"\n top: \"inception_resnet_v2_a1_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a1_3x3_2_reduce\"\n top: \"inception_resnet_v2_a1_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a1_3x3_2_reduce\"\n top: \"inception_resnet_v2_a1_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a1_3x3_2_reduce\"\n top: \"inception_resnet_v2_a1_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a1_3x3_2_reduce\"\n top: \"inception_resnet_v2_a1_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 48\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a1_3x3_2\"\n top: \"inception_resnet_v2_a1_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a1_3x3_2\"\n top: \"inception_resnet_v2_a1_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a1_3x3_2\"\n top: \"inception_resnet_v2_a1_3x3_2\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a1_3x3_2\"\n top: \"inception_resnet_v2_a1_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a1_3x3_3\"\n top: \"inception_resnet_v2_a1_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a1_3x3_3\"\n top: \"inception_resnet_v2_a1_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a1_3x3_3\"\n top: \"inception_resnet_v2_a1_3x3_3\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_a1_1x1\"\n bottom: \"inception_resnet_v2_a1_3x3\"\n bottom: \"inception_resnet_v2_a1_3x3_3\"\n top: \"inception_resnet_v2_a1_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_a1_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a1_concat\"\n top: \"inception_resnet_v2_a1_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a1_1x1_2\"\n top: \"inception_resnet_v2_a1_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a1_1x1_2\"\n top: \"inception_resnet_v2_a1_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a1_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_stem3\"\n bottom: \"inception_resnet_v2_a1_1x1_2\"\n top: \"inception_resnet_v2_a1_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a1_residual_eltwise\"\n top: \"inception_resnet_v2_a2_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a2_1x1\"\n top: \"inception_resnet_v2_a2_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a2_1x1\"\n top: \"inception_resnet_v2_a2_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a2_1x1\"\n top: \"inception_resnet_v2_a2_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a1_residual_eltwise\"\n top: \"inception_resnet_v2_a2_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a2_3x3_reduce\"\n top: \"inception_resnet_v2_a2_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a2_3x3_reduce\"\n top: \"inception_resnet_v2_a2_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a2_3x3_reduce\"\n top: \"inception_resnet_v2_a2_3x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a2_3x3_reduce\"\n top: \"inception_resnet_v2_a2_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a2_3x3\"\n top: \"inception_resnet_v2_a2_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a2_3x3\"\n top: \"inception_resnet_v2_a2_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a2_3x3\"\n top: \"inception_resnet_v2_a2_3x3\"\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a1_residual_eltwise\"\n top: \"inception_resnet_v2_a2_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a2_3x3_2_reduce\"\n top: \"inception_resnet_v2_a2_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a2_3x3_2_reduce\"\n top: \"inception_resnet_v2_a2_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a2_3x3_2_reduce\"\n top: \"inception_resnet_v2_a2_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a2_3x3_2_reduce\"\n top: \"inception_resnet_v2_a2_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 48\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a2_3x3_2\"\n top: \"inception_resnet_v2_a2_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a2_3x3_2\"\n top: \"inception_resnet_v2_a2_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a2_3x3_2\"\n top: \"inception_resnet_v2_a2_3x3_2\"\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a2_3x3_2\"\n top: \"inception_resnet_v2_a2_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a2_3x3_3\"\n top: \"inception_resnet_v2_a2_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a2_3x3_3\"\n top: \"inception_resnet_v2_a2_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a2_3x3_3\"\n top: \"inception_resnet_v2_a2_3x3_3\"\n}\nlayer {\n name: \"inception_resnet_v2_a2_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_a2_1x1\"\n bottom: \"inception_resnet_v2_a2_3x3\"\n bottom: \"inception_resnet_v2_a2_3x3_3\"\n top: \"inception_resnet_v2_a2_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_a2_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a2_concat\"\n top: \"inception_resnet_v2_a2_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a2_1x1_2\"\n top: \"inception_resnet_v2_a2_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a2_1x1_2\"\n top: \"inception_resnet_v2_a2_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a2_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_a1_residual_eltwise\"\n bottom: \"inception_resnet_v2_a2_1x1_2\"\n top: \"inception_resnet_v2_a2_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a2_residual_eltwise\"\n top: \"inception_resnet_v2_a3_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a3_1x1\"\n top: \"inception_resnet_v2_a3_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a3_1x1\"\n top: \"inception_resnet_v2_a3_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a3_1x1\"\n top: \"inception_resnet_v2_a3_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a2_residual_eltwise\"\n top: \"inception_resnet_v2_a3_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a3_3x3_reduce\"\n top: \"inception_resnet_v2_a3_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a3_3x3_reduce\"\n top: \"inception_resnet_v2_a3_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a3_3x3_reduce\"\n top: \"inception_resnet_v2_a3_3x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a3_3x3_reduce\"\n top: \"inception_resnet_v2_a3_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a3_3x3\"\n top: \"inception_resnet_v2_a3_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a3_3x3\"\n top: \"inception_resnet_v2_a3_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a3_3x3\"\n top: \"inception_resnet_v2_a3_3x3\"\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a2_residual_eltwise\"\n top: \"inception_resnet_v2_a3_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a3_3x3_2_reduce\"\n top: \"inception_resnet_v2_a3_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a3_3x3_2_reduce\"\n top: \"inception_resnet_v2_a3_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a3_3x3_2_reduce\"\n top: \"inception_resnet_v2_a3_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a3_3x3_2_reduce\"\n top: \"inception_resnet_v2_a3_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 48\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a3_3x3_2\"\n top: \"inception_resnet_v2_a3_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a3_3x3_2\"\n top: \"inception_resnet_v2_a3_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a3_3x3_2\"\n top: \"inception_resnet_v2_a3_3x3_2\"\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a3_3x3_2\"\n top: \"inception_resnet_v2_a3_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a3_3x3_3\"\n top: \"inception_resnet_v2_a3_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a3_3x3_3\"\n top: \"inception_resnet_v2_a3_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a3_3x3_3\"\n top: \"inception_resnet_v2_a3_3x3_3\"\n}\nlayer {\n name: \"inception_resnet_v2_a3_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_a3_1x1\"\n bottom: \"inception_resnet_v2_a3_3x3\"\n bottom: \"inception_resnet_v2_a3_3x3_3\"\n top: \"inception_resnet_v2_a3_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_a3_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a3_concat\"\n top: \"inception_resnet_v2_a3_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a3_1x1_2\"\n top: \"inception_resnet_v2_a3_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a3_1x1_2\"\n top: \"inception_resnet_v2_a3_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a3_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_a2_residual_eltwise\"\n bottom: \"inception_resnet_v2_a3_1x1_2\"\n top: \"inception_resnet_v2_a3_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a3_residual_eltwise\"\n top: \"inception_resnet_v2_a4_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a4_1x1\"\n top: \"inception_resnet_v2_a4_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a4_1x1\"\n top: \"inception_resnet_v2_a4_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a4_1x1\"\n top: \"inception_resnet_v2_a4_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a3_residual_eltwise\"\n top: \"inception_resnet_v2_a4_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a4_3x3_reduce\"\n top: \"inception_resnet_v2_a4_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a4_3x3_reduce\"\n top: \"inception_resnet_v2_a4_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a4_3x3_reduce\"\n top: \"inception_resnet_v2_a4_3x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a4_3x3_reduce\"\n top: \"inception_resnet_v2_a4_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a4_3x3\"\n top: \"inception_resnet_v2_a4_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a4_3x3\"\n top: \"inception_resnet_v2_a4_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a4_3x3\"\n top: \"inception_resnet_v2_a4_3x3\"\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a3_residual_eltwise\"\n top: \"inception_resnet_v2_a4_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a4_3x3_2_reduce\"\n top: \"inception_resnet_v2_a4_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a4_3x3_2_reduce\"\n top: \"inception_resnet_v2_a4_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a4_3x3_2_reduce\"\n top: \"inception_resnet_v2_a4_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a4_3x3_2_reduce\"\n top: \"inception_resnet_v2_a4_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 48\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a4_3x3_2\"\n top: \"inception_resnet_v2_a4_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a4_3x3_2\"\n top: \"inception_resnet_v2_a4_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a4_3x3_2\"\n top: \"inception_resnet_v2_a4_3x3_2\"\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a4_3x3_2\"\n top: \"inception_resnet_v2_a4_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a4_3x3_3\"\n top: \"inception_resnet_v2_a4_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a4_3x3_3\"\n top: \"inception_resnet_v2_a4_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a4_3x3_3\"\n top: \"inception_resnet_v2_a4_3x3_3\"\n}\nlayer {\n name: \"inception_resnet_v2_a4_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_a4_1x1\"\n bottom: \"inception_resnet_v2_a4_3x3\"\n bottom: \"inception_resnet_v2_a4_3x3_3\"\n top: \"inception_resnet_v2_a4_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_a4_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a4_concat\"\n top: \"inception_resnet_v2_a4_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a4_1x1_2\"\n top: \"inception_resnet_v2_a4_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a4_1x1_2\"\n top: \"inception_resnet_v2_a4_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a4_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_a3_residual_eltwise\"\n bottom: \"inception_resnet_v2_a4_1x1_2\"\n top: \"inception_resnet_v2_a4_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a4_residual_eltwise\"\n top: \"inception_resnet_v2_a5_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a5_1x1\"\n top: \"inception_resnet_v2_a5_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a5_1x1\"\n top: \"inception_resnet_v2_a5_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a5_1x1\"\n top: \"inception_resnet_v2_a5_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a4_residual_eltwise\"\n top: \"inception_resnet_v2_a5_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a5_3x3_reduce\"\n top: \"inception_resnet_v2_a5_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a5_3x3_reduce\"\n top: \"inception_resnet_v2_a5_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a5_3x3_reduce\"\n top: \"inception_resnet_v2_a5_3x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a5_3x3_reduce\"\n top: \"inception_resnet_v2_a5_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a5_3x3\"\n top: \"inception_resnet_v2_a5_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a5_3x3\"\n top: \"inception_resnet_v2_a5_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a5_3x3\"\n top: \"inception_resnet_v2_a5_3x3\"\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a4_residual_eltwise\"\n top: \"inception_resnet_v2_a5_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 32\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a5_3x3_2_reduce\"\n top: \"inception_resnet_v2_a5_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a5_3x3_2_reduce\"\n top: \"inception_resnet_v2_a5_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a5_3x3_2_reduce\"\n top: \"inception_resnet_v2_a5_3x3_2_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a5_3x3_2_reduce\"\n top: \"inception_resnet_v2_a5_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 48\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a5_3x3_2\"\n top: \"inception_resnet_v2_a5_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a5_3x3_2\"\n top: \"inception_resnet_v2_a5_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a5_3x3_2\"\n top: \"inception_resnet_v2_a5_3x3_2\"\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a5_3x3_2\"\n top: \"inception_resnet_v2_a5_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a5_3x3_3\"\n top: \"inception_resnet_v2_a5_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a5_3x3_3\"\n top: \"inception_resnet_v2_a5_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_a5_3x3_3\"\n top: \"inception_resnet_v2_a5_3x3_3\"\n}\nlayer {\n name: \"inception_resnet_v2_a5_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_a5_1x1\"\n bottom: \"inception_resnet_v2_a5_3x3\"\n bottom: \"inception_resnet_v2_a5_3x3_3\"\n top: \"inception_resnet_v2_a5_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_a5_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a5_concat\"\n top: \"inception_resnet_v2_a5_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_a5_1x1_2\"\n top: \"inception_resnet_v2_a5_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_a5_1x1_2\"\n top: \"inception_resnet_v2_a5_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_a5_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_a4_residual_eltwise\"\n bottom: \"inception_resnet_v2_a5_1x1_2\"\n top: \"inception_resnet_v2_a5_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"reduction_a_pool\"\n type: \"Pooling\"\n bottom: \"inception_resnet_v2_a5_residual_eltwise\"\n top: \"reduction_a_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"reduction_a_3x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a5_residual_eltwise\"\n top: \"reduction_a_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3\"\n top: \"reduction_a_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3\"\n top: \"reduction_a_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3\"\n top: \"reduction_a_3x3\"\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_a5_residual_eltwise\"\n top: \"reduction_a_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2_reduce\"\n}\nlayer {\n name: \"reduction_a_3x3_2\"\n type: \"Convolution\"\n bottom: \"reduction_a_3x3_2_reduce\"\n top: \"reduction_a_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_2\"\n}\nlayer {\n name: \"reduction_a_3x3_3\"\n type: \"Convolution\"\n bottom: \"reduction_a_3x3_2\"\n top: \"reduction_a_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_a_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_a_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_a_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_3x3_3\"\n}\nlayer {\n name: \"reduction_a_concat\"\n type: \"Concat\"\n bottom: \"reduction_a_pool\"\n bottom: \"reduction_a_3x3\"\n bottom: \"reduction_a_3x3_3\"\n top: \"reduction_a_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x1\"\n type: \"Convolution\"\n bottom: \"reduction_a_concat\"\n top: \"inception_resnet_v2_b1_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b1_1x1\"\n top: \"inception_resnet_v2_b1_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b1_1x1\"\n top: \"inception_resnet_v2_b1_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b1_1x1\"\n top: \"inception_resnet_v2_b1_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"reduction_a_concat\"\n top: \"inception_resnet_v2_b1_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b1_1x7_reduce\"\n top: \"inception_resnet_v2_b1_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b1_1x7_reduce\"\n top: \"inception_resnet_v2_b1_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b1_1x7_reduce\"\n top: \"inception_resnet_v2_b1_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b1_1x7_reduce\"\n top: \"inception_resnet_v2_b1_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b1_1x7\"\n top: \"inception_resnet_v2_b1_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b1_1x7\"\n top: \"inception_resnet_v2_b1_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b1_1x7\"\n top: \"inception_resnet_v2_b1_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b1_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b1_1x7\"\n top: \"inception_resnet_v2_b1_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b1_7x1\"\n top: \"inception_resnet_v2_b1_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b1_7x1\"\n top: \"inception_resnet_v2_b1_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b1_7x1\"\n top: \"inception_resnet_v2_b1_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b1_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b1_1x1\"\n bottom: \"inception_resnet_v2_b1_7x1\"\n top: \"inception_resnet_v2_b1_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b1_concat\"\n top: \"inception_resnet_v2_b1_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b1_1x1_2\"\n top: \"inception_resnet_v2_b1_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b1_1x1_2\"\n top: \"inception_resnet_v2_b1_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b1_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"reduction_a_concat\"\n bottom: \"inception_resnet_v2_b1_1x1_2\"\n top: \"inception_resnet_v2_b1_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b1_residual_eltwise\"\n top: \"inception_resnet_v2_b2_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b2_1x1\"\n top: \"inception_resnet_v2_b2_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b2_1x1\"\n top: \"inception_resnet_v2_b2_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b2_1x1\"\n top: \"inception_resnet_v2_b2_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b1_residual_eltwise\"\n top: \"inception_resnet_v2_b2_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b2_1x7_reduce\"\n top: \"inception_resnet_v2_b2_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b2_1x7_reduce\"\n top: \"inception_resnet_v2_b2_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b2_1x7_reduce\"\n top: \"inception_resnet_v2_b2_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b2_1x7_reduce\"\n top: \"inception_resnet_v2_b2_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b2_1x7\"\n top: \"inception_resnet_v2_b2_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b2_1x7\"\n top: \"inception_resnet_v2_b2_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b2_1x7\"\n top: \"inception_resnet_v2_b2_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b2_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b2_1x7\"\n top: \"inception_resnet_v2_b2_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b2_7x1\"\n top: \"inception_resnet_v2_b2_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b2_7x1\"\n top: \"inception_resnet_v2_b2_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b2_7x1\"\n top: \"inception_resnet_v2_b2_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b2_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b2_1x1\"\n bottom: \"inception_resnet_v2_b2_7x1\"\n top: \"inception_resnet_v2_b2_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b2_concat\"\n top: \"inception_resnet_v2_b2_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b2_1x1_2\"\n top: \"inception_resnet_v2_b2_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b2_1x1_2\"\n top: \"inception_resnet_v2_b2_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b2_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b1_residual_eltwise\"\n bottom: \"inception_resnet_v2_b2_1x1_2\"\n top: \"inception_resnet_v2_b2_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b2_residual_eltwise\"\n top: \"inception_resnet_v2_b3_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b3_1x1\"\n top: \"inception_resnet_v2_b3_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b3_1x1\"\n top: \"inception_resnet_v2_b3_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b3_1x1\"\n top: \"inception_resnet_v2_b3_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b2_residual_eltwise\"\n top: \"inception_resnet_v2_b3_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b3_1x7_reduce\"\n top: \"inception_resnet_v2_b3_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b3_1x7_reduce\"\n top: \"inception_resnet_v2_b3_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b3_1x7_reduce\"\n top: \"inception_resnet_v2_b3_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b3_1x7_reduce\"\n top: \"inception_resnet_v2_b3_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b3_1x7\"\n top: \"inception_resnet_v2_b3_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b3_1x7\"\n top: \"inception_resnet_v2_b3_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b3_1x7\"\n top: \"inception_resnet_v2_b3_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b3_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b3_1x7\"\n top: \"inception_resnet_v2_b3_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b3_7x1\"\n top: \"inception_resnet_v2_b3_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b3_7x1\"\n top: \"inception_resnet_v2_b3_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b3_7x1\"\n top: \"inception_resnet_v2_b3_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b3_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b3_1x1\"\n bottom: \"inception_resnet_v2_b3_7x1\"\n top: \"inception_resnet_v2_b3_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b3_concat\"\n top: \"inception_resnet_v2_b3_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b3_1x1_2\"\n top: \"inception_resnet_v2_b3_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b3_1x1_2\"\n top: \"inception_resnet_v2_b3_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b3_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b2_residual_eltwise\"\n bottom: \"inception_resnet_v2_b3_1x1_2\"\n top: \"inception_resnet_v2_b3_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b3_residual_eltwise\"\n top: \"inception_resnet_v2_b4_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b4_1x1\"\n top: \"inception_resnet_v2_b4_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b4_1x1\"\n top: \"inception_resnet_v2_b4_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b4_1x1\"\n top: \"inception_resnet_v2_b4_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b3_residual_eltwise\"\n top: \"inception_resnet_v2_b4_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b4_1x7_reduce\"\n top: \"inception_resnet_v2_b4_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b4_1x7_reduce\"\n top: \"inception_resnet_v2_b4_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b4_1x7_reduce\"\n top: \"inception_resnet_v2_b4_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b4_1x7_reduce\"\n top: \"inception_resnet_v2_b4_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b4_1x7\"\n top: \"inception_resnet_v2_b4_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b4_1x7\"\n top: \"inception_resnet_v2_b4_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b4_1x7\"\n top: \"inception_resnet_v2_b4_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b4_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b4_1x7\"\n top: \"inception_resnet_v2_b4_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b4_7x1\"\n top: \"inception_resnet_v2_b4_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b4_7x1\"\n top: \"inception_resnet_v2_b4_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b4_7x1\"\n top: \"inception_resnet_v2_b4_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b4_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b4_1x1\"\n bottom: \"inception_resnet_v2_b4_7x1\"\n top: \"inception_resnet_v2_b4_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b4_concat\"\n top: \"inception_resnet_v2_b4_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b4_1x1_2\"\n top: \"inception_resnet_v2_b4_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b4_1x1_2\"\n top: \"inception_resnet_v2_b4_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b4_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b3_residual_eltwise\"\n bottom: \"inception_resnet_v2_b4_1x1_2\"\n top: \"inception_resnet_v2_b4_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b4_residual_eltwise\"\n top: \"inception_resnet_v2_b5_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b5_1x1\"\n top: \"inception_resnet_v2_b5_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b5_1x1\"\n top: \"inception_resnet_v2_b5_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b5_1x1\"\n top: \"inception_resnet_v2_b5_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b4_residual_eltwise\"\n top: \"inception_resnet_v2_b5_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b5_1x7_reduce\"\n top: \"inception_resnet_v2_b5_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b5_1x7_reduce\"\n top: \"inception_resnet_v2_b5_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b5_1x7_reduce\"\n top: \"inception_resnet_v2_b5_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b5_1x7_reduce\"\n top: \"inception_resnet_v2_b5_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b5_1x7\"\n top: \"inception_resnet_v2_b5_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b5_1x7\"\n top: \"inception_resnet_v2_b5_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b5_1x7\"\n top: \"inception_resnet_v2_b5_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b5_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b5_1x7\"\n top: \"inception_resnet_v2_b5_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b5_7x1\"\n top: \"inception_resnet_v2_b5_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b5_7x1\"\n top: \"inception_resnet_v2_b5_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b5_7x1\"\n top: \"inception_resnet_v2_b5_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b5_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b5_1x1\"\n bottom: \"inception_resnet_v2_b5_7x1\"\n top: \"inception_resnet_v2_b5_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b5_concat\"\n top: \"inception_resnet_v2_b5_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b5_1x1_2\"\n top: \"inception_resnet_v2_b5_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b5_1x1_2\"\n top: \"inception_resnet_v2_b5_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b5_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b4_residual_eltwise\"\n bottom: \"inception_resnet_v2_b5_1x1_2\"\n top: \"inception_resnet_v2_b5_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b5_residual_eltwise\"\n top: \"inception_resnet_v2_b6_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b6_1x1\"\n top: \"inception_resnet_v2_b6_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b6_1x1\"\n top: \"inception_resnet_v2_b6_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b6_1x1\"\n top: \"inception_resnet_v2_b6_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b5_residual_eltwise\"\n top: \"inception_resnet_v2_b6_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b6_1x7_reduce\"\n top: \"inception_resnet_v2_b6_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b6_1x7_reduce\"\n top: \"inception_resnet_v2_b6_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b6_1x7_reduce\"\n top: \"inception_resnet_v2_b6_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b6_1x7_reduce\"\n top: \"inception_resnet_v2_b6_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b6_1x7\"\n top: \"inception_resnet_v2_b6_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b6_1x7\"\n top: \"inception_resnet_v2_b6_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b6_1x7\"\n top: \"inception_resnet_v2_b6_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b6_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b6_1x7\"\n top: \"inception_resnet_v2_b6_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b6_7x1\"\n top: \"inception_resnet_v2_b6_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b6_7x1\"\n top: \"inception_resnet_v2_b6_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b6_7x1\"\n top: \"inception_resnet_v2_b6_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b6_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b6_1x1\"\n bottom: \"inception_resnet_v2_b6_7x1\"\n top: \"inception_resnet_v2_b6_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b6_concat\"\n top: \"inception_resnet_v2_b6_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b6_1x1_2\"\n top: \"inception_resnet_v2_b6_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b6_1x1_2\"\n top: \"inception_resnet_v2_b6_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b6_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b5_residual_eltwise\"\n bottom: \"inception_resnet_v2_b6_1x1_2\"\n top: \"inception_resnet_v2_b6_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b6_residual_eltwise\"\n top: \"inception_resnet_v2_b7_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b7_1x1\"\n top: \"inception_resnet_v2_b7_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b7_1x1\"\n top: \"inception_resnet_v2_b7_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b7_1x1\"\n top: \"inception_resnet_v2_b7_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b6_residual_eltwise\"\n top: \"inception_resnet_v2_b7_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b7_1x7_reduce\"\n top: \"inception_resnet_v2_b7_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b7_1x7_reduce\"\n top: \"inception_resnet_v2_b7_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b7_1x7_reduce\"\n top: \"inception_resnet_v2_b7_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b7_1x7_reduce\"\n top: \"inception_resnet_v2_b7_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b7_1x7\"\n top: \"inception_resnet_v2_b7_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b7_1x7\"\n top: \"inception_resnet_v2_b7_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b7_1x7\"\n top: \"inception_resnet_v2_b7_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b7_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b7_1x7\"\n top: \"inception_resnet_v2_b7_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b7_7x1\"\n top: \"inception_resnet_v2_b7_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b7_7x1\"\n top: \"inception_resnet_v2_b7_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b7_7x1\"\n top: \"inception_resnet_v2_b7_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b7_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b7_1x1\"\n bottom: \"inception_resnet_v2_b7_7x1\"\n top: \"inception_resnet_v2_b7_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b7_concat\"\n top: \"inception_resnet_v2_b7_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b7_1x1_2\"\n top: \"inception_resnet_v2_b7_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b7_1x1_2\"\n top: \"inception_resnet_v2_b7_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b7_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b6_residual_eltwise\"\n bottom: \"inception_resnet_v2_b7_1x1_2\"\n top: \"inception_resnet_v2_b7_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b7_residual_eltwise\"\n top: \"inception_resnet_v2_b8_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b8_1x1\"\n top: \"inception_resnet_v2_b8_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b8_1x1\"\n top: \"inception_resnet_v2_b8_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b8_1x1\"\n top: \"inception_resnet_v2_b8_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b7_residual_eltwise\"\n top: \"inception_resnet_v2_b8_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b8_1x7_reduce\"\n top: \"inception_resnet_v2_b8_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b8_1x7_reduce\"\n top: \"inception_resnet_v2_b8_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b8_1x7_reduce\"\n top: \"inception_resnet_v2_b8_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b8_1x7_reduce\"\n top: \"inception_resnet_v2_b8_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b8_1x7\"\n top: \"inception_resnet_v2_b8_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b8_1x7\"\n top: \"inception_resnet_v2_b8_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b8_1x7\"\n top: \"inception_resnet_v2_b8_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b8_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b8_1x7\"\n top: \"inception_resnet_v2_b8_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b8_7x1\"\n top: \"inception_resnet_v2_b8_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b8_7x1\"\n top: \"inception_resnet_v2_b8_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b8_7x1\"\n top: \"inception_resnet_v2_b8_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b8_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b8_1x1\"\n bottom: \"inception_resnet_v2_b8_7x1\"\n top: \"inception_resnet_v2_b8_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b8_concat\"\n top: \"inception_resnet_v2_b8_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b8_1x1_2\"\n top: \"inception_resnet_v2_b8_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b8_1x1_2\"\n top: \"inception_resnet_v2_b8_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b8_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b7_residual_eltwise\"\n bottom: \"inception_resnet_v2_b8_1x1_2\"\n top: \"inception_resnet_v2_b8_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b8_residual_eltwise\"\n top: \"inception_resnet_v2_b9_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b9_1x1\"\n top: \"inception_resnet_v2_b9_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b9_1x1\"\n top: \"inception_resnet_v2_b9_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b9_1x1\"\n top: \"inception_resnet_v2_b9_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b8_residual_eltwise\"\n top: \"inception_resnet_v2_b9_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b9_1x7_reduce\"\n top: \"inception_resnet_v2_b9_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b9_1x7_reduce\"\n top: \"inception_resnet_v2_b9_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b9_1x7_reduce\"\n top: \"inception_resnet_v2_b9_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b9_1x7_reduce\"\n top: \"inception_resnet_v2_b9_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b9_1x7\"\n top: \"inception_resnet_v2_b9_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b9_1x7\"\n top: \"inception_resnet_v2_b9_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b9_1x7\"\n top: \"inception_resnet_v2_b9_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b9_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b9_1x7\"\n top: \"inception_resnet_v2_b9_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b9_7x1\"\n top: \"inception_resnet_v2_b9_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b9_7x1\"\n top: \"inception_resnet_v2_b9_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b9_7x1\"\n top: \"inception_resnet_v2_b9_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b9_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b9_1x1\"\n bottom: \"inception_resnet_v2_b9_7x1\"\n top: \"inception_resnet_v2_b9_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b9_concat\"\n top: \"inception_resnet_v2_b9_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b9_1x1_2\"\n top: \"inception_resnet_v2_b9_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b9_1x1_2\"\n top: \"inception_resnet_v2_b9_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b9_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b8_residual_eltwise\"\n bottom: \"inception_resnet_v2_b9_1x1_2\"\n top: \"inception_resnet_v2_b9_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b9_residual_eltwise\"\n top: \"inception_resnet_v2_b10_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b10_1x1\"\n top: \"inception_resnet_v2_b10_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b10_1x1\"\n top: \"inception_resnet_v2_b10_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b10_1x1\"\n top: \"inception_resnet_v2_b10_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b9_residual_eltwise\"\n top: \"inception_resnet_v2_b10_1x7_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 128\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b10_1x7_reduce\"\n top: \"inception_resnet_v2_b10_1x7_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b10_1x7_reduce\"\n top: \"inception_resnet_v2_b10_1x7_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b10_1x7_reduce\"\n top: \"inception_resnet_v2_b10_1x7_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b10_1x7_reduce\"\n top: \"inception_resnet_v2_b10_1x7\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 160\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 3\n kernel_h: 1\n kernel_w: 7\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b10_1x7\"\n top: \"inception_resnet_v2_b10_1x7\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b10_1x7\"\n top: \"inception_resnet_v2_b10_1x7\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x7_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b10_1x7\"\n top: \"inception_resnet_v2_b10_1x7\"\n}\nlayer {\n name: \"inception_resnet_v2_b10_7x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b10_1x7\"\n top: \"inception_resnet_v2_b10_7x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 3\n pad_w: 0\n kernel_h: 7\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_7x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b10_7x1\"\n top: \"inception_resnet_v2_b10_7x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_7x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b10_7x1\"\n top: \"inception_resnet_v2_b10_7x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_7x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_b10_7x1\"\n top: \"inception_resnet_v2_b10_7x1\"\n}\nlayer {\n name: \"inception_resnet_v2_b10_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_b10_1x1\"\n bottom: \"inception_resnet_v2_b10_7x1\"\n top: \"inception_resnet_v2_b10_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b10_concat\"\n top: \"inception_resnet_v2_b10_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 1152\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_b10_1x1_2\"\n top: \"inception_resnet_v2_b10_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_b10_1x1_2\"\n top: \"inception_resnet_v2_b10_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_b10_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_b9_residual_eltwise\"\n bottom: \"inception_resnet_v2_b10_1x1_2\"\n top: \"inception_resnet_v2_b10_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"reduction_b_pool\"\n type: \"Pooling\"\n bottom: \"inception_resnet_v2_b10_residual_eltwise\"\n top: \"reduction_b_pool\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b10_residual_eltwise\"\n top: \"reduction_b_3x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3_reduce\"\n}\nlayer {\n name: \"reduction_b_3x3\"\n type: \"Convolution\"\n bottom: \"reduction_b_3x3_reduce\"\n top: \"reduction_b_3x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 384\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3\"\n top: \"reduction_b_3x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3\"\n top: \"reduction_b_3x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3\"\n top: \"reduction_b_3x3\"\n}\nlayer {\n name: \"reduction_b_3x3_2_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b10_residual_eltwise\"\n top: \"reduction_b_3x3_2_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_2_reduce\"\n top: \"reduction_b_3x3_2_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_reduce_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_2_reduce\"\n top: \"reduction_b_3x3_2_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_reduce_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_2_reduce\"\n top: \"reduction_b_3x3_2_reduce\"\n}\nlayer {\n name: \"reduction_b_3x3_2\"\n type: \"Convolution\"\n bottom: \"reduction_b_3x3_2_reduce\"\n top: \"reduction_b_3x3_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_2\"\n top: \"reduction_b_3x3_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_2\"\n top: \"reduction_b_3x3_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_2_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_2\"\n top: \"reduction_b_3x3_2\"\n}\nlayer {\n name: \"reduction_b_3x3_3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_b10_residual_eltwise\"\n top: \"reduction_b_3x3_3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_3_reduce\"\n top: \"reduction_b_3x3_3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_3_reduce_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_3_reduce\"\n top: \"reduction_b_3x3_3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_3_reduce\"\n top: \"reduction_b_3x3_3_reduce\"\n}\nlayer {\n name: \"reduction_b_3x3_3\"\n type: \"Convolution\"\n bottom: \"reduction_b_3x3_3_reduce\"\n top: \"reduction_b_3x3_3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_3_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_3\"\n top: \"reduction_b_3x3_3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_3_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_3\"\n top: \"reduction_b_3x3_3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_3_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_3\"\n top: \"reduction_b_3x3_3\"\n}\nlayer {\n name: \"reduction_b_3x3_4\"\n type: \"Convolution\"\n bottom: \"reduction_b_3x3_3\"\n top: \"reduction_b_3x3_4\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n pad: 0\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"reduction_b_3x3_4_bn\"\n type: \"BatchNorm\"\n bottom: \"reduction_b_3x3_4\"\n top: \"reduction_b_3x3_4\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"reduction_b_3x3_4_scale\"\n type: \"Scale\"\n bottom: \"reduction_b_3x3_4\"\n top: \"reduction_b_3x3_4\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"reduction_b_3x3_4_relu\"\n type: \"ReLU\"\n bottom: \"reduction_b_3x3_4\"\n top: \"reduction_b_3x3_4\"\n}\nlayer {\n name: \"reduction_b_concat\"\n type: \"Concat\"\n bottom: \"reduction_b_pool\"\n bottom: \"reduction_b_3x3\"\n bottom: \"reduction_b_3x3_2\"\n bottom: \"reduction_b_3x3_4\"\n top: \"reduction_b_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x1\"\n type: \"Convolution\"\n bottom: \"reduction_b_concat\"\n top: \"inception_resnet_v2_c1_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c1_1x1\"\n top: \"inception_resnet_v2_c1_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c1_1x1\"\n top: \"inception_resnet_v2_c1_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c1_1x1\"\n top: \"inception_resnet_v2_c1_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3_reduce\"\n type: \"Convolution\"\n bottom: \"reduction_b_concat\"\n top: \"inception_resnet_v2_c1_1x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c1_1x3_reduce\"\n top: \"inception_resnet_v2_c1_1x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c1_1x3_reduce\"\n top: \"inception_resnet_v2_c1_1x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c1_1x3_reduce\"\n top: \"inception_resnet_v2_c1_1x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c1_1x3_reduce\"\n top: \"inception_resnet_v2_c1_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c1_1x3\"\n top: \"inception_resnet_v2_c1_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c1_1x3\"\n top: \"inception_resnet_v2_c1_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c1_1x3\"\n top: \"inception_resnet_v2_c1_1x3\"\n}\nlayer {\n name: \"inception_resnet_v2_c1_3x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c1_1x3\"\n top: \"inception_resnet_v2_c1_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c1_3x1\"\n top: \"inception_resnet_v2_c1_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c1_3x1\"\n top: \"inception_resnet_v2_c1_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c1_3x1\"\n top: \"inception_resnet_v2_c1_3x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c1_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_c1_1x1\"\n bottom: \"inception_resnet_v2_c1_3x1\"\n top: \"inception_resnet_v2_c1_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c1_concat\"\n top: \"inception_resnet_v2_c1_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 2048\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c1_1x1_2\"\n top: \"inception_resnet_v2_c1_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c1_1x1_2\"\n top: \"inception_resnet_v2_c1_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c1_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"reduction_b_concat\"\n bottom: \"inception_resnet_v2_c1_1x1_2\"\n top: \"inception_resnet_v2_c1_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c1_residual_eltwise\"\n top: \"inception_resnet_v2_c2_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c2_1x1\"\n top: \"inception_resnet_v2_c2_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c2_1x1\"\n top: \"inception_resnet_v2_c2_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c2_1x1\"\n top: \"inception_resnet_v2_c2_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c1_residual_eltwise\"\n top: \"inception_resnet_v2_c2_1x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c2_1x3_reduce\"\n top: \"inception_resnet_v2_c2_1x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c2_1x3_reduce\"\n top: \"inception_resnet_v2_c2_1x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c2_1x3_reduce\"\n top: \"inception_resnet_v2_c2_1x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c2_1x3_reduce\"\n top: \"inception_resnet_v2_c2_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c2_1x3\"\n top: \"inception_resnet_v2_c2_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c2_1x3\"\n top: \"inception_resnet_v2_c2_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c2_1x3\"\n top: \"inception_resnet_v2_c2_1x3\"\n}\nlayer {\n name: \"inception_resnet_v2_c2_3x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c2_1x3\"\n top: \"inception_resnet_v2_c2_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c2_3x1\"\n top: \"inception_resnet_v2_c2_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c2_3x1\"\n top: \"inception_resnet_v2_c2_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c2_3x1\"\n top: \"inception_resnet_v2_c2_3x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c2_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_c2_1x1\"\n bottom: \"inception_resnet_v2_c2_3x1\"\n top: \"inception_resnet_v2_c2_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c2_concat\"\n top: \"inception_resnet_v2_c2_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 2048\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c2_1x1_2\"\n top: \"inception_resnet_v2_c2_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c2_1x1_2\"\n top: \"inception_resnet_v2_c2_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c2_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_c1_residual_eltwise\"\n bottom: \"inception_resnet_v2_c2_1x1_2\"\n top: \"inception_resnet_v2_c2_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c2_residual_eltwise\"\n top: \"inception_resnet_v2_c3_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c3_1x1\"\n top: \"inception_resnet_v2_c3_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c3_1x1\"\n top: \"inception_resnet_v2_c3_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c3_1x1\"\n top: \"inception_resnet_v2_c3_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c2_residual_eltwise\"\n top: \"inception_resnet_v2_c3_1x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c3_1x3_reduce\"\n top: \"inception_resnet_v2_c3_1x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c3_1x3_reduce\"\n top: \"inception_resnet_v2_c3_1x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c3_1x3_reduce\"\n top: \"inception_resnet_v2_c3_1x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c3_1x3_reduce\"\n top: \"inception_resnet_v2_c3_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c3_1x3\"\n top: \"inception_resnet_v2_c3_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c3_1x3\"\n top: \"inception_resnet_v2_c3_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c3_1x3\"\n top: \"inception_resnet_v2_c3_1x3\"\n}\nlayer {\n name: \"inception_resnet_v2_c3_3x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c3_1x3\"\n top: \"inception_resnet_v2_c3_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c3_3x1\"\n top: \"inception_resnet_v2_c3_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c3_3x1\"\n top: \"inception_resnet_v2_c3_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c3_3x1\"\n top: \"inception_resnet_v2_c3_3x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c3_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_c3_1x1\"\n bottom: \"inception_resnet_v2_c3_3x1\"\n top: \"inception_resnet_v2_c3_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c3_concat\"\n top: \"inception_resnet_v2_c3_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 2048\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c3_1x1_2\"\n top: \"inception_resnet_v2_c3_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c3_1x1_2\"\n top: \"inception_resnet_v2_c3_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c3_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_c2_residual_eltwise\"\n bottom: \"inception_resnet_v2_c3_1x1_2\"\n top: \"inception_resnet_v2_c3_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c3_residual_eltwise\"\n top: \"inception_resnet_v2_c4_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c4_1x1\"\n top: \"inception_resnet_v2_c4_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c4_1x1\"\n top: \"inception_resnet_v2_c4_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c4_1x1\"\n top: \"inception_resnet_v2_c4_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c3_residual_eltwise\"\n top: \"inception_resnet_v2_c4_1x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c4_1x3_reduce\"\n top: \"inception_resnet_v2_c4_1x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c4_1x3_reduce\"\n top: \"inception_resnet_v2_c4_1x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c4_1x3_reduce\"\n top: \"inception_resnet_v2_c4_1x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c4_1x3_reduce\"\n top: \"inception_resnet_v2_c4_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c4_1x3\"\n top: \"inception_resnet_v2_c4_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c4_1x3\"\n top: \"inception_resnet_v2_c4_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c4_1x3\"\n top: \"inception_resnet_v2_c4_1x3\"\n}\nlayer {\n name: \"inception_resnet_v2_c4_3x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c4_1x3\"\n top: \"inception_resnet_v2_c4_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c4_3x1\"\n top: \"inception_resnet_v2_c4_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c4_3x1\"\n top: \"inception_resnet_v2_c4_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c4_3x1\"\n top: \"inception_resnet_v2_c4_3x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c4_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_c4_1x1\"\n bottom: \"inception_resnet_v2_c4_3x1\"\n top: \"inception_resnet_v2_c4_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c4_concat\"\n top: \"inception_resnet_v2_c4_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 2048\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c4_1x1_2\"\n top: \"inception_resnet_v2_c4_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c4_1x1_2\"\n top: \"inception_resnet_v2_c4_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c4_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_c3_residual_eltwise\"\n bottom: \"inception_resnet_v2_c4_1x1_2\"\n top: \"inception_resnet_v2_c4_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c4_residual_eltwise\"\n top: \"inception_resnet_v2_c5_1x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c5_1x1\"\n top: \"inception_resnet_v2_c5_1x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c5_1x1\"\n top: \"inception_resnet_v2_c5_1x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c5_1x1\"\n top: \"inception_resnet_v2_c5_1x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3_reduce\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c4_residual_eltwise\"\n top: \"inception_resnet_v2_c5_1x3_reduce\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 192\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3_reduce_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c5_1x3_reduce\"\n top: \"inception_resnet_v2_c5_1x3_reduce\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3_reduce_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c5_1x3_reduce\"\n top: \"inception_resnet_v2_c5_1x3_reduce\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3_reduce_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c5_1x3_reduce\"\n top: \"inception_resnet_v2_c5_1x3_reduce\"\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c5_1x3_reduce\"\n top: \"inception_resnet_v2_c5_1x3\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 224\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 0\n pad_w: 1\n kernel_h: 1\n kernel_w: 3\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c5_1x3\"\n top: \"inception_resnet_v2_c5_1x3\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c5_1x3\"\n top: \"inception_resnet_v2_c5_1x3\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x3_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c5_1x3\"\n top: \"inception_resnet_v2_c5_1x3\"\n}\nlayer {\n name: \"inception_resnet_v2_c5_3x1\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c5_1x3\"\n top: \"inception_resnet_v2_c5_3x1\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 256\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n pad_h: 1\n pad_w: 0\n kernel_h: 3\n kernel_w: 1\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_3x1_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c5_3x1\"\n top: \"inception_resnet_v2_c5_3x1\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_3x1_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c5_3x1\"\n top: \"inception_resnet_v2_c5_3x1\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_3x1_relu\"\n type: \"ReLU\"\n bottom: \"inception_resnet_v2_c5_3x1\"\n top: \"inception_resnet_v2_c5_3x1\"\n}\nlayer {\n name: \"inception_resnet_v2_c5_concat\"\n type: \"Concat\"\n bottom: \"inception_resnet_v2_c5_1x1\"\n bottom: \"inception_resnet_v2_c5_3x1\"\n top: \"inception_resnet_v2_c5_concat\"\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x1_2\"\n type: \"Convolution\"\n bottom: \"inception_resnet_v2_c5_concat\"\n top: \"inception_resnet_v2_c5_1x1_2\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n convolution_param {\n num_output: 2048\n pad: 0\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"xavier\"\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0.2\n }\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x1_2_bn\"\n type: \"BatchNorm\"\n bottom: \"inception_resnet_v2_c5_1x1_2\"\n top: \"inception_resnet_v2_c5_1x1_2\"\n batch_norm_param {\n use_global_stats: false\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_1x1_2_scale\"\n type: \"Scale\"\n bottom: \"inception_resnet_v2_c5_1x1_2\"\n top: \"inception_resnet_v2_c5_1x1_2\"\n scale_param {\n bias_term: true\n }\n}\nlayer {\n name: \"inception_resnet_v2_c5_residual_eltwise\"\n type: \"Eltwise\"\n bottom: \"inception_resnet_v2_c4_residual_eltwise\"\n bottom: \"inception_resnet_v2_c5_1x1_2\"\n top: \"inception_resnet_v2_c5_residual_eltwise\"\n eltwise_param {\n operation: SUM\n }\n}\nlayer {\n name: \"pool_8x8_s1\"\n type: \"Pooling\"\n bottom: \"inception_resnet_v2_c5_residual_eltwise\"\n top: \"pool_8x8_s1\"\n pooling_param {\n pool: AVE\n global_pooling: true\n }\n}\nlayer {\n name: \"pool_8x8_s1_drop\"\n type: \"Dropout\"\n bottom: \"pool_8x8_s1\"\n top: \"pool_8x8_s1_drop\"\n dropout_param {\n dropout_ratio: 0.2\n }\n}\nlayer {\n name: \"classifier\"\n type: \"InnerProduct\"\n bottom: \"pool_8x8_s1_drop\"\n top: \"classifier\"\n param {\n lr_mult: 1\n decay_mult: 1\n }\n param {\n lr_mult: 2\n decay_mult: 0\n }\n inner_product_param {\n num_output: 1000\n weight_filler {\n type: \"xavier\"\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"classifier\"\n top: \"loss\"\n}\n" }, { "path": "presets/nin.prototxt", "content": "name: \"nin_imagenet\"\nlayers {\n top: \"data\"\n top: \"label\"\n name: \"data\"\n type: DATA\n data_param {\n source: \"/home/linmin/IMAGENET-LMDB/imagenet-train-lmdb\"\n backend: LMDB\n batch_size: 64\n }\n transform_param {\n crop_size: 224\n mirror: true\n mean_file: \"/home/linmin/IMAGENET-LMDB/imagenet-train-mean\"\n }\n include: { phase: TRAIN }\n}\nlayers {\n top: \"data\"\n top: \"label\"\n name: \"data\"\n type: DATA\n data_param {\n source: \"/home/linmin/IMAGENET-LMDB/imagenet-val-lmdb\"\n backend: LMDB\n batch_size: 89\n }\n transform_param {\n crop_size: 224\n mirror: false\n mean_file: \"/home/linmin/IMAGENET-LMDB/imagenet-train-mean\"\n }\n include: { phase: TEST }\n}\nlayers {\n bottom: \"data\"\n top: \"conv1\"\n name: \"conv1\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 96\n kernel_size: 11\n stride: 4\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"conv1\"\n top: \"conv1\"\n name: \"relu0\"\n type: RELU\n}\nlayers {\n bottom: \"conv1\"\n top: \"cccp1\"\n name: \"cccp1\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 96\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp1\"\n top: \"cccp1\"\n name: \"relu1\"\n type: RELU\n}\nlayers {\n bottom: \"cccp1\"\n top: \"cccp2\"\n name: \"cccp2\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 96\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp2\"\n top: \"cccp2\"\n name: \"relu2\"\n type: RELU\n}\nlayers {\n bottom: \"cccp2\"\n top: \"pool0\"\n name: \"pool0\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayers {\n bottom: \"pool0\"\n top: \"conv2\"\n name: \"conv2\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 256\n pad: 2\n kernel_size: 5\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"conv2\"\n top: \"conv2\"\n name: \"relu3\"\n type: RELU\n}\nlayers {\n bottom: \"conv2\"\n top: \"cccp3\"\n name: \"cccp3\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 256\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp3\"\n top: \"cccp3\"\n name: \"relu5\"\n type: RELU\n}\nlayers {\n bottom: \"cccp3\"\n top: \"cccp4\"\n name: \"cccp4\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 256\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp4\"\n top: \"cccp4\"\n name: \"relu6\"\n type: RELU\n}\nlayers {\n bottom: \"cccp4\"\n top: \"pool2\"\n name: \"pool2\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayers {\n bottom: \"pool2\"\n top: \"conv3\"\n name: \"conv3\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 384\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"conv3\"\n top: \"conv3\"\n name: \"relu7\"\n type: RELU\n}\nlayers {\n bottom: \"conv3\"\n top: \"cccp5\"\n name: \"cccp5\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 384\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp5\"\n top: \"cccp5\"\n name: \"relu8\"\n type: RELU\n}\nlayers {\n bottom: \"cccp5\"\n top: \"cccp6\"\n name: \"cccp6\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 384\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp6\"\n top: \"cccp6\"\n name: \"relu9\"\n type: RELU\n}\nlayers {\n bottom: \"cccp6\"\n top: \"pool3\"\n name: \"pool3\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayers {\n bottom: \"pool3\"\n top: \"pool3\"\n name: \"drop\"\n type: DROPOUT\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayers {\n bottom: \"pool3\"\n top: \"conv4-1024\"\n name: \"conv4-1024\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 1024\n pad: 1\n kernel_size: 3\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"conv4-1024\"\n top: \"conv4-1024\"\n name: \"relu10\"\n type: RELU\n}\nlayers {\n bottom: \"conv4-1024\"\n top: \"cccp7-1024\"\n name: \"cccp7-1024\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 1024\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.05\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp7-1024\"\n top: \"cccp7-1024\"\n name: \"relu11\"\n type: RELU\n}\nlayers {\n bottom: \"cccp7-1024\"\n top: \"cccp8-1024\"\n name: \"cccp8-1024\"\n type: CONVOLUTION\n blobs_lr: 1\n blobs_lr: 2\n weight_decay: 1\n weight_decay: 0\n convolution_param {\n num_output: 1000\n kernel_size: 1\n stride: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0\n std: 0.01\n }\n bias_filler {\n type: \"constant\"\n value: 0\n }\n }\n}\nlayers {\n bottom: \"cccp8-1024\"\n top: \"cccp8-1024\"\n name: \"relu12\"\n type: RELU\n}\nlayers {\n bottom: \"cccp8-1024\"\n top: \"pool4\"\n name: \"pool4\"\n type: POOLING\n pooling_param {\n pool: AVE\n kernel_size: 6\n stride: 1\n }\n}\nlayers {\n name: \"accuracy\"\n type: ACCURACY\n bottom: \"pool4\"\n bottom: \"label\"\n top: \"accuracy\"\n include: { phase: TEST }\n}\nlayers {\n bottom: \"pool4\"\n bottom: \"label\"\n name: \"loss\"\n type: SOFTMAX_LOSS\n include: { phase: TRAIN }\n}\n" }, { "path": "presets/resnet-152.prototxt", "content": "name: \"ResNet-152\"\ninput: \"data\"\ninput_dim: 1\ninput_dim: 3\ninput_dim: 224\ninput_dim: 224\n\nlayer {\n\tbottom: \"data\"\n\ttop: \"conv1\"\n\tname: \"conv1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 7\n\t\tpad: 3\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"bn_conv1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"scale_conv1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"conv1\"\n\tbottom: \"conv1\"\n\tname: \"conv1_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"pool1\"\n\tname: \"pool1\"\n\ttype: \"Pooling\"\n\tpooling_param {\n\t\tkernel_size: 3\n\t\tstride: 2\n\t\tpool: MAX\n\t}\n}\n\nlayer {\n\tbottom: \"pool1\"\n\ttop: \"res2a_branch1\"\n\tname: \"res2a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\ttop: \"res2a_branch1\"\n\tname: \"bn2a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\ttop: \"res2a_branch1\"\n\tname: \"scale2a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"pool1\"\n\ttop: \"res2a_branch2a\"\n\tname: \"res2a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"bn2a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"scale2a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res2a_branch2a\"\n\tbottom: \"res2a_branch2a\"\n\tname: \"res2a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2b\"\n\tname: \"res2a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"bn2a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"scale2a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res2a_branch2b\"\n\tbottom: \"res2a_branch2b\"\n\tname: \"res2a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2c\"\n\tname: \"res2a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a_branch2c\"\n\tname: \"bn2a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a_branch2c\"\n\tname: \"scale2a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a\"\n\tname: \"res2a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2a\"\n\ttop: \"res2a\"\n\tname: \"res2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"res2b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"bn2b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"scale2b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res2b_branch2a\"\n\tbottom: \"res2b_branch2a\"\n\tname: \"res2b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2b\"\n\tname: \"res2b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"bn2b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"scale2b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res2b_branch2b\"\n\tbottom: \"res2b_branch2b\"\n\tname: \"res2b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2c\"\n\tname: \"res2b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b_branch2c\"\n\tname: \"bn2b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b_branch2c\"\n\tname: \"scale2b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a\"\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b\"\n\tname: \"res2b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2b\"\n\ttop: \"res2b\"\n\tname: \"res2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b\"\n\ttop: \"res2c_branch2a\"\n\tname: \"res2c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"bn2c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"scale2c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res2c_branch2a\"\n\tbottom: \"res2c_branch2a\"\n\tname: \"res2c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2b\"\n\tname: \"res2c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"bn2c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"scale2c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res2c_branch2b\"\n\tbottom: \"res2c_branch2b\"\n\tname: \"res2c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2c\"\n\tname: \"res2c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c_branch2c\"\n\tname: \"bn2c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c_branch2c\"\n\tname: \"scale2c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b\"\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c\"\n\tname: \"res2c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res2c\"\n\tname: \"res2c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res3a_branch1\"\n\tname: \"res3a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\ttop: \"res3a_branch1\"\n\tname: \"bn3a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\ttop: \"res3a_branch1\"\n\tname: \"scale3a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res3a_branch2a\"\n\tname: \"res3a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"bn3a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"scale3a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3a_branch2a\"\n\tbottom: \"res3a_branch2a\"\n\tname: \"res3a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2b\"\n\tname: \"res3a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"bn3a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"scale3a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3a_branch2b\"\n\tbottom: \"res3a_branch2b\"\n\tname: \"res3a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2c\"\n\tname: \"res3a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a_branch2c\"\n\tname: \"bn3a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a_branch2c\"\n\tname: \"scale3a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a\"\n\tname: \"res3a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3a\"\n\ttop: \"res3a\"\n\tname: \"res3a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a\"\n\ttop: \"res3b1_branch2a\"\n\tname: \"res3b1_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b1_branch2a\"\n\ttop: \"res3b1_branch2a\"\n\tname: \"bn3b1_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b1_branch2a\"\n\ttop: \"res3b1_branch2a\"\n\tname: \"scale3b1_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b1_branch2a\"\n\tbottom: \"res3b1_branch2a\"\n\tname: \"res3b1_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b1_branch2a\"\n\ttop: \"res3b1_branch2b\"\n\tname: \"res3b1_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b1_branch2b\"\n\ttop: \"res3b1_branch2b\"\n\tname: \"bn3b1_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b1_branch2b\"\n\ttop: \"res3b1_branch2b\"\n\tname: \"scale3b1_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b1_branch2b\"\n\tbottom: \"res3b1_branch2b\"\n\tname: \"res3b1_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b1_branch2b\"\n\ttop: \"res3b1_branch2c\"\n\tname: \"res3b1_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b1_branch2c\"\n\ttop: \"res3b1_branch2c\"\n\tname: \"bn3b1_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b1_branch2c\"\n\ttop: \"res3b1_branch2c\"\n\tname: \"scale3b1_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a\"\n\tbottom: \"res3b1_branch2c\"\n\ttop: \"res3b1\"\n\tname: \"res3b1\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b1\"\n\ttop: \"res3b1\"\n\tname: \"res3b1_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b1\"\n\ttop: \"res3b2_branch2a\"\n\tname: \"res3b2_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b2_branch2a\"\n\ttop: \"res3b2_branch2a\"\n\tname: \"bn3b2_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b2_branch2a\"\n\ttop: \"res3b2_branch2a\"\n\tname: \"scale3b2_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b2_branch2a\"\n\tbottom: \"res3b2_branch2a\"\n\tname: \"res3b2_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b2_branch2a\"\n\ttop: \"res3b2_branch2b\"\n\tname: \"res3b2_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b2_branch2b\"\n\ttop: \"res3b2_branch2b\"\n\tname: \"bn3b2_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b2_branch2b\"\n\ttop: \"res3b2_branch2b\"\n\tname: \"scale3b2_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b2_branch2b\"\n\tbottom: \"res3b2_branch2b\"\n\tname: \"res3b2_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b2_branch2b\"\n\ttop: \"res3b2_branch2c\"\n\tname: \"res3b2_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b2_branch2c\"\n\ttop: \"res3b2_branch2c\"\n\tname: \"bn3b2_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b2_branch2c\"\n\ttop: \"res3b2_branch2c\"\n\tname: \"scale3b2_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b1\"\n\tbottom: \"res3b2_branch2c\"\n\ttop: \"res3b2\"\n\tname: \"res3b2\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b2\"\n\ttop: \"res3b2\"\n\tname: \"res3b2_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b2\"\n\ttop: \"res3b3_branch2a\"\n\tname: \"res3b3_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b3_branch2a\"\n\ttop: \"res3b3_branch2a\"\n\tname: \"bn3b3_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b3_branch2a\"\n\ttop: \"res3b3_branch2a\"\n\tname: \"scale3b3_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b3_branch2a\"\n\tbottom: \"res3b3_branch2a\"\n\tname: \"res3b3_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b3_branch2a\"\n\ttop: \"res3b3_branch2b\"\n\tname: \"res3b3_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b3_branch2b\"\n\ttop: \"res3b3_branch2b\"\n\tname: \"bn3b3_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b3_branch2b\"\n\ttop: \"res3b3_branch2b\"\n\tname: \"scale3b3_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b3_branch2b\"\n\tbottom: \"res3b3_branch2b\"\n\tname: \"res3b3_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b3_branch2b\"\n\ttop: \"res3b3_branch2c\"\n\tname: \"res3b3_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b3_branch2c\"\n\ttop: \"res3b3_branch2c\"\n\tname: \"bn3b3_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b3_branch2c\"\n\ttop: \"res3b3_branch2c\"\n\tname: \"scale3b3_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b2\"\n\tbottom: \"res3b3_branch2c\"\n\ttop: \"res3b3\"\n\tname: \"res3b3\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b3\"\n\ttop: \"res3b3\"\n\tname: \"res3b3_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b3\"\n\ttop: \"res3b4_branch2a\"\n\tname: \"res3b4_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b4_branch2a\"\n\ttop: \"res3b4_branch2a\"\n\tname: \"bn3b4_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b4_branch2a\"\n\ttop: \"res3b4_branch2a\"\n\tname: \"scale3b4_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b4_branch2a\"\n\tbottom: \"res3b4_branch2a\"\n\tname: \"res3b4_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b4_branch2a\"\n\ttop: \"res3b4_branch2b\"\n\tname: \"res3b4_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b4_branch2b\"\n\ttop: \"res3b4_branch2b\"\n\tname: \"bn3b4_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b4_branch2b\"\n\ttop: \"res3b4_branch2b\"\n\tname: \"scale3b4_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b4_branch2b\"\n\tbottom: \"res3b4_branch2b\"\n\tname: \"res3b4_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b4_branch2b\"\n\ttop: \"res3b4_branch2c\"\n\tname: \"res3b4_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b4_branch2c\"\n\ttop: \"res3b4_branch2c\"\n\tname: \"bn3b4_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b4_branch2c\"\n\ttop: \"res3b4_branch2c\"\n\tname: \"scale3b4_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b3\"\n\tbottom: \"res3b4_branch2c\"\n\ttop: \"res3b4\"\n\tname: \"res3b4\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b4\"\n\ttop: \"res3b4\"\n\tname: \"res3b4_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b4\"\n\ttop: \"res3b5_branch2a\"\n\tname: \"res3b5_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b5_branch2a\"\n\ttop: \"res3b5_branch2a\"\n\tname: \"bn3b5_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b5_branch2a\"\n\ttop: \"res3b5_branch2a\"\n\tname: \"scale3b5_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b5_branch2a\"\n\tbottom: \"res3b5_branch2a\"\n\tname: \"res3b5_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b5_branch2a\"\n\ttop: \"res3b5_branch2b\"\n\tname: \"res3b5_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b5_branch2b\"\n\ttop: \"res3b5_branch2b\"\n\tname: \"bn3b5_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b5_branch2b\"\n\ttop: \"res3b5_branch2b\"\n\tname: \"scale3b5_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b5_branch2b\"\n\tbottom: \"res3b5_branch2b\"\n\tname: \"res3b5_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b5_branch2b\"\n\ttop: \"res3b5_branch2c\"\n\tname: \"res3b5_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b5_branch2c\"\n\ttop: \"res3b5_branch2c\"\n\tname: \"bn3b5_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b5_branch2c\"\n\ttop: \"res3b5_branch2c\"\n\tname: \"scale3b5_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b4\"\n\tbottom: \"res3b5_branch2c\"\n\ttop: \"res3b5\"\n\tname: \"res3b5\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b5\"\n\ttop: \"res3b5\"\n\tname: \"res3b5_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b5\"\n\ttop: \"res3b6_branch2a\"\n\tname: \"res3b6_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b6_branch2a\"\n\ttop: \"res3b6_branch2a\"\n\tname: \"bn3b6_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b6_branch2a\"\n\ttop: \"res3b6_branch2a\"\n\tname: \"scale3b6_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b6_branch2a\"\n\tbottom: \"res3b6_branch2a\"\n\tname: \"res3b6_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b6_branch2a\"\n\ttop: \"res3b6_branch2b\"\n\tname: \"res3b6_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b6_branch2b\"\n\ttop: \"res3b6_branch2b\"\n\tname: \"bn3b6_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b6_branch2b\"\n\ttop: \"res3b6_branch2b\"\n\tname: \"scale3b6_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b6_branch2b\"\n\tbottom: \"res3b6_branch2b\"\n\tname: \"res3b6_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b6_branch2b\"\n\ttop: \"res3b6_branch2c\"\n\tname: \"res3b6_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b6_branch2c\"\n\ttop: \"res3b6_branch2c\"\n\tname: \"bn3b6_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b6_branch2c\"\n\ttop: \"res3b6_branch2c\"\n\tname: \"scale3b6_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b5\"\n\tbottom: \"res3b6_branch2c\"\n\ttop: \"res3b6\"\n\tname: \"res3b6\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b6\"\n\ttop: \"res3b6\"\n\tname: \"res3b6_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b6\"\n\ttop: \"res3b7_branch2a\"\n\tname: \"res3b7_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b7_branch2a\"\n\ttop: \"res3b7_branch2a\"\n\tname: \"bn3b7_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b7_branch2a\"\n\ttop: \"res3b7_branch2a\"\n\tname: \"scale3b7_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b7_branch2a\"\n\tbottom: \"res3b7_branch2a\"\n\tname: \"res3b7_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b7_branch2a\"\n\ttop: \"res3b7_branch2b\"\n\tname: \"res3b7_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b7_branch2b\"\n\ttop: \"res3b7_branch2b\"\n\tname: \"bn3b7_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b7_branch2b\"\n\ttop: \"res3b7_branch2b\"\n\tname: \"scale3b7_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res3b7_branch2b\"\n\tbottom: \"res3b7_branch2b\"\n\tname: \"res3b7_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b7_branch2b\"\n\ttop: \"res3b7_branch2c\"\n\tname: \"res3b7_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b7_branch2c\"\n\ttop: \"res3b7_branch2c\"\n\tname: \"bn3b7_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b7_branch2c\"\n\ttop: \"res3b7_branch2c\"\n\tname: \"scale3b7_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b6\"\n\tbottom: \"res3b7_branch2c\"\n\ttop: \"res3b7\"\n\tname: \"res3b7\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b7\"\n\ttop: \"res3b7\"\n\tname: \"res3b7_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b7\"\n\ttop: \"res4a_branch1\"\n\tname: \"res4a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\ttop: \"res4a_branch1\"\n\tname: \"bn4a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\ttop: \"res4a_branch1\"\n\tname: \"scale4a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b7\"\n\ttop: \"res4a_branch2a\"\n\tname: \"res4a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"bn4a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"scale4a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4a_branch2a\"\n\tbottom: \"res4a_branch2a\"\n\tname: \"res4a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2b\"\n\tname: \"res4a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"bn4a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"scale4a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4a_branch2b\"\n\tbottom: \"res4a_branch2b\"\n\tname: \"res4a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2c\"\n\tname: \"res4a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a_branch2c\"\n\tname: \"bn4a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a_branch2c\"\n\tname: \"scale4a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a\"\n\tname: \"res4a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4a\"\n\ttop: \"res4a\"\n\tname: \"res4a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a\"\n\ttop: \"res4b1_branch2a\"\n\tname: \"res4b1_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b1_branch2a\"\n\ttop: \"res4b1_branch2a\"\n\tname: \"bn4b1_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b1_branch2a\"\n\ttop: \"res4b1_branch2a\"\n\tname: \"scale4b1_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b1_branch2a\"\n\tbottom: \"res4b1_branch2a\"\n\tname: \"res4b1_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b1_branch2a\"\n\ttop: \"res4b1_branch2b\"\n\tname: \"res4b1_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b1_branch2b\"\n\ttop: \"res4b1_branch2b\"\n\tname: \"bn4b1_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b1_branch2b\"\n\ttop: \"res4b1_branch2b\"\n\tname: \"scale4b1_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b1_branch2b\"\n\tbottom: \"res4b1_branch2b\"\n\tname: \"res4b1_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b1_branch2b\"\n\ttop: \"res4b1_branch2c\"\n\tname: \"res4b1_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b1_branch2c\"\n\ttop: \"res4b1_branch2c\"\n\tname: \"bn4b1_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b1_branch2c\"\n\ttop: \"res4b1_branch2c\"\n\tname: \"scale4b1_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a\"\n\tbottom: \"res4b1_branch2c\"\n\ttop: \"res4b1\"\n\tname: \"res4b1\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b1\"\n\ttop: \"res4b1\"\n\tname: \"res4b1_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b1\"\n\ttop: \"res4b2_branch2a\"\n\tname: \"res4b2_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b2_branch2a\"\n\ttop: \"res4b2_branch2a\"\n\tname: \"bn4b2_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b2_branch2a\"\n\ttop: \"res4b2_branch2a\"\n\tname: \"scale4b2_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b2_branch2a\"\n\tbottom: \"res4b2_branch2a\"\n\tname: \"res4b2_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b2_branch2a\"\n\ttop: \"res4b2_branch2b\"\n\tname: \"res4b2_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b2_branch2b\"\n\ttop: \"res4b2_branch2b\"\n\tname: \"bn4b2_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b2_branch2b\"\n\ttop: \"res4b2_branch2b\"\n\tname: \"scale4b2_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b2_branch2b\"\n\tbottom: \"res4b2_branch2b\"\n\tname: \"res4b2_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b2_branch2b\"\n\ttop: \"res4b2_branch2c\"\n\tname: \"res4b2_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b2_branch2c\"\n\ttop: \"res4b2_branch2c\"\n\tname: \"bn4b2_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b2_branch2c\"\n\ttop: \"res4b2_branch2c\"\n\tname: \"scale4b2_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b1\"\n\tbottom: \"res4b2_branch2c\"\n\ttop: \"res4b2\"\n\tname: \"res4b2\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b2\"\n\ttop: \"res4b2\"\n\tname: \"res4b2_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b2\"\n\ttop: \"res4b3_branch2a\"\n\tname: \"res4b3_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b3_branch2a\"\n\ttop: \"res4b3_branch2a\"\n\tname: \"bn4b3_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b3_branch2a\"\n\ttop: \"res4b3_branch2a\"\n\tname: \"scale4b3_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b3_branch2a\"\n\tbottom: \"res4b3_branch2a\"\n\tname: \"res4b3_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b3_branch2a\"\n\ttop: \"res4b3_branch2b\"\n\tname: \"res4b3_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b3_branch2b\"\n\ttop: \"res4b3_branch2b\"\n\tname: \"bn4b3_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b3_branch2b\"\n\ttop: \"res4b3_branch2b\"\n\tname: \"scale4b3_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b3_branch2b\"\n\tbottom: \"res4b3_branch2b\"\n\tname: \"res4b3_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b3_branch2b\"\n\ttop: \"res4b3_branch2c\"\n\tname: \"res4b3_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b3_branch2c\"\n\ttop: \"res4b3_branch2c\"\n\tname: \"bn4b3_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b3_branch2c\"\n\ttop: \"res4b3_branch2c\"\n\tname: \"scale4b3_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b2\"\n\tbottom: \"res4b3_branch2c\"\n\ttop: \"res4b3\"\n\tname: \"res4b3\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b3\"\n\ttop: \"res4b3\"\n\tname: \"res4b3_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b3\"\n\ttop: \"res4b4_branch2a\"\n\tname: \"res4b4_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b4_branch2a\"\n\ttop: \"res4b4_branch2a\"\n\tname: \"bn4b4_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b4_branch2a\"\n\ttop: \"res4b4_branch2a\"\n\tname: \"scale4b4_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b4_branch2a\"\n\tbottom: \"res4b4_branch2a\"\n\tname: \"res4b4_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b4_branch2a\"\n\ttop: \"res4b4_branch2b\"\n\tname: \"res4b4_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b4_branch2b\"\n\ttop: \"res4b4_branch2b\"\n\tname: \"bn4b4_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b4_branch2b\"\n\ttop: \"res4b4_branch2b\"\n\tname: \"scale4b4_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b4_branch2b\"\n\tbottom: \"res4b4_branch2b\"\n\tname: \"res4b4_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b4_branch2b\"\n\ttop: \"res4b4_branch2c\"\n\tname: \"res4b4_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b4_branch2c\"\n\ttop: \"res4b4_branch2c\"\n\tname: \"bn4b4_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b4_branch2c\"\n\ttop: \"res4b4_branch2c\"\n\tname: \"scale4b4_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b3\"\n\tbottom: \"res4b4_branch2c\"\n\ttop: \"res4b4\"\n\tname: \"res4b4\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b4\"\n\ttop: \"res4b4\"\n\tname: \"res4b4_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b4\"\n\ttop: \"res4b5_branch2a\"\n\tname: \"res4b5_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b5_branch2a\"\n\ttop: \"res4b5_branch2a\"\n\tname: \"bn4b5_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b5_branch2a\"\n\ttop: \"res4b5_branch2a\"\n\tname: \"scale4b5_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b5_branch2a\"\n\tbottom: \"res4b5_branch2a\"\n\tname: \"res4b5_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b5_branch2a\"\n\ttop: \"res4b5_branch2b\"\n\tname: \"res4b5_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b5_branch2b\"\n\ttop: \"res4b5_branch2b\"\n\tname: \"bn4b5_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b5_branch2b\"\n\ttop: \"res4b5_branch2b\"\n\tname: \"scale4b5_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b5_branch2b\"\n\tbottom: \"res4b5_branch2b\"\n\tname: \"res4b5_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b5_branch2b\"\n\ttop: \"res4b5_branch2c\"\n\tname: \"res4b5_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b5_branch2c\"\n\ttop: \"res4b5_branch2c\"\n\tname: \"bn4b5_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b5_branch2c\"\n\ttop: \"res4b5_branch2c\"\n\tname: \"scale4b5_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b4\"\n\tbottom: \"res4b5_branch2c\"\n\ttop: \"res4b5\"\n\tname: \"res4b5\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b5\"\n\ttop: \"res4b5\"\n\tname: \"res4b5_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b5\"\n\ttop: \"res4b6_branch2a\"\n\tname: \"res4b6_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b6_branch2a\"\n\ttop: \"res4b6_branch2a\"\n\tname: \"bn4b6_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b6_branch2a\"\n\ttop: \"res4b6_branch2a\"\n\tname: \"scale4b6_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b6_branch2a\"\n\tbottom: \"res4b6_branch2a\"\n\tname: \"res4b6_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b6_branch2a\"\n\ttop: \"res4b6_branch2b\"\n\tname: \"res4b6_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b6_branch2b\"\n\ttop: \"res4b6_branch2b\"\n\tname: \"bn4b6_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b6_branch2b\"\n\ttop: \"res4b6_branch2b\"\n\tname: \"scale4b6_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b6_branch2b\"\n\tbottom: \"res4b6_branch2b\"\n\tname: \"res4b6_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b6_branch2b\"\n\ttop: \"res4b6_branch2c\"\n\tname: \"res4b6_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b6_branch2c\"\n\ttop: \"res4b6_branch2c\"\n\tname: \"bn4b6_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b6_branch2c\"\n\ttop: \"res4b6_branch2c\"\n\tname: \"scale4b6_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b5\"\n\tbottom: \"res4b6_branch2c\"\n\ttop: \"res4b6\"\n\tname: \"res4b6\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b6\"\n\ttop: \"res4b6\"\n\tname: \"res4b6_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b6\"\n\ttop: \"res4b7_branch2a\"\n\tname: \"res4b7_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b7_branch2a\"\n\ttop: \"res4b7_branch2a\"\n\tname: \"bn4b7_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b7_branch2a\"\n\ttop: \"res4b7_branch2a\"\n\tname: \"scale4b7_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b7_branch2a\"\n\tbottom: \"res4b7_branch2a\"\n\tname: \"res4b7_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b7_branch2a\"\n\ttop: \"res4b7_branch2b\"\n\tname: \"res4b7_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b7_branch2b\"\n\ttop: \"res4b7_branch2b\"\n\tname: \"bn4b7_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b7_branch2b\"\n\ttop: \"res4b7_branch2b\"\n\tname: \"scale4b7_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b7_branch2b\"\n\tbottom: \"res4b7_branch2b\"\n\tname: \"res4b7_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b7_branch2b\"\n\ttop: \"res4b7_branch2c\"\n\tname: \"res4b7_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b7_branch2c\"\n\ttop: \"res4b7_branch2c\"\n\tname: \"bn4b7_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b7_branch2c\"\n\ttop: \"res4b7_branch2c\"\n\tname: \"scale4b7_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b6\"\n\tbottom: \"res4b7_branch2c\"\n\ttop: \"res4b7\"\n\tname: \"res4b7\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b7\"\n\ttop: \"res4b7\"\n\tname: \"res4b7_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b7\"\n\ttop: \"res4b8_branch2a\"\n\tname: \"res4b8_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b8_branch2a\"\n\ttop: \"res4b8_branch2a\"\n\tname: \"bn4b8_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b8_branch2a\"\n\ttop: \"res4b8_branch2a\"\n\tname: \"scale4b8_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b8_branch2a\"\n\tbottom: \"res4b8_branch2a\"\n\tname: \"res4b8_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b8_branch2a\"\n\ttop: \"res4b8_branch2b\"\n\tname: \"res4b8_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b8_branch2b\"\n\ttop: \"res4b8_branch2b\"\n\tname: \"bn4b8_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b8_branch2b\"\n\ttop: \"res4b8_branch2b\"\n\tname: \"scale4b8_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b8_branch2b\"\n\tbottom: \"res4b8_branch2b\"\n\tname: \"res4b8_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b8_branch2b\"\n\ttop: \"res4b8_branch2c\"\n\tname: \"res4b8_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b8_branch2c\"\n\ttop: \"res4b8_branch2c\"\n\tname: \"bn4b8_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b8_branch2c\"\n\ttop: \"res4b8_branch2c\"\n\tname: \"scale4b8_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b7\"\n\tbottom: \"res4b8_branch2c\"\n\ttop: \"res4b8\"\n\tname: \"res4b8\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b8\"\n\ttop: \"res4b8\"\n\tname: \"res4b8_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b8\"\n\ttop: \"res4b9_branch2a\"\n\tname: \"res4b9_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b9_branch2a\"\n\ttop: \"res4b9_branch2a\"\n\tname: \"bn4b9_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b9_branch2a\"\n\ttop: \"res4b9_branch2a\"\n\tname: \"scale4b9_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b9_branch2a\"\n\tbottom: \"res4b9_branch2a\"\n\tname: \"res4b9_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b9_branch2a\"\n\ttop: \"res4b9_branch2b\"\n\tname: \"res4b9_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b9_branch2b\"\n\ttop: \"res4b9_branch2b\"\n\tname: \"bn4b9_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b9_branch2b\"\n\ttop: \"res4b9_branch2b\"\n\tname: \"scale4b9_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b9_branch2b\"\n\tbottom: \"res4b9_branch2b\"\n\tname: \"res4b9_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b9_branch2b\"\n\ttop: \"res4b9_branch2c\"\n\tname: \"res4b9_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b9_branch2c\"\n\ttop: \"res4b9_branch2c\"\n\tname: \"bn4b9_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b9_branch2c\"\n\ttop: \"res4b9_branch2c\"\n\tname: \"scale4b9_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b8\"\n\tbottom: \"res4b9_branch2c\"\n\ttop: \"res4b9\"\n\tname: \"res4b9\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b9\"\n\ttop: \"res4b9\"\n\tname: \"res4b9_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b9\"\n\ttop: \"res4b10_branch2a\"\n\tname: \"res4b10_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b10_branch2a\"\n\ttop: \"res4b10_branch2a\"\n\tname: \"bn4b10_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b10_branch2a\"\n\ttop: \"res4b10_branch2a\"\n\tname: \"scale4b10_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b10_branch2a\"\n\tbottom: \"res4b10_branch2a\"\n\tname: \"res4b10_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b10_branch2a\"\n\ttop: \"res4b10_branch2b\"\n\tname: \"res4b10_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b10_branch2b\"\n\ttop: \"res4b10_branch2b\"\n\tname: \"bn4b10_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b10_branch2b\"\n\ttop: \"res4b10_branch2b\"\n\tname: \"scale4b10_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b10_branch2b\"\n\tbottom: \"res4b10_branch2b\"\n\tname: \"res4b10_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b10_branch2b\"\n\ttop: \"res4b10_branch2c\"\n\tname: \"res4b10_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b10_branch2c\"\n\ttop: \"res4b10_branch2c\"\n\tname: \"bn4b10_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b10_branch2c\"\n\ttop: \"res4b10_branch2c\"\n\tname: \"scale4b10_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b9\"\n\tbottom: \"res4b10_branch2c\"\n\ttop: \"res4b10\"\n\tname: \"res4b10\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b10\"\n\ttop: \"res4b10\"\n\tname: \"res4b10_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b10\"\n\ttop: \"res4b11_branch2a\"\n\tname: \"res4b11_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b11_branch2a\"\n\ttop: \"res4b11_branch2a\"\n\tname: \"bn4b11_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b11_branch2a\"\n\ttop: \"res4b11_branch2a\"\n\tname: \"scale4b11_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b11_branch2a\"\n\tbottom: \"res4b11_branch2a\"\n\tname: \"res4b11_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b11_branch2a\"\n\ttop: \"res4b11_branch2b\"\n\tname: \"res4b11_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b11_branch2b\"\n\ttop: \"res4b11_branch2b\"\n\tname: \"bn4b11_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b11_branch2b\"\n\ttop: \"res4b11_branch2b\"\n\tname: \"scale4b11_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b11_branch2b\"\n\tbottom: \"res4b11_branch2b\"\n\tname: \"res4b11_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b11_branch2b\"\n\ttop: \"res4b11_branch2c\"\n\tname: \"res4b11_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b11_branch2c\"\n\ttop: \"res4b11_branch2c\"\n\tname: \"bn4b11_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b11_branch2c\"\n\ttop: \"res4b11_branch2c\"\n\tname: \"scale4b11_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b10\"\n\tbottom: \"res4b11_branch2c\"\n\ttop: \"res4b11\"\n\tname: \"res4b11\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b11\"\n\ttop: \"res4b11\"\n\tname: \"res4b11_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b11\"\n\ttop: \"res4b12_branch2a\"\n\tname: \"res4b12_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b12_branch2a\"\n\ttop: \"res4b12_branch2a\"\n\tname: \"bn4b12_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b12_branch2a\"\n\ttop: \"res4b12_branch2a\"\n\tname: \"scale4b12_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b12_branch2a\"\n\tbottom: \"res4b12_branch2a\"\n\tname: \"res4b12_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b12_branch2a\"\n\ttop: \"res4b12_branch2b\"\n\tname: \"res4b12_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b12_branch2b\"\n\ttop: \"res4b12_branch2b\"\n\tname: \"bn4b12_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b12_branch2b\"\n\ttop: \"res4b12_branch2b\"\n\tname: \"scale4b12_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b12_branch2b\"\n\tbottom: \"res4b12_branch2b\"\n\tname: \"res4b12_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b12_branch2b\"\n\ttop: \"res4b12_branch2c\"\n\tname: \"res4b12_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b12_branch2c\"\n\ttop: \"res4b12_branch2c\"\n\tname: \"bn4b12_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b12_branch2c\"\n\ttop: \"res4b12_branch2c\"\n\tname: \"scale4b12_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b11\"\n\tbottom: \"res4b12_branch2c\"\n\ttop: \"res4b12\"\n\tname: \"res4b12\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b12\"\n\ttop: \"res4b12\"\n\tname: \"res4b12_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b12\"\n\ttop: \"res4b13_branch2a\"\n\tname: \"res4b13_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b13_branch2a\"\n\ttop: \"res4b13_branch2a\"\n\tname: \"bn4b13_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b13_branch2a\"\n\ttop: \"res4b13_branch2a\"\n\tname: \"scale4b13_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b13_branch2a\"\n\tbottom: \"res4b13_branch2a\"\n\tname: \"res4b13_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b13_branch2a\"\n\ttop: \"res4b13_branch2b\"\n\tname: \"res4b13_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b13_branch2b\"\n\ttop: \"res4b13_branch2b\"\n\tname: \"bn4b13_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b13_branch2b\"\n\ttop: \"res4b13_branch2b\"\n\tname: \"scale4b13_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b13_branch2b\"\n\tbottom: \"res4b13_branch2b\"\n\tname: \"res4b13_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b13_branch2b\"\n\ttop: \"res4b13_branch2c\"\n\tname: \"res4b13_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b13_branch2c\"\n\ttop: \"res4b13_branch2c\"\n\tname: \"bn4b13_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b13_branch2c\"\n\ttop: \"res4b13_branch2c\"\n\tname: \"scale4b13_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b12\"\n\tbottom: \"res4b13_branch2c\"\n\ttop: \"res4b13\"\n\tname: \"res4b13\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b13\"\n\ttop: \"res4b13\"\n\tname: \"res4b13_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b13\"\n\ttop: \"res4b14_branch2a\"\n\tname: \"res4b14_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b14_branch2a\"\n\ttop: \"res4b14_branch2a\"\n\tname: \"bn4b14_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b14_branch2a\"\n\ttop: \"res4b14_branch2a\"\n\tname: \"scale4b14_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b14_branch2a\"\n\tbottom: \"res4b14_branch2a\"\n\tname: \"res4b14_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b14_branch2a\"\n\ttop: \"res4b14_branch2b\"\n\tname: \"res4b14_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b14_branch2b\"\n\ttop: \"res4b14_branch2b\"\n\tname: \"bn4b14_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b14_branch2b\"\n\ttop: \"res4b14_branch2b\"\n\tname: \"scale4b14_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b14_branch2b\"\n\tbottom: \"res4b14_branch2b\"\n\tname: \"res4b14_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b14_branch2b\"\n\ttop: \"res4b14_branch2c\"\n\tname: \"res4b14_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b14_branch2c\"\n\ttop: \"res4b14_branch2c\"\n\tname: \"bn4b14_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b14_branch2c\"\n\ttop: \"res4b14_branch2c\"\n\tname: \"scale4b14_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b13\"\n\tbottom: \"res4b14_branch2c\"\n\ttop: \"res4b14\"\n\tname: \"res4b14\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b14\"\n\ttop: \"res4b14\"\n\tname: \"res4b14_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b14\"\n\ttop: \"res4b15_branch2a\"\n\tname: \"res4b15_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b15_branch2a\"\n\ttop: \"res4b15_branch2a\"\n\tname: \"bn4b15_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b15_branch2a\"\n\ttop: \"res4b15_branch2a\"\n\tname: \"scale4b15_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b15_branch2a\"\n\tbottom: \"res4b15_branch2a\"\n\tname: \"res4b15_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b15_branch2a\"\n\ttop: \"res4b15_branch2b\"\n\tname: \"res4b15_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b15_branch2b\"\n\ttop: \"res4b15_branch2b\"\n\tname: \"bn4b15_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b15_branch2b\"\n\ttop: \"res4b15_branch2b\"\n\tname: \"scale4b15_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b15_branch2b\"\n\tbottom: \"res4b15_branch2b\"\n\tname: \"res4b15_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b15_branch2b\"\n\ttop: \"res4b15_branch2c\"\n\tname: \"res4b15_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b15_branch2c\"\n\ttop: \"res4b15_branch2c\"\n\tname: \"bn4b15_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b15_branch2c\"\n\ttop: \"res4b15_branch2c\"\n\tname: \"scale4b15_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b14\"\n\tbottom: \"res4b15_branch2c\"\n\ttop: \"res4b15\"\n\tname: \"res4b15\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b15\"\n\ttop: \"res4b15\"\n\tname: \"res4b15_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b15\"\n\ttop: \"res4b16_branch2a\"\n\tname: \"res4b16_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b16_branch2a\"\n\ttop: \"res4b16_branch2a\"\n\tname: \"bn4b16_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b16_branch2a\"\n\ttop: \"res4b16_branch2a\"\n\tname: \"scale4b16_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b16_branch2a\"\n\tbottom: \"res4b16_branch2a\"\n\tname: \"res4b16_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b16_branch2a\"\n\ttop: \"res4b16_branch2b\"\n\tname: \"res4b16_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b16_branch2b\"\n\ttop: \"res4b16_branch2b\"\n\tname: \"bn4b16_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b16_branch2b\"\n\ttop: \"res4b16_branch2b\"\n\tname: \"scale4b16_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b16_branch2b\"\n\tbottom: \"res4b16_branch2b\"\n\tname: \"res4b16_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b16_branch2b\"\n\ttop: \"res4b16_branch2c\"\n\tname: \"res4b16_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b16_branch2c\"\n\ttop: \"res4b16_branch2c\"\n\tname: \"bn4b16_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b16_branch2c\"\n\ttop: \"res4b16_branch2c\"\n\tname: \"scale4b16_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b15\"\n\tbottom: \"res4b16_branch2c\"\n\ttop: \"res4b16\"\n\tname: \"res4b16\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b16\"\n\ttop: \"res4b16\"\n\tname: \"res4b16_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b16\"\n\ttop: \"res4b17_branch2a\"\n\tname: \"res4b17_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b17_branch2a\"\n\ttop: \"res4b17_branch2a\"\n\tname: \"bn4b17_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b17_branch2a\"\n\ttop: \"res4b17_branch2a\"\n\tname: \"scale4b17_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b17_branch2a\"\n\tbottom: \"res4b17_branch2a\"\n\tname: \"res4b17_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b17_branch2a\"\n\ttop: \"res4b17_branch2b\"\n\tname: \"res4b17_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b17_branch2b\"\n\ttop: \"res4b17_branch2b\"\n\tname: \"bn4b17_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b17_branch2b\"\n\ttop: \"res4b17_branch2b\"\n\tname: \"scale4b17_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b17_branch2b\"\n\tbottom: \"res4b17_branch2b\"\n\tname: \"res4b17_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b17_branch2b\"\n\ttop: \"res4b17_branch2c\"\n\tname: \"res4b17_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b17_branch2c\"\n\ttop: \"res4b17_branch2c\"\n\tname: \"bn4b17_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b17_branch2c\"\n\ttop: \"res4b17_branch2c\"\n\tname: \"scale4b17_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b16\"\n\tbottom: \"res4b17_branch2c\"\n\ttop: \"res4b17\"\n\tname: \"res4b17\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b17\"\n\ttop: \"res4b17\"\n\tname: \"res4b17_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b17\"\n\ttop: \"res4b18_branch2a\"\n\tname: \"res4b18_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b18_branch2a\"\n\ttop: \"res4b18_branch2a\"\n\tname: \"bn4b18_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b18_branch2a\"\n\ttop: \"res4b18_branch2a\"\n\tname: \"scale4b18_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b18_branch2a\"\n\tbottom: \"res4b18_branch2a\"\n\tname: \"res4b18_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b18_branch2a\"\n\ttop: \"res4b18_branch2b\"\n\tname: \"res4b18_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b18_branch2b\"\n\ttop: \"res4b18_branch2b\"\n\tname: \"bn4b18_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b18_branch2b\"\n\ttop: \"res4b18_branch2b\"\n\tname: \"scale4b18_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b18_branch2b\"\n\tbottom: \"res4b18_branch2b\"\n\tname: \"res4b18_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b18_branch2b\"\n\ttop: \"res4b18_branch2c\"\n\tname: \"res4b18_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b18_branch2c\"\n\ttop: \"res4b18_branch2c\"\n\tname: \"bn4b18_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b18_branch2c\"\n\ttop: \"res4b18_branch2c\"\n\tname: \"scale4b18_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b17\"\n\tbottom: \"res4b18_branch2c\"\n\ttop: \"res4b18\"\n\tname: \"res4b18\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b18\"\n\ttop: \"res4b18\"\n\tname: \"res4b18_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b18\"\n\ttop: \"res4b19_branch2a\"\n\tname: \"res4b19_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b19_branch2a\"\n\ttop: \"res4b19_branch2a\"\n\tname: \"bn4b19_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b19_branch2a\"\n\ttop: \"res4b19_branch2a\"\n\tname: \"scale4b19_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b19_branch2a\"\n\tbottom: \"res4b19_branch2a\"\n\tname: \"res4b19_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b19_branch2a\"\n\ttop: \"res4b19_branch2b\"\n\tname: \"res4b19_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b19_branch2b\"\n\ttop: \"res4b19_branch2b\"\n\tname: \"bn4b19_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b19_branch2b\"\n\ttop: \"res4b19_branch2b\"\n\tname: \"scale4b19_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b19_branch2b\"\n\tbottom: \"res4b19_branch2b\"\n\tname: \"res4b19_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b19_branch2b\"\n\ttop: \"res4b19_branch2c\"\n\tname: \"res4b19_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b19_branch2c\"\n\ttop: \"res4b19_branch2c\"\n\tname: \"bn4b19_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b19_branch2c\"\n\ttop: \"res4b19_branch2c\"\n\tname: \"scale4b19_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b18\"\n\tbottom: \"res4b19_branch2c\"\n\ttop: \"res4b19\"\n\tname: \"res4b19\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b19\"\n\ttop: \"res4b19\"\n\tname: \"res4b19_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b19\"\n\ttop: \"res4b20_branch2a\"\n\tname: \"res4b20_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b20_branch2a\"\n\ttop: \"res4b20_branch2a\"\n\tname: \"bn4b20_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b20_branch2a\"\n\ttop: \"res4b20_branch2a\"\n\tname: \"scale4b20_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b20_branch2a\"\n\tbottom: \"res4b20_branch2a\"\n\tname: \"res4b20_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b20_branch2a\"\n\ttop: \"res4b20_branch2b\"\n\tname: \"res4b20_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b20_branch2b\"\n\ttop: \"res4b20_branch2b\"\n\tname: \"bn4b20_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b20_branch2b\"\n\ttop: \"res4b20_branch2b\"\n\tname: \"scale4b20_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b20_branch2b\"\n\tbottom: \"res4b20_branch2b\"\n\tname: \"res4b20_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b20_branch2b\"\n\ttop: \"res4b20_branch2c\"\n\tname: \"res4b20_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b20_branch2c\"\n\ttop: \"res4b20_branch2c\"\n\tname: \"bn4b20_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b20_branch2c\"\n\ttop: \"res4b20_branch2c\"\n\tname: \"scale4b20_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b19\"\n\tbottom: \"res4b20_branch2c\"\n\ttop: \"res4b20\"\n\tname: \"res4b20\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b20\"\n\ttop: \"res4b20\"\n\tname: \"res4b20_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b20\"\n\ttop: \"res4b21_branch2a\"\n\tname: \"res4b21_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b21_branch2a\"\n\ttop: \"res4b21_branch2a\"\n\tname: \"bn4b21_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b21_branch2a\"\n\ttop: \"res4b21_branch2a\"\n\tname: \"scale4b21_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b21_branch2a\"\n\tbottom: \"res4b21_branch2a\"\n\tname: \"res4b21_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b21_branch2a\"\n\ttop: \"res4b21_branch2b\"\n\tname: \"res4b21_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b21_branch2b\"\n\ttop: \"res4b21_branch2b\"\n\tname: \"bn4b21_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b21_branch2b\"\n\ttop: \"res4b21_branch2b\"\n\tname: \"scale4b21_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b21_branch2b\"\n\tbottom: \"res4b21_branch2b\"\n\tname: \"res4b21_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b21_branch2b\"\n\ttop: \"res4b21_branch2c\"\n\tname: \"res4b21_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b21_branch2c\"\n\ttop: \"res4b21_branch2c\"\n\tname: \"bn4b21_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b21_branch2c\"\n\ttop: \"res4b21_branch2c\"\n\tname: \"scale4b21_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b20\"\n\tbottom: \"res4b21_branch2c\"\n\ttop: \"res4b21\"\n\tname: \"res4b21\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b21\"\n\ttop: \"res4b21\"\n\tname: \"res4b21_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b21\"\n\ttop: \"res4b22_branch2a\"\n\tname: \"res4b22_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b22_branch2a\"\n\ttop: \"res4b22_branch2a\"\n\tname: \"bn4b22_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b22_branch2a\"\n\ttop: \"res4b22_branch2a\"\n\tname: \"scale4b22_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b22_branch2a\"\n\tbottom: \"res4b22_branch2a\"\n\tname: \"res4b22_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b22_branch2a\"\n\ttop: \"res4b22_branch2b\"\n\tname: \"res4b22_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b22_branch2b\"\n\ttop: \"res4b22_branch2b\"\n\tname: \"bn4b22_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b22_branch2b\"\n\ttop: \"res4b22_branch2b\"\n\tname: \"scale4b22_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b22_branch2b\"\n\tbottom: \"res4b22_branch2b\"\n\tname: \"res4b22_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b22_branch2b\"\n\ttop: \"res4b22_branch2c\"\n\tname: \"res4b22_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b22_branch2c\"\n\ttop: \"res4b22_branch2c\"\n\tname: \"bn4b22_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b22_branch2c\"\n\ttop: \"res4b22_branch2c\"\n\tname: \"scale4b22_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b21\"\n\tbottom: \"res4b22_branch2c\"\n\ttop: \"res4b22\"\n\tname: \"res4b22\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b22\"\n\ttop: \"res4b22\"\n\tname: \"res4b22_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b22\"\n\ttop: \"res4b23_branch2a\"\n\tname: \"res4b23_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b23_branch2a\"\n\ttop: \"res4b23_branch2a\"\n\tname: \"bn4b23_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b23_branch2a\"\n\ttop: \"res4b23_branch2a\"\n\tname: \"scale4b23_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b23_branch2a\"\n\tbottom: \"res4b23_branch2a\"\n\tname: \"res4b23_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b23_branch2a\"\n\ttop: \"res4b23_branch2b\"\n\tname: \"res4b23_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b23_branch2b\"\n\ttop: \"res4b23_branch2b\"\n\tname: \"bn4b23_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b23_branch2b\"\n\ttop: \"res4b23_branch2b\"\n\tname: \"scale4b23_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b23_branch2b\"\n\tbottom: \"res4b23_branch2b\"\n\tname: \"res4b23_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b23_branch2b\"\n\ttop: \"res4b23_branch2c\"\n\tname: \"res4b23_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b23_branch2c\"\n\ttop: \"res4b23_branch2c\"\n\tname: \"bn4b23_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b23_branch2c\"\n\ttop: \"res4b23_branch2c\"\n\tname: \"scale4b23_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b22\"\n\tbottom: \"res4b23_branch2c\"\n\ttop: \"res4b23\"\n\tname: \"res4b23\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b23\"\n\ttop: \"res4b23\"\n\tname: \"res4b23_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b23\"\n\ttop: \"res4b24_branch2a\"\n\tname: \"res4b24_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b24_branch2a\"\n\ttop: \"res4b24_branch2a\"\n\tname: \"bn4b24_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b24_branch2a\"\n\ttop: \"res4b24_branch2a\"\n\tname: \"scale4b24_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b24_branch2a\"\n\tbottom: \"res4b24_branch2a\"\n\tname: \"res4b24_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b24_branch2a\"\n\ttop: \"res4b24_branch2b\"\n\tname: \"res4b24_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b24_branch2b\"\n\ttop: \"res4b24_branch2b\"\n\tname: \"bn4b24_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b24_branch2b\"\n\ttop: \"res4b24_branch2b\"\n\tname: \"scale4b24_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b24_branch2b\"\n\tbottom: \"res4b24_branch2b\"\n\tname: \"res4b24_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b24_branch2b\"\n\ttop: \"res4b24_branch2c\"\n\tname: \"res4b24_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b24_branch2c\"\n\ttop: \"res4b24_branch2c\"\n\tname: \"bn4b24_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b24_branch2c\"\n\ttop: \"res4b24_branch2c\"\n\tname: \"scale4b24_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b23\"\n\tbottom: \"res4b24_branch2c\"\n\ttop: \"res4b24\"\n\tname: \"res4b24\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b24\"\n\ttop: \"res4b24\"\n\tname: \"res4b24_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b24\"\n\ttop: \"res4b25_branch2a\"\n\tname: \"res4b25_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b25_branch2a\"\n\ttop: \"res4b25_branch2a\"\n\tname: \"bn4b25_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b25_branch2a\"\n\ttop: \"res4b25_branch2a\"\n\tname: \"scale4b25_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b25_branch2a\"\n\tbottom: \"res4b25_branch2a\"\n\tname: \"res4b25_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b25_branch2a\"\n\ttop: \"res4b25_branch2b\"\n\tname: \"res4b25_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b25_branch2b\"\n\ttop: \"res4b25_branch2b\"\n\tname: \"bn4b25_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b25_branch2b\"\n\ttop: \"res4b25_branch2b\"\n\tname: \"scale4b25_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b25_branch2b\"\n\tbottom: \"res4b25_branch2b\"\n\tname: \"res4b25_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b25_branch2b\"\n\ttop: \"res4b25_branch2c\"\n\tname: \"res4b25_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b25_branch2c\"\n\ttop: \"res4b25_branch2c\"\n\tname: \"bn4b25_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b25_branch2c\"\n\ttop: \"res4b25_branch2c\"\n\tname: \"scale4b25_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b24\"\n\tbottom: \"res4b25_branch2c\"\n\ttop: \"res4b25\"\n\tname: \"res4b25\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b25\"\n\ttop: \"res4b25\"\n\tname: \"res4b25_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b25\"\n\ttop: \"res4b26_branch2a\"\n\tname: \"res4b26_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b26_branch2a\"\n\ttop: \"res4b26_branch2a\"\n\tname: \"bn4b26_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b26_branch2a\"\n\ttop: \"res4b26_branch2a\"\n\tname: \"scale4b26_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b26_branch2a\"\n\tbottom: \"res4b26_branch2a\"\n\tname: \"res4b26_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b26_branch2a\"\n\ttop: \"res4b26_branch2b\"\n\tname: \"res4b26_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b26_branch2b\"\n\ttop: \"res4b26_branch2b\"\n\tname: \"bn4b26_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b26_branch2b\"\n\ttop: \"res4b26_branch2b\"\n\tname: \"scale4b26_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b26_branch2b\"\n\tbottom: \"res4b26_branch2b\"\n\tname: \"res4b26_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b26_branch2b\"\n\ttop: \"res4b26_branch2c\"\n\tname: \"res4b26_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b26_branch2c\"\n\ttop: \"res4b26_branch2c\"\n\tname: \"bn4b26_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b26_branch2c\"\n\ttop: \"res4b26_branch2c\"\n\tname: \"scale4b26_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b25\"\n\tbottom: \"res4b26_branch2c\"\n\ttop: \"res4b26\"\n\tname: \"res4b26\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b26\"\n\ttop: \"res4b26\"\n\tname: \"res4b26_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b26\"\n\ttop: \"res4b27_branch2a\"\n\tname: \"res4b27_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b27_branch2a\"\n\ttop: \"res4b27_branch2a\"\n\tname: \"bn4b27_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b27_branch2a\"\n\ttop: \"res4b27_branch2a\"\n\tname: \"scale4b27_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b27_branch2a\"\n\tbottom: \"res4b27_branch2a\"\n\tname: \"res4b27_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b27_branch2a\"\n\ttop: \"res4b27_branch2b\"\n\tname: \"res4b27_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b27_branch2b\"\n\ttop: \"res4b27_branch2b\"\n\tname: \"bn4b27_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b27_branch2b\"\n\ttop: \"res4b27_branch2b\"\n\tname: \"scale4b27_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b27_branch2b\"\n\tbottom: \"res4b27_branch2b\"\n\tname: \"res4b27_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b27_branch2b\"\n\ttop: \"res4b27_branch2c\"\n\tname: \"res4b27_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b27_branch2c\"\n\ttop: \"res4b27_branch2c\"\n\tname: \"bn4b27_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b27_branch2c\"\n\ttop: \"res4b27_branch2c\"\n\tname: \"scale4b27_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b26\"\n\tbottom: \"res4b27_branch2c\"\n\ttop: \"res4b27\"\n\tname: \"res4b27\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b27\"\n\ttop: \"res4b27\"\n\tname: \"res4b27_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b27\"\n\ttop: \"res4b28_branch2a\"\n\tname: \"res4b28_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b28_branch2a\"\n\ttop: \"res4b28_branch2a\"\n\tname: \"bn4b28_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b28_branch2a\"\n\ttop: \"res4b28_branch2a\"\n\tname: \"scale4b28_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b28_branch2a\"\n\tbottom: \"res4b28_branch2a\"\n\tname: \"res4b28_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b28_branch2a\"\n\ttop: \"res4b28_branch2b\"\n\tname: \"res4b28_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b28_branch2b\"\n\ttop: \"res4b28_branch2b\"\n\tname: \"bn4b28_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b28_branch2b\"\n\ttop: \"res4b28_branch2b\"\n\tname: \"scale4b28_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b28_branch2b\"\n\tbottom: \"res4b28_branch2b\"\n\tname: \"res4b28_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b28_branch2b\"\n\ttop: \"res4b28_branch2c\"\n\tname: \"res4b28_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b28_branch2c\"\n\ttop: \"res4b28_branch2c\"\n\tname: \"bn4b28_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b28_branch2c\"\n\ttop: \"res4b28_branch2c\"\n\tname: \"scale4b28_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b27\"\n\tbottom: \"res4b28_branch2c\"\n\ttop: \"res4b28\"\n\tname: \"res4b28\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b28\"\n\ttop: \"res4b28\"\n\tname: \"res4b28_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b28\"\n\ttop: \"res4b29_branch2a\"\n\tname: \"res4b29_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b29_branch2a\"\n\ttop: \"res4b29_branch2a\"\n\tname: \"bn4b29_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b29_branch2a\"\n\ttop: \"res4b29_branch2a\"\n\tname: \"scale4b29_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b29_branch2a\"\n\tbottom: \"res4b29_branch2a\"\n\tname: \"res4b29_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b29_branch2a\"\n\ttop: \"res4b29_branch2b\"\n\tname: \"res4b29_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b29_branch2b\"\n\ttop: \"res4b29_branch2b\"\n\tname: \"bn4b29_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b29_branch2b\"\n\ttop: \"res4b29_branch2b\"\n\tname: \"scale4b29_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b29_branch2b\"\n\tbottom: \"res4b29_branch2b\"\n\tname: \"res4b29_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b29_branch2b\"\n\ttop: \"res4b29_branch2c\"\n\tname: \"res4b29_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b29_branch2c\"\n\ttop: \"res4b29_branch2c\"\n\tname: \"bn4b29_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b29_branch2c\"\n\ttop: \"res4b29_branch2c\"\n\tname: \"scale4b29_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b28\"\n\tbottom: \"res4b29_branch2c\"\n\ttop: \"res4b29\"\n\tname: \"res4b29\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b29\"\n\ttop: \"res4b29\"\n\tname: \"res4b29_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b29\"\n\ttop: \"res4b30_branch2a\"\n\tname: \"res4b30_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b30_branch2a\"\n\ttop: \"res4b30_branch2a\"\n\tname: \"bn4b30_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b30_branch2a\"\n\ttop: \"res4b30_branch2a\"\n\tname: \"scale4b30_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b30_branch2a\"\n\tbottom: \"res4b30_branch2a\"\n\tname: \"res4b30_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b30_branch2a\"\n\ttop: \"res4b30_branch2b\"\n\tname: \"res4b30_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b30_branch2b\"\n\ttop: \"res4b30_branch2b\"\n\tname: \"bn4b30_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b30_branch2b\"\n\ttop: \"res4b30_branch2b\"\n\tname: \"scale4b30_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b30_branch2b\"\n\tbottom: \"res4b30_branch2b\"\n\tname: \"res4b30_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b30_branch2b\"\n\ttop: \"res4b30_branch2c\"\n\tname: \"res4b30_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b30_branch2c\"\n\ttop: \"res4b30_branch2c\"\n\tname: \"bn4b30_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b30_branch2c\"\n\ttop: \"res4b30_branch2c\"\n\tname: \"scale4b30_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b29\"\n\tbottom: \"res4b30_branch2c\"\n\ttop: \"res4b30\"\n\tname: \"res4b30\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b30\"\n\ttop: \"res4b30\"\n\tname: \"res4b30_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b30\"\n\ttop: \"res4b31_branch2a\"\n\tname: \"res4b31_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b31_branch2a\"\n\ttop: \"res4b31_branch2a\"\n\tname: \"bn4b31_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b31_branch2a\"\n\ttop: \"res4b31_branch2a\"\n\tname: \"scale4b31_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b31_branch2a\"\n\tbottom: \"res4b31_branch2a\"\n\tname: \"res4b31_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b31_branch2a\"\n\ttop: \"res4b31_branch2b\"\n\tname: \"res4b31_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b31_branch2b\"\n\ttop: \"res4b31_branch2b\"\n\tname: \"bn4b31_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b31_branch2b\"\n\ttop: \"res4b31_branch2b\"\n\tname: \"scale4b31_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b31_branch2b\"\n\tbottom: \"res4b31_branch2b\"\n\tname: \"res4b31_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b31_branch2b\"\n\ttop: \"res4b31_branch2c\"\n\tname: \"res4b31_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b31_branch2c\"\n\ttop: \"res4b31_branch2c\"\n\tname: \"bn4b31_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b31_branch2c\"\n\ttop: \"res4b31_branch2c\"\n\tname: \"scale4b31_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b30\"\n\tbottom: \"res4b31_branch2c\"\n\ttop: \"res4b31\"\n\tname: \"res4b31\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b31\"\n\ttop: \"res4b31\"\n\tname: \"res4b31_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b31\"\n\ttop: \"res4b32_branch2a\"\n\tname: \"res4b32_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b32_branch2a\"\n\ttop: \"res4b32_branch2a\"\n\tname: \"bn4b32_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b32_branch2a\"\n\ttop: \"res4b32_branch2a\"\n\tname: \"scale4b32_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b32_branch2a\"\n\tbottom: \"res4b32_branch2a\"\n\tname: \"res4b32_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b32_branch2a\"\n\ttop: \"res4b32_branch2b\"\n\tname: \"res4b32_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b32_branch2b\"\n\ttop: \"res4b32_branch2b\"\n\tname: \"bn4b32_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b32_branch2b\"\n\ttop: \"res4b32_branch2b\"\n\tname: \"scale4b32_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b32_branch2b\"\n\tbottom: \"res4b32_branch2b\"\n\tname: \"res4b32_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b32_branch2b\"\n\ttop: \"res4b32_branch2c\"\n\tname: \"res4b32_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b32_branch2c\"\n\ttop: \"res4b32_branch2c\"\n\tname: \"bn4b32_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b32_branch2c\"\n\ttop: \"res4b32_branch2c\"\n\tname: \"scale4b32_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b31\"\n\tbottom: \"res4b32_branch2c\"\n\ttop: \"res4b32\"\n\tname: \"res4b32\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b32\"\n\ttop: \"res4b32\"\n\tname: \"res4b32_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b32\"\n\ttop: \"res4b33_branch2a\"\n\tname: \"res4b33_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b33_branch2a\"\n\ttop: \"res4b33_branch2a\"\n\tname: \"bn4b33_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b33_branch2a\"\n\ttop: \"res4b33_branch2a\"\n\tname: \"scale4b33_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b33_branch2a\"\n\tbottom: \"res4b33_branch2a\"\n\tname: \"res4b33_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b33_branch2a\"\n\ttop: \"res4b33_branch2b\"\n\tname: \"res4b33_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b33_branch2b\"\n\ttop: \"res4b33_branch2b\"\n\tname: \"bn4b33_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b33_branch2b\"\n\ttop: \"res4b33_branch2b\"\n\tname: \"scale4b33_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b33_branch2b\"\n\tbottom: \"res4b33_branch2b\"\n\tname: \"res4b33_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b33_branch2b\"\n\ttop: \"res4b33_branch2c\"\n\tname: \"res4b33_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b33_branch2c\"\n\ttop: \"res4b33_branch2c\"\n\tname: \"bn4b33_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b33_branch2c\"\n\ttop: \"res4b33_branch2c\"\n\tname: \"scale4b33_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b32\"\n\tbottom: \"res4b33_branch2c\"\n\ttop: \"res4b33\"\n\tname: \"res4b33\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b33\"\n\ttop: \"res4b33\"\n\tname: \"res4b33_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b33\"\n\ttop: \"res4b34_branch2a\"\n\tname: \"res4b34_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b34_branch2a\"\n\ttop: \"res4b34_branch2a\"\n\tname: \"bn4b34_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b34_branch2a\"\n\ttop: \"res4b34_branch2a\"\n\tname: \"scale4b34_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b34_branch2a\"\n\tbottom: \"res4b34_branch2a\"\n\tname: \"res4b34_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b34_branch2a\"\n\ttop: \"res4b34_branch2b\"\n\tname: \"res4b34_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b34_branch2b\"\n\ttop: \"res4b34_branch2b\"\n\tname: \"bn4b34_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b34_branch2b\"\n\ttop: \"res4b34_branch2b\"\n\tname: \"scale4b34_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b34_branch2b\"\n\tbottom: \"res4b34_branch2b\"\n\tname: \"res4b34_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b34_branch2b\"\n\ttop: \"res4b34_branch2c\"\n\tname: \"res4b34_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b34_branch2c\"\n\ttop: \"res4b34_branch2c\"\n\tname: \"bn4b34_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b34_branch2c\"\n\ttop: \"res4b34_branch2c\"\n\tname: \"scale4b34_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b33\"\n\tbottom: \"res4b34_branch2c\"\n\ttop: \"res4b34\"\n\tname: \"res4b34\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b34\"\n\ttop: \"res4b34\"\n\tname: \"res4b34_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b34\"\n\ttop: \"res4b35_branch2a\"\n\tname: \"res4b35_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b35_branch2a\"\n\ttop: \"res4b35_branch2a\"\n\tname: \"bn4b35_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b35_branch2a\"\n\ttop: \"res4b35_branch2a\"\n\tname: \"scale4b35_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b35_branch2a\"\n\tbottom: \"res4b35_branch2a\"\n\tname: \"res4b35_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b35_branch2a\"\n\ttop: \"res4b35_branch2b\"\n\tname: \"res4b35_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b35_branch2b\"\n\ttop: \"res4b35_branch2b\"\n\tname: \"bn4b35_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b35_branch2b\"\n\ttop: \"res4b35_branch2b\"\n\tname: \"scale4b35_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res4b35_branch2b\"\n\tbottom: \"res4b35_branch2b\"\n\tname: \"res4b35_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b35_branch2b\"\n\ttop: \"res4b35_branch2c\"\n\tname: \"res4b35_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b35_branch2c\"\n\ttop: \"res4b35_branch2c\"\n\tname: \"bn4b35_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b35_branch2c\"\n\ttop: \"res4b35_branch2c\"\n\tname: \"scale4b35_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b34\"\n\tbottom: \"res4b35_branch2c\"\n\ttop: \"res4b35\"\n\tname: \"res4b35\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b35\"\n\ttop: \"res4b35\"\n\tname: \"res4b35_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b35\"\n\ttop: \"res5a_branch1\"\n\tname: \"res5a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\ttop: \"res5a_branch1\"\n\tname: \"bn5a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\ttop: \"res5a_branch1\"\n\tname: \"scale5a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b35\"\n\ttop: \"res5a_branch2a\"\n\tname: \"res5a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"bn5a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"scale5a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res5a_branch2a\"\n\tbottom: \"res5a_branch2a\"\n\tname: \"res5a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2b\"\n\tname: \"res5a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"bn5a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"scale5a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res5a_branch2b\"\n\tbottom: \"res5a_branch2b\"\n\tname: \"res5a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2c\"\n\tname: \"res5a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a_branch2c\"\n\tname: \"bn5a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a_branch2c\"\n\tname: \"scale5a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a\"\n\tname: \"res5a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5a\"\n\ttop: \"res5a\"\n\tname: \"res5a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"res5b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"bn5b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"scale5b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res5b_branch2a\"\n\tbottom: \"res5b_branch2a\"\n\tname: \"res5b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2b\"\n\tname: \"res5b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"bn5b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"scale5b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res5b_branch2b\"\n\tbottom: \"res5b_branch2b\"\n\tname: \"res5b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2c\"\n\tname: \"res5b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b_branch2c\"\n\tname: \"bn5b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b_branch2c\"\n\tname: \"scale5b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a\"\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b\"\n\tname: \"res5b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5b\"\n\ttop: \"res5b\"\n\tname: \"res5b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b\"\n\ttop: \"res5c_branch2a\"\n\tname: \"res5c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"bn5c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"scale5c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res5c_branch2a\"\n\tbottom: \"res5c_branch2a\"\n\tname: \"res5c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2b\"\n\tname: \"res5c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"bn5c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"scale5c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\ttop: \"res5c_branch2b\"\n\tbottom: \"res5c_branch2b\"\n\tname: \"res5c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2c\"\n\tname: \"res5c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c_branch2c\"\n\tname: \"bn5c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c_branch2c\"\n\tname: \"scale5c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b\"\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c\"\n\tname: \"res5c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5c\"\n\ttop: \"res5c\"\n\tname: \"res5c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c\"\n\ttop: \"pool5\"\n\tname: \"pool5\"\n\ttype: \"Pooling\"\n\tpooling_param {\n\t\tkernel_size: 7\n\t\tstride: 1\n\t\tpool: AVE\n\t}\n}\n\nlayer {\n\tbottom: \"pool5\"\n\ttop: \"fc1000\"\n\tname: \"fc1000\"\n\ttype: \"InnerProduct\"\n\tinner_product_param {\n\t\tnum_output: 1000\n\t}\n}\n\nlayer {\n\tbottom: \"fc1000\"\n\ttop: \"prob\"\n\tname: \"prob\"\n\ttype: \"Softmax\"\n}\n\n" }, { "path": "presets/resnet-50.prototxt", "content": "name: \"ResNet-50\"\ninput: \"data\"\ninput_dim: 1\ninput_dim: 3\ninput_dim: 224\ninput_dim: 224\n\nlayer {\n\tbottom: \"data\"\n\ttop: \"conv1\"\n\tname: \"conv1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 7\n\t\tpad: 3\n\t\tstride: 2\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"bn_conv1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"scale_conv1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"conv1\"\n\tname: \"conv1_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"conv1\"\n\ttop: \"pool1\"\n\tname: \"pool1\"\n\ttype: \"Pooling\"\n\tpooling_param {\n\t\tkernel_size: 3\n\t\tstride: 2\n\t\tpool: MAX\n\t}\n}\n\nlayer {\n\tbottom: \"pool1\"\n\ttop: \"res2a_branch1\"\n\tname: \"res2a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\ttop: \"res2a_branch1\"\n\tname: \"bn2a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\ttop: \"res2a_branch1\"\n\tname: \"scale2a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"pool1\"\n\ttop: \"res2a_branch2a\"\n\tname: \"res2a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"bn2a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"scale2a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2a\"\n\tname: \"res2a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a_branch2a\"\n\ttop: \"res2a_branch2b\"\n\tname: \"res2a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"bn2a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"scale2a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2b\"\n\tname: \"res2a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a_branch2b\"\n\ttop: \"res2a_branch2c\"\n\tname: \"res2a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a_branch2c\"\n\tname: \"bn2a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a_branch2c\"\n\tname: \"scale2a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a_branch1\"\n\tbottom: \"res2a_branch2c\"\n\ttop: \"res2a\"\n\tname: \"res2a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2a\"\n\ttop: \"res2a\"\n\tname: \"res2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"res2b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"bn2b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"scale2b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2a\"\n\tname: \"res2b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b_branch2a\"\n\ttop: \"res2b_branch2b\"\n\tname: \"res2b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"bn2b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"scale2b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2b\"\n\tname: \"res2b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b_branch2b\"\n\ttop: \"res2b_branch2c\"\n\tname: \"res2b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b_branch2c\"\n\tname: \"bn2b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b_branch2c\"\n\tname: \"scale2b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2a\"\n\tbottom: \"res2b_branch2c\"\n\ttop: \"res2b\"\n\tname: \"res2b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2b\"\n\ttop: \"res2b\"\n\tname: \"res2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2b\"\n\ttop: \"res2c_branch2a\"\n\tname: \"res2c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"bn2c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"scale2c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2a\"\n\tname: \"res2c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c_branch2a\"\n\ttop: \"res2c_branch2b\"\n\tname: \"res2c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 64\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"bn2c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"scale2c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2b\"\n\tname: \"res2c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c_branch2b\"\n\ttop: \"res2c_branch2c\"\n\tname: \"res2c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c_branch2c\"\n\tname: \"bn2c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c_branch2c\"\n\tname: \"scale2c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2b\"\n\tbottom: \"res2c_branch2c\"\n\ttop: \"res2c\"\n\tname: \"res2c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res2c\"\n\tname: \"res2c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res3a_branch1\"\n\tname: \"res3a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\ttop: \"res3a_branch1\"\n\tname: \"bn3a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\ttop: \"res3a_branch1\"\n\tname: \"scale3a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res2c\"\n\ttop: \"res3a_branch2a\"\n\tname: \"res3a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"bn3a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"scale3a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2a\"\n\tname: \"res3a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a_branch2a\"\n\ttop: \"res3a_branch2b\"\n\tname: \"res3a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"bn3a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"scale3a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2b\"\n\tname: \"res3a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a_branch2b\"\n\ttop: \"res3a_branch2c\"\n\tname: \"res3a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a_branch2c\"\n\tname: \"bn3a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a_branch2c\"\n\tname: \"scale3a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a_branch1\"\n\tbottom: \"res3a_branch2c\"\n\ttop: \"res3a\"\n\tname: \"res3a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3a\"\n\ttop: \"res3a\"\n\tname: \"res3a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"res3b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"bn3b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"scale3b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2a\"\n\tname: \"res3b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b_branch2a\"\n\ttop: \"res3b_branch2b\"\n\tname: \"res3b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2b\"\n\tname: \"bn3b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2b\"\n\tname: \"scale3b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2b\"\n\tname: \"res3b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b_branch2b\"\n\ttop: \"res3b_branch2c\"\n\tname: \"res3b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2c\"\n\ttop: \"res3b_branch2c\"\n\tname: \"bn3b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b_branch2c\"\n\ttop: \"res3b_branch2c\"\n\tname: \"scale3b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3a\"\n\tbottom: \"res3b_branch2c\"\n\ttop: \"res3b\"\n\tname: \"res3b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3b\"\n\ttop: \"res3b\"\n\tname: \"res3b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3b\"\n\ttop: \"res3c_branch2a\"\n\tname: \"res3c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2a\"\n\tname: \"bn3c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2a\"\n\tname: \"scale3c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2a\"\n\tname: \"res3c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3c_branch2a\"\n\ttop: \"res3c_branch2b\"\n\tname: \"res3c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2b\"\n\tname: \"bn3c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2b\"\n\tname: \"scale3c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2b\"\n\tname: \"res3c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3c_branch2b\"\n\ttop: \"res3c_branch2c\"\n\tname: \"res3c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2c\"\n\ttop: \"res3c_branch2c\"\n\tname: \"bn3c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c_branch2c\"\n\ttop: \"res3c_branch2c\"\n\tname: \"scale3c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3b\"\n\tbottom: \"res3c_branch2c\"\n\ttop: \"res3c\"\n\tname: \"res3c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3c\"\n\ttop: \"res3c\"\n\tname: \"res3c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3c\"\n\ttop: \"res3d_branch2a\"\n\tname: \"res3d_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2a\"\n\tname: \"bn3d_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2a\"\n\tname: \"scale3d_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2a\"\n\tname: \"res3d_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3d_branch2a\"\n\ttop: \"res3d_branch2b\"\n\tname: \"res3d_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 128\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2b\"\n\tname: \"bn3d_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2b\"\n\tname: \"scale3d_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2b\"\n\tname: \"res3d_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3d_branch2b\"\n\ttop: \"res3d_branch2c\"\n\tname: \"res3d_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2c\"\n\ttop: \"res3d_branch2c\"\n\tname: \"bn3d_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d_branch2c\"\n\ttop: \"res3d_branch2c\"\n\tname: \"scale3d_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3c\"\n\tbottom: \"res3d_branch2c\"\n\ttop: \"res3d\"\n\tname: \"res3d\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res3d\"\n\ttop: \"res3d\"\n\tname: \"res3d_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res3d\"\n\ttop: \"res4a_branch1\"\n\tname: \"res4a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\ttop: \"res4a_branch1\"\n\tname: \"bn4a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\ttop: \"res4a_branch1\"\n\tname: \"scale4a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res3d\"\n\ttop: \"res4a_branch2a\"\n\tname: \"res4a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"bn4a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"scale4a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2a\"\n\tname: \"res4a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a_branch2a\"\n\ttop: \"res4a_branch2b\"\n\tname: \"res4a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"bn4a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"scale4a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2b\"\n\tname: \"res4a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a_branch2b\"\n\ttop: \"res4a_branch2c\"\n\tname: \"res4a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a_branch2c\"\n\tname: \"bn4a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a_branch2c\"\n\tname: \"scale4a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a_branch1\"\n\tbottom: \"res4a_branch2c\"\n\ttop: \"res4a\"\n\tname: \"res4a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4a\"\n\ttop: \"res4a\"\n\tname: \"res4a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"res4b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"bn4b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"scale4b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2a\"\n\tname: \"res4b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b_branch2a\"\n\ttop: \"res4b_branch2b\"\n\tname: \"res4b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2b\"\n\tname: \"bn4b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2b\"\n\tname: \"scale4b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2b\"\n\tname: \"res4b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b_branch2b\"\n\ttop: \"res4b_branch2c\"\n\tname: \"res4b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2c\"\n\ttop: \"res4b_branch2c\"\n\tname: \"bn4b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b_branch2c\"\n\ttop: \"res4b_branch2c\"\n\tname: \"scale4b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4a\"\n\tbottom: \"res4b_branch2c\"\n\ttop: \"res4b\"\n\tname: \"res4b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4b\"\n\ttop: \"res4b\"\n\tname: \"res4b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4b\"\n\ttop: \"res4c_branch2a\"\n\tname: \"res4c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2a\"\n\tname: \"bn4c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2a\"\n\tname: \"scale4c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2a\"\n\tname: \"res4c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4c_branch2a\"\n\ttop: \"res4c_branch2b\"\n\tname: \"res4c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2b\"\n\tname: \"bn4c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2b\"\n\tname: \"scale4c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2b\"\n\tname: \"res4c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4c_branch2b\"\n\ttop: \"res4c_branch2c\"\n\tname: \"res4c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2c\"\n\ttop: \"res4c_branch2c\"\n\tname: \"bn4c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c_branch2c\"\n\ttop: \"res4c_branch2c\"\n\tname: \"scale4c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4b\"\n\tbottom: \"res4c_branch2c\"\n\ttop: \"res4c\"\n\tname: \"res4c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4c\"\n\ttop: \"res4c\"\n\tname: \"res4c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4c\"\n\ttop: \"res4d_branch2a\"\n\tname: \"res4d_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2a\"\n\tname: \"bn4d_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2a\"\n\tname: \"scale4d_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2a\"\n\tname: \"res4d_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4d_branch2a\"\n\ttop: \"res4d_branch2b\"\n\tname: \"res4d_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2b\"\n\tname: \"bn4d_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2b\"\n\tname: \"scale4d_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2b\"\n\tname: \"res4d_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4d_branch2b\"\n\ttop: \"res4d_branch2c\"\n\tname: \"res4d_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2c\"\n\ttop: \"res4d_branch2c\"\n\tname: \"bn4d_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d_branch2c\"\n\ttop: \"res4d_branch2c\"\n\tname: \"scale4d_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4c\"\n\tbottom: \"res4d_branch2c\"\n\ttop: \"res4d\"\n\tname: \"res4d\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4d\"\n\ttop: \"res4d\"\n\tname: \"res4d_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4d\"\n\ttop: \"res4e_branch2a\"\n\tname: \"res4e_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2a\"\n\tname: \"bn4e_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2a\"\n\tname: \"scale4e_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2a\"\n\tname: \"res4e_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4e_branch2a\"\n\ttop: \"res4e_branch2b\"\n\tname: \"res4e_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2b\"\n\tname: \"bn4e_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2b\"\n\tname: \"scale4e_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2b\"\n\tname: \"res4e_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4e_branch2b\"\n\ttop: \"res4e_branch2c\"\n\tname: \"res4e_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2c\"\n\ttop: \"res4e_branch2c\"\n\tname: \"bn4e_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e_branch2c\"\n\ttop: \"res4e_branch2c\"\n\tname: \"scale4e_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4d\"\n\tbottom: \"res4e_branch2c\"\n\ttop: \"res4e\"\n\tname: \"res4e\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4e\"\n\ttop: \"res4e\"\n\tname: \"res4e_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4e\"\n\ttop: \"res4f_branch2a\"\n\tname: \"res4f_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2a\"\n\tname: \"bn4f_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2a\"\n\tname: \"scale4f_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2a\"\n\tname: \"res4f_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4f_branch2a\"\n\ttop: \"res4f_branch2b\"\n\tname: \"res4f_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 256\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2b\"\n\tname: \"bn4f_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2b\"\n\tname: \"scale4f_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2b\"\n\tname: \"res4f_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4f_branch2b\"\n\ttop: \"res4f_branch2c\"\n\tname: \"res4f_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 1024\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2c\"\n\ttop: \"res4f_branch2c\"\n\tname: \"bn4f_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f_branch2c\"\n\ttop: \"res4f_branch2c\"\n\tname: \"scale4f_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4e\"\n\tbottom: \"res4f_branch2c\"\n\ttop: \"res4f\"\n\tname: \"res4f\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res4f\"\n\ttop: \"res4f\"\n\tname: \"res4f_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res4f\"\n\ttop: \"res5a_branch1\"\n\tname: \"res5a_branch1\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\ttop: \"res5a_branch1\"\n\tname: \"bn5a_branch1\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\ttop: \"res5a_branch1\"\n\tname: \"scale5a_branch1\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res4f\"\n\ttop: \"res5a_branch2a\"\n\tname: \"res5a_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 2\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"bn5a_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"scale5a_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2a\"\n\tname: \"res5a_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a_branch2a\"\n\ttop: \"res5a_branch2b\"\n\tname: \"res5a_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"bn5a_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"scale5a_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2b\"\n\tname: \"res5a_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a_branch2b\"\n\ttop: \"res5a_branch2c\"\n\tname: \"res5a_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a_branch2c\"\n\tname: \"bn5a_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a_branch2c\"\n\tname: \"scale5a_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a_branch1\"\n\tbottom: \"res5a_branch2c\"\n\ttop: \"res5a\"\n\tname: \"res5a\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5a\"\n\ttop: \"res5a\"\n\tname: \"res5a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"res5b_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"bn5b_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"scale5b_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2a\"\n\tname: \"res5b_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b_branch2a\"\n\ttop: \"res5b_branch2b\"\n\tname: \"res5b_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"bn5b_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"scale5b_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2b\"\n\tname: \"res5b_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b_branch2b\"\n\ttop: \"res5b_branch2c\"\n\tname: \"res5b_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b_branch2c\"\n\tname: \"bn5b_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b_branch2c\"\n\tname: \"scale5b_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5a\"\n\tbottom: \"res5b_branch2c\"\n\ttop: \"res5b\"\n\tname: \"res5b\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5b\"\n\ttop: \"res5b\"\n\tname: \"res5b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5b\"\n\ttop: \"res5c_branch2a\"\n\tname: \"res5c_branch2a\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"bn5c_branch2a\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"scale5c_branch2a\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2a\"\n\tname: \"res5c_branch2a_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c_branch2a\"\n\ttop: \"res5c_branch2b\"\n\tname: \"res5c_branch2b\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 512\n\t\tkernel_size: 3\n\t\tpad: 1\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"bn5c_branch2b\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"scale5c_branch2b\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2b\"\n\tname: \"res5c_branch2b_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c_branch2b\"\n\ttop: \"res5c_branch2c\"\n\tname: \"res5c_branch2c\"\n\ttype: \"Convolution\"\n\tconvolution_param {\n\t\tnum_output: 2048\n\t\tkernel_size: 1\n\t\tpad: 0\n\t\tstride: 1\n\t\tbias_term: false\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c_branch2c\"\n\tname: \"bn5c_branch2c\"\n\ttype: \"BatchNorm\"\n\tbatch_norm_param {\n\t\tuse_global_stats: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c_branch2c\"\n\tname: \"scale5c_branch2c\"\n\ttype: \"Scale\"\n\tscale_param {\n\t\tbias_term: true\n\t}\n}\n\nlayer {\n\tbottom: \"res5b\"\n\tbottom: \"res5c_branch2c\"\n\ttop: \"res5c\"\n\tname: \"res5c\"\n\ttype: \"Eltwise\"\n}\n\nlayer {\n\tbottom: \"res5c\"\n\ttop: \"res5c\"\n\tname: \"res5c_relu\"\n\ttype: \"ReLU\"\n}\n\nlayer {\n\tbottom: \"res5c\"\n\ttop: \"pool5\"\n\tname: \"pool5\"\n\ttype: \"Pooling\"\n\tpooling_param {\n\t\tkernel_size: 7\n\t\tstride: 1\n\t\tpool: AVE\n\t}\n}\n\nlayer {\n\tbottom: \"pool5\"\n\ttop: \"fc1000\"\n\tname: \"fc1000\"\n\ttype: \"InnerProduct\"\n\tinner_product_param {\n\t\tnum_output: 1000\n\t}\n}\n\nlayer {\n\tbottom: \"fc1000\"\n\ttop: \"prob\"\n\tname: \"prob\"\n\ttype: \"Softmax\"\n}" }, { "path": "presets/sq11b2a_e3.prototxt", "content": "name: \"SqueezeNet v1.1 b2a ext3\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n input_param {\n shape: {\n dim: 1\n dim: 3\n dim: 256\n dim: 256\n }\n }\n}\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"relu_conv1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n}\nlayer {\n name: \"fire2/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"conv1\"\n top: \"fire2/squeeze3x3\"\n convolution_param {\n num_output: 16\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/squeeze3x3\"\n}\nlayer {\n name: \"fire2/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire2/expand1x1\"\n top: \"fire2/expand1x1\"\n}\nlayer {\n name: \"fire2/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/expand3x3\"\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/expand3x3\"\n}\nlayer {\n name: \"fire2/concat\"\n type: \"Concat\"\n bottom: \"fire2/expand1x1\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/concat\"\n}\nlayer {\n name: \"fire3/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire2/concat\"\n top: \"fire3/squeeze1x1\"\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/squeeze1x1\"\n}\nlayer {\n name: \"fire3/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire3/expand1x1\"\n top: \"fire3/expand1x1\"\n}\nlayer {\n name: \"fire3/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand3x3\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/expand3x3\"\n}\nlayer {\n name: \"fire3/concat\"\n type: \"Concat\"\n bottom: \"fire3/expand1x1\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/concat\"\n}\nlayer {\n name: \"fire4/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire3/concat\"\n top: \"fire4/squeeze3x3\"\n convolution_param {\n num_output: 32\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/squeeze3x3\"\n}\nlayer {\n name: \"fire4/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire4/expand1x1\"\n top: \"fire4/expand1x1\"\n}\nlayer {\n name: \"fire4/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/expand3x3\"\n convolution_param {\n num_output: 128\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/expand3x3\"\n}\nlayer {\n name: \"fire4/concat\"\n type: \"Concat\"\n bottom: \"fire4/expand1x1\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/concat\"\n}\nlayer {\n name: \"fire5/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire4/concat\"\n top: \"fire5/squeeze1x1\"\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/squeeze1x1\"\n}\nlayer {\n name: \"fire5/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire5/expand1x1\"\n top: \"fire5/expand1x1\"\n}\nlayer {\n name: \"fire5/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand3x3\"\n convolution_param {\n num_output: 128\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/expand3x3\"\n}\nlayer {\n name: \"fire5/concat\"\n type: \"Concat\"\n bottom: \"fire5/expand1x1\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/concat\"\n}\nlayer {\n name: \"fire6/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire5/concat\"\n top: \"fire6/squeeze3x3\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/squeeze3x3\"\n}\nlayer {\n name: \"fire6/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire6/expand1x1\"\n top: \"fire6/expand1x1\"\n}\nlayer {\n name: \"fire6/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/expand3x3\"\n convolution_param {\n num_output: 256\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/expand3x3\"\n}\nlayer {\n name: \"fire6/concat\"\n type: \"Concat\"\n bottom: \"fire6/expand1x1\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/concat\"\n}\nlayer {\n name: \"fire7/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire6/concat\"\n top: \"fire7/squeeze1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/squeeze1x1\"\n}\nlayer {\n name: \"fire7/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire7/expand1x1\"\n top: \"fire7/expand1x1\"\n}\nlayer {\n name: \"fire7/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand3x3\"\n convolution_param {\n num_output: 256\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/expand3x3\"\n}\nlayer {\n name: \"fire7/concat\"\n type: \"Concat\"\n bottom: \"fire7/expand1x1\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/concat\"\n}\nlayer {\n name: \"fire8/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire7/concat\"\n top: \"fire8/squeeze3x3\"\n convolution_param {\n num_output: 128\n kernel_size: 3\n pad: 1\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/squeeze3x3\"\n}\nlayer {\n name: \"fire8/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/expand1x1\"\n convolution_param {\n num_output: 512\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire8/expand1x1\"\n top: \"fire8/expand1x1\"\n}\nlayer {\n name: \"fire8/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/expand3x3\"\n convolution_param {\n num_output: 512\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/expand3x3\"\n}\nlayer {\n name: \"fire8/concat\"\n type: \"Concat\"\n bottom: \"fire8/expand1x1\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/concat\"\n}\nlayer {\n name: \"fire9/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire8/concat\"\n top: \"fire9/squeeze1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/squeeze1x1\"\n}\nlayer {\n name: \"fire9/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand1x1\"\n convolution_param {\n num_output: 512\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire9/expand1x1\"\n top: \"fire9/expand1x1\"\n}\nlayer {\n name: \"fire9/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand3x3\"\n convolution_param {\n num_output: 512\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/expand3x3\"\n}\nlayer {\n name: \"fire9/concat\"\n type: \"Concat\"\n bottom: \"fire9/expand1x1\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/concat\"\n}\nlayer {\n name: \"drop9\"\n type: \"Dropout\"\n bottom: \"fire9/concat\"\n top: \"fire9/concat\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"conv10\"\n type: \"Convolution\"\n bottom: \"fire9/concat\"\n top: \"conv10\"\n convolution_param {\n num_output: 1000\n kernel_size: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0.0\n std: 0.01\n }\n }\n}\nlayer {\n name: \"conv10/relu\"\n type: \"ReLU\"\n bottom: \"conv10\"\n top: \"conv10\"\n}\nlayer {\n name: \"pool10\"\n type: \"Pooling\"\n bottom: \"conv10\"\n top: \"pool10\"\n pooling_param {\n pool: AVE\n global_pooling: true\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"pool10\"\n top: \"loss\"\n include {\n phase: TRAIN\n }\n}" }, { "path": "presets/squeezenet.prototxt", "content": "name: \"SqueezeNet\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n input_param {\n shape: {\n dim: 1\n dim: 3\n dim: 227\n dim: 227\n }\n }\n}\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n convolution_param {\n num_output: 96\n kernel_size: 7\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"relu_conv1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv1\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fire2/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"fire2/squeeze1x1\"\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire2/squeeze1x1\"\n top: \"fire2/squeeze1x1\"\n}\nlayer {\n name: \"fire2/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze1x1\"\n top: \"fire2/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire2/expand1x1\"\n top: \"fire2/expand1x1\"\n}\nlayer {\n name: \"fire2/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze1x1\"\n top: \"fire2/expand3x3\"\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/expand3x3\"\n}\nlayer {\n name: \"fire2/concat\"\n type: \"Concat\"\n bottom: \"fire2/expand1x1\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/concat\"\n}\nlayer {\n name: \"fire3/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire2/concat\"\n top: \"fire3/squeeze1x1\"\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/squeeze1x1\"\n}\nlayer {\n name: \"fire3/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire3/expand1x1\"\n top: \"fire3/expand1x1\"\n}\nlayer {\n name: \"fire3/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand3x3\"\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/expand3x3\"\n}\nlayer {\n name: \"fire3/concat\"\n type: \"Concat\"\n bottom: \"fire3/expand1x1\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/concat\"\n}\nlayer {\n name: \"fire4/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire3/concat\"\n top: \"fire4/squeeze1x1\"\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire4/squeeze1x1\"\n top: \"fire4/squeeze1x1\"\n}\nlayer {\n name: \"fire4/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze1x1\"\n top: \"fire4/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire4/expand1x1\"\n top: \"fire4/expand1x1\"\n}\nlayer {\n name: \"fire4/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze1x1\"\n top: \"fire4/expand3x3\"\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/expand3x3\"\n}\nlayer {\n name: \"fire4/concat\"\n type: \"Concat\"\n bottom: \"fire4/expand1x1\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/concat\"\n}\nlayer {\n name: \"pool4\"\n type: \"Pooling\"\n bottom: \"fire4/concat\"\n top: \"pool4\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fire5/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"pool4\"\n top: \"fire5/squeeze1x1\"\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/squeeze1x1\"\n}\nlayer {\n name: \"fire5/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire5/expand1x1\"\n top: \"fire5/expand1x1\"\n}\nlayer {\n name: \"fire5/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand3x3\"\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/expand3x3\"\n}\nlayer {\n name: \"fire5/concat\"\n type: \"Concat\"\n bottom: \"fire5/expand1x1\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/concat\"\n}\nlayer {\n name: \"fire6/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire5/concat\"\n top: \"fire6/squeeze1x1\"\n convolution_param {\n num_output: 48\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire6/squeeze1x1\"\n top: \"fire6/squeeze1x1\"\n}\nlayer {\n name: \"fire6/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze1x1\"\n top: \"fire6/expand1x1\"\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire6/expand1x1\"\n top: \"fire6/expand1x1\"\n}\nlayer {\n name: \"fire6/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze1x1\"\n top: \"fire6/expand3x3\"\n convolution_param {\n num_output: 192\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/expand3x3\"\n}\nlayer {\n name: \"fire6/concat\"\n type: \"Concat\"\n bottom: \"fire6/expand1x1\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/concat\"\n}\nlayer {\n name: \"fire7/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire6/concat\"\n top: \"fire7/squeeze1x1\"\n convolution_param {\n num_output: 48\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/squeeze1x1\"\n}\nlayer {\n name: \"fire7/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand1x1\"\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire7/expand1x1\"\n top: \"fire7/expand1x1\"\n}\nlayer {\n name: \"fire7/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand3x3\"\n convolution_param {\n num_output: 192\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/expand3x3\"\n}\nlayer {\n name: \"fire7/concat\"\n type: \"Concat\"\n bottom: \"fire7/expand1x1\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/concat\"\n}\nlayer {\n name: \"fire8/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire7/concat\"\n top: \"fire8/squeeze1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire8/squeeze1x1\"\n top: \"fire8/squeeze1x1\"\n}\nlayer {\n name: \"fire8/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze1x1\"\n top: \"fire8/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire8/expand1x1\"\n top: \"fire8/expand1x1\"\n}\nlayer {\n name: \"fire8/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze1x1\"\n top: \"fire8/expand3x3\"\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/expand3x3\"\n}\nlayer {\n name: \"fire8/concat\"\n type: \"Concat\"\n bottom: \"fire8/expand1x1\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/concat\"\n}\nlayer {\n name: \"pool8\"\n type: \"Pooling\"\n bottom: \"fire8/concat\"\n top: \"pool8\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fire9/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"pool8\"\n top: \"fire9/squeeze1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/squeeze1x1\"\n}\nlayer {\n name: \"fire9/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire9/expand1x1\"\n top: \"fire9/expand1x1\"\n}\nlayer {\n name: \"fire9/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand3x3\"\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/expand3x3\"\n}\nlayer {\n name: \"fire9/concat\"\n type: \"Concat\"\n bottom: \"fire9/expand1x1\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/concat\"\n}\nlayer {\n name: \"drop9\"\n type: \"Dropout\"\n bottom: \"fire9/concat\"\n top: \"fire9/concat\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"conv10\"\n type: \"Convolution\"\n bottom: \"fire9/concat\"\n top: \"conv10\"\n convolution_param {\n num_output: 1000\n pad: 1\n kernel_size: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0.0\n std: 0.01\n }\n }\n}\nlayer {\n name: \"relu_conv10\"\n type: \"ReLU\"\n bottom: \"conv10\"\n top: \"conv10\"\n}\nlayer {\n name: \"pool10\"\n type: \"Pooling\"\n bottom: \"conv10\"\n top: \"pool10\"\n pooling_param {\n pool: AVE\n global_pooling: true\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"pool10\"\n top: \"loss\"\n include {\n phase: TRAIN\n }\n}" }, { "path": "presets/squeezenet_v11.prototxt", "content": "name: \"SqueezeNet v1.1\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n input_param {\n shape: {\n dim: 1\n dim: 3\n dim: 227\n dim: 227\n }\n }\n}\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"relu_conv1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n}\nlayer {\n name: \"pool1\"\n type: \"Pooling\"\n bottom: \"conv1\"\n top: \"pool1\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fire2/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"pool1\"\n top: \"fire2/squeeze1x1\"\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire2/squeeze1x1\"\n top: \"fire2/squeeze1x1\"\n}\nlayer {\n name: \"fire2/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze1x1\"\n top: \"fire2/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire2/expand1x1\"\n top: \"fire2/expand1x1\"\n}\nlayer {\n name: \"fire2/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze1x1\"\n top: \"fire2/expand3x3\"\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/expand3x3\"\n}\nlayer {\n name: \"fire2/concat\"\n type: \"Concat\"\n bottom: \"fire2/expand1x1\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/concat\"\n}\nlayer {\n name: \"fire3/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire2/concat\"\n top: \"fire3/squeeze1x1\"\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/squeeze1x1\"\n}\nlayer {\n name: \"fire3/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire3/expand1x1\"\n top: \"fire3/expand1x1\"\n}\nlayer {\n name: \"fire3/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand3x3\"\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/expand3x3\"\n}\nlayer {\n name: \"fire3/concat\"\n type: \"Concat\"\n bottom: \"fire3/expand1x1\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/concat\"\n}\nlayer {\n name: \"pool3\"\n type: \"Pooling\"\n bottom: \"fire3/concat\"\n top: \"pool3\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fire4/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"pool3\"\n top: \"fire4/squeeze1x1\"\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire4/squeeze1x1\"\n top: \"fire4/squeeze1x1\"\n}\nlayer {\n name: \"fire4/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze1x1\"\n top: \"fire4/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire4/expand1x1\"\n top: \"fire4/expand1x1\"\n}\nlayer {\n name: \"fire4/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze1x1\"\n top: \"fire4/expand3x3\"\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/expand3x3\"\n}\nlayer {\n name: \"fire4/concat\"\n type: \"Concat\"\n bottom: \"fire4/expand1x1\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/concat\"\n}\nlayer {\n name: \"fire5/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire4/concat\"\n top: \"fire5/squeeze1x1\"\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/squeeze1x1\"\n}\nlayer {\n name: \"fire5/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire5/expand1x1\"\n top: \"fire5/expand1x1\"\n}\nlayer {\n name: \"fire5/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand3x3\"\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/expand3x3\"\n}\nlayer {\n name: \"fire5/concat\"\n type: \"Concat\"\n bottom: \"fire5/expand1x1\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/concat\"\n}\nlayer {\n name: \"pool5\"\n type: \"Pooling\"\n bottom: \"fire5/concat\"\n top: \"pool5\"\n pooling_param {\n pool: MAX\n kernel_size: 3\n stride: 2\n }\n}\nlayer {\n name: \"fire6/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"pool5\"\n top: \"fire6/squeeze1x1\"\n convolution_param {\n num_output: 48\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire6/squeeze1x1\"\n top: \"fire6/squeeze1x1\"\n}\nlayer {\n name: \"fire6/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze1x1\"\n top: \"fire6/expand1x1\"\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire6/expand1x1\"\n top: \"fire6/expand1x1\"\n}\nlayer {\n name: \"fire6/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze1x1\"\n top: \"fire6/expand3x3\"\n convolution_param {\n num_output: 192\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/expand3x3\"\n}\nlayer {\n name: \"fire6/concat\"\n type: \"Concat\"\n bottom: \"fire6/expand1x1\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/concat\"\n}\nlayer {\n name: \"fire7/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire6/concat\"\n top: \"fire7/squeeze1x1\"\n convolution_param {\n num_output: 48\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/squeeze1x1\"\n}\nlayer {\n name: \"fire7/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand1x1\"\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire7/expand1x1\"\n top: \"fire7/expand1x1\"\n}\nlayer {\n name: \"fire7/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand3x3\"\n convolution_param {\n num_output: 192\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/expand3x3\"\n}\nlayer {\n name: \"fire7/concat\"\n type: \"Concat\"\n bottom: \"fire7/expand1x1\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/concat\"\n}\nlayer {\n name: \"fire8/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire7/concat\"\n top: \"fire8/squeeze1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire8/squeeze1x1\"\n top: \"fire8/squeeze1x1\"\n}\nlayer {\n name: \"fire8/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze1x1\"\n top: \"fire8/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire8/expand1x1\"\n top: \"fire8/expand1x1\"\n}\nlayer {\n name: \"fire8/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze1x1\"\n top: \"fire8/expand3x3\"\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/expand3x3\"\n}\nlayer {\n name: \"fire8/concat\"\n type: \"Concat\"\n bottom: \"fire8/expand1x1\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/concat\"\n}\nlayer {\n name: \"fire9/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire8/concat\"\n top: \"fire9/squeeze1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/squeeze1x1\"\n}\nlayer {\n name: \"fire9/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire9/expand1x1\"\n top: \"fire9/expand1x1\"\n}\nlayer {\n name: \"fire9/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand3x3\"\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/expand3x3\"\n}\nlayer {\n name: \"fire9/concat\"\n type: \"Concat\"\n bottom: \"fire9/expand1x1\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/concat\"\n}\nlayer {\n name: \"drop9\"\n type: \"Dropout\"\n bottom: \"fire9/concat\"\n top: \"fire9/concat\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayer {\n name: \"conv10\"\n type: \"Convolution\"\n bottom: \"fire9/concat\"\n top: \"conv10\"\n convolution_param {\n num_output: 1000\n kernel_size: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0.0\n std: 0.01\n }\n }\n}\nlayer {\n name: \"relu_conv10\"\n type: \"ReLU\"\n bottom: \"conv10\"\n top: \"conv10\"\n}\nlayer {\n name: \"pool10\"\n type: \"Pooling\"\n bottom: \"conv10\"\n top: \"pool10\"\n pooling_param {\n pool: AVE\n global_pooling: true\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"pool10\"\n top: \"loss\"\n include {\n phase: TRAIN\n }\n}" }, { "path": "presets/vgg-16.prototxt", "content": "name: \"VGG_ILSVRC_16_layers\"\nlayer {\n type: \"data\"\n name: \"data\"\n top: \"data\"\n input_param: {\n shape: {\n dim: 10\n dim: 3\n dim: 224\n dim: 224\n }\n }\n\n}\nlayers {\n bottom: \"data\"\n top: \"conv1_1\"\n name: \"conv1_1\"\n type: CONVOLUTION\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv1_1\"\n top: \"conv1_1\"\n name: \"relu1_1\"\n type: RELU\n}\nlayers {\n bottom: \"conv1_1\"\n top: \"conv1_2\"\n name: \"conv1_2\"\n type: CONVOLUTION\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv1_2\"\n top: \"conv1_2\"\n name: \"relu1_2\"\n type: RELU\n}\nlayers {\n bottom: \"conv1_2\"\n top: \"pool1\"\n name: \"pool1\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayers {\n bottom: \"pool1\"\n top: \"conv2_1\"\n name: \"conv2_1\"\n type: CONVOLUTION\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv2_1\"\n top: \"conv2_1\"\n name: \"relu2_1\"\n type: RELU\n}\nlayers {\n bottom: \"conv2_1\"\n top: \"conv2_2\"\n name: \"conv2_2\"\n type: CONVOLUTION\n convolution_param {\n num_output: 128\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv2_2\"\n top: \"conv2_2\"\n name: \"relu2_2\"\n type: RELU\n}\nlayers {\n bottom: \"conv2_2\"\n top: \"pool2\"\n name: \"pool2\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayers {\n bottom: \"pool2\"\n top: \"conv3_1\"\n name: \"conv3_1\"\n type: CONVOLUTION\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv3_1\"\n top: \"conv3_1\"\n name: \"relu3_1\"\n type: RELU\n}\nlayers {\n bottom: \"conv3_1\"\n top: \"conv3_2\"\n name: \"conv3_2\"\n type: CONVOLUTION\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv3_2\"\n top: \"conv3_2\"\n name: \"relu3_2\"\n type: RELU\n}\nlayers {\n bottom: \"conv3_2\"\n top: \"conv3_3\"\n name: \"conv3_3\"\n type: CONVOLUTION\n convolution_param {\n num_output: 256\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv3_3\"\n top: \"conv3_3\"\n name: \"relu3_3\"\n type: RELU\n}\nlayers {\n bottom: \"conv3_3\"\n top: \"pool3\"\n name: \"pool3\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayers {\n bottom: \"pool3\"\n top: \"conv4_1\"\n name: \"conv4_1\"\n type: CONVOLUTION\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv4_1\"\n top: \"conv4_1\"\n name: \"relu4_1\"\n type: RELU\n}\nlayers {\n bottom: \"conv4_1\"\n top: \"conv4_2\"\n name: \"conv4_2\"\n type: CONVOLUTION\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv4_2\"\n top: \"conv4_2\"\n name: \"relu4_2\"\n type: RELU\n}\nlayers {\n bottom: \"conv4_2\"\n top: \"conv4_3\"\n name: \"conv4_3\"\n type: CONVOLUTION\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv4_3\"\n top: \"conv4_3\"\n name: \"relu4_3\"\n type: RELU\n}\nlayers {\n bottom: \"conv4_3\"\n top: \"pool4\"\n name: \"pool4\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayers {\n bottom: \"pool4\"\n top: \"conv5_1\"\n name: \"conv5_1\"\n type: CONVOLUTION\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv5_1\"\n top: \"conv5_1\"\n name: \"relu5_1\"\n type: RELU\n}\nlayers {\n bottom: \"conv5_1\"\n top: \"conv5_2\"\n name: \"conv5_2\"\n type: CONVOLUTION\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv5_2\"\n top: \"conv5_2\"\n name: \"relu5_2\"\n type: RELU\n}\nlayers {\n bottom: \"conv5_2\"\n top: \"conv5_3\"\n name: \"conv5_3\"\n type: CONVOLUTION\n convolution_param {\n num_output: 512\n pad: 1\n kernel_size: 3\n }\n}\nlayers {\n bottom: \"conv5_3\"\n top: \"conv5_3\"\n name: \"relu5_3\"\n type: RELU\n}\nlayers {\n bottom: \"conv5_3\"\n top: \"pool5\"\n name: \"pool5\"\n type: POOLING\n pooling_param {\n pool: MAX\n kernel_size: 2\n stride: 2\n }\n}\nlayers {\n bottom: \"pool5\"\n top: \"fc6\"\n name: \"fc6\"\n type: INNER_PRODUCT\n inner_product_param {\n num_output: 4096\n }\n}\nlayers {\n bottom: \"fc6\"\n top: \"fc6\"\n name: \"relu6\"\n type: RELU\n}\nlayers {\n bottom: \"fc6\"\n top: \"fc6\"\n name: \"drop6\"\n type: DROPOUT\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayers {\n bottom: \"fc6\"\n top: \"fc7\"\n name: \"fc7\"\n type: INNER_PRODUCT\n inner_product_param {\n num_output: 4096\n }\n}\nlayers {\n bottom: \"fc7\"\n top: \"fc7\"\n name: \"relu7\"\n type: RELU\n}\nlayers {\n bottom: \"fc7\"\n top: \"fc7\"\n name: \"drop7\"\n type: DROPOUT\n dropout_param {\n dropout_ratio: 0.5\n }\n}\nlayers {\n bottom: \"fc7\"\n top: \"fc8\"\n name: \"fc8\"\n type: INNER_PRODUCT\n inner_product_param {\n num_output: 1000\n }\n}\nlayers {\n bottom: \"fc8\"\n top: \"prob\"\n name: \"prob\"\n type: SOFTMAX\n}\n" }, { "path": "presets/zynqnet.prototxt", "content": "name: \"ZynqNet\"\nlayer {\n name: \"data\"\n type: \"Data\"\n top: \"data\"\n input_param {\n shape: {\n dim: 1\n dim: 3\n dim: 256\n dim: 256\n }\n }\n}\nlayer {\n name: \"conv1\"\n type: \"Convolution\"\n bottom: \"data\"\n top: \"conv1\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"relu_conv1\"\n type: \"ReLU\"\n bottom: \"conv1\"\n top: \"conv1\"\n}\nlayer {\n name: \"fire2/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"conv1\"\n top: \"fire2/squeeze3x3\"\n convolution_param {\n num_output: 16\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/squeeze3x3\"\n}\nlayer {\n name: \"fire2/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire2/expand1x1\"\n top: \"fire2/expand1x1\"\n}\nlayer {\n name: \"fire2/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire2/squeeze3x3\"\n top: \"fire2/expand3x3\"\n convolution_param {\n num_output: 64\n pad: 1\n kernel_size: 3\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire2/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/expand3x3\"\n}\nlayer {\n name: \"fire2/concat\"\n type: \"Concat\"\n bottom: \"fire2/expand1x1\"\n bottom: \"fire2/expand3x3\"\n top: \"fire2/concat\"\n}\nlayer {\n name: \"fire3/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire2/concat\"\n top: \"fire3/squeeze1x1\"\n convolution_param {\n num_output: 16\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/squeeze1x1\"\n}\nlayer {\n name: \"fire3/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire3/expand1x1\"\n top: \"fire3/expand1x1\"\n}\nlayer {\n name: \"fire3/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire3/squeeze1x1\"\n top: \"fire3/expand3x3\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire3/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/expand3x3\"\n}\nlayer {\n name: \"fire3/concat\"\n type: \"Concat\"\n bottom: \"fire3/expand1x1\"\n bottom: \"fire3/expand3x3\"\n top: \"fire3/concat\"\n}\nlayer {\n name: \"fire4/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire3/concat\"\n top: \"fire4/squeeze3x3\"\n convolution_param {\n num_output: 32\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/squeeze3x3\"\n}\nlayer {\n name: \"fire4/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire4/expand1x1\"\n top: \"fire4/expand1x1\"\n}\nlayer {\n name: \"fire4/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire4/squeeze3x3\"\n top: \"fire4/expand3x3\"\n convolution_param {\n num_output: 128\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire4/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/expand3x3\"\n}\nlayer {\n name: \"fire4/concat\"\n type: \"Concat\"\n bottom: \"fire4/expand1x1\"\n bottom: \"fire4/expand3x3\"\n top: \"fire4/concat\"\n}\nlayer {\n name: \"fire5/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire4/concat\"\n top: \"fire5/squeeze1x1\"\n convolution_param {\n num_output: 32\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/squeeze1x1\"\n}\nlayer {\n name: \"fire5/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand1x1\"\n convolution_param {\n num_output: 128\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire5/expand1x1\"\n top: \"fire5/expand1x1\"\n}\nlayer {\n name: \"fire5/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire5/squeeze1x1\"\n top: \"fire5/expand3x3\"\n convolution_param {\n num_output: 128\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire5/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/expand3x3\"\n}\nlayer {\n name: \"fire5/concat\"\n type: \"Concat\"\n bottom: \"fire5/expand1x1\"\n bottom: \"fire5/expand3x3\"\n top: \"fire5/concat\"\n}\nlayer {\n name: \"fire6/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire5/concat\"\n top: \"fire6/squeeze3x3\"\n convolution_param {\n num_output: 64\n kernel_size: 3\n stride: 2\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/squeeze3x3\"\n}\nlayer {\n name: \"fire6/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire6/expand1x1\"\n top: \"fire6/expand1x1\"\n}\nlayer {\n name: \"fire6/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire6/squeeze3x3\"\n top: \"fire6/expand3x3\"\n convolution_param {\n num_output: 256\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire6/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/expand3x3\"\n}\nlayer {\n name: \"fire6/concat\"\n type: \"Concat\"\n bottom: \"fire6/expand1x1\"\n bottom: \"fire6/expand3x3\"\n top: \"fire6/concat\"\n}\nlayer {\n name: \"fire7/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire6/concat\"\n top: \"fire7/squeeze1x1\"\n convolution_param {\n num_output: 64\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/squeeze1x1\"\n}\nlayer {\n name: \"fire7/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand1x1\"\n convolution_param {\n num_output: 192\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire7/expand1x1\"\n top: \"fire7/expand1x1\"\n}\nlayer {\n name: \"fire7/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire7/squeeze1x1\"\n top: \"fire7/expand3x3\"\n convolution_param {\n num_output: 192\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire7/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/expand3x3\"\n}\nlayer {\n name: \"fire7/concat\"\n type: \"Concat\"\n bottom: \"fire7/expand1x1\"\n bottom: \"fire7/expand3x3\"\n top: \"fire7/concat\"\n}\nlayer {\n name: \"fire8/squeeze3x3\"\n type: \"Convolution\"\n bottom: \"fire7/concat\"\n top: \"fire8/squeeze3x3\"\n convolution_param {\n num_output: 112\n kernel_size: 3\n pad: 1\n stride: 2\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_squeeze3x3\"\n type: \"ReLU\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/squeeze3x3\"\n}\nlayer {\n name: \"fire8/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/expand1x1\"\n convolution_param {\n num_output: 256\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire8/expand1x1\"\n top: \"fire8/expand1x1\"\n}\nlayer {\n name: \"fire8/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire8/squeeze3x3\"\n top: \"fire8/expand3x3\"\n convolution_param {\n num_output: 256\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire8/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/expand3x3\"\n}\nlayer {\n name: \"fire8/concat\"\n type: \"Concat\"\n bottom: \"fire8/expand1x1\"\n bottom: \"fire8/expand3x3\"\n top: \"fire8/concat\"\n}\nlayer {\n name: \"fire9/squeeze1x1\"\n type: \"Convolution\"\n bottom: \"fire8/concat\"\n top: \"fire9/squeeze1x1\"\n convolution_param {\n num_output: 112\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_squeeze1x1\"\n type: \"ReLU\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/squeeze1x1\"\n}\nlayer {\n name: \"fire9/expand1x1\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand1x1\"\n convolution_param {\n num_output: 368\n kernel_size: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand1x1\"\n type: \"ReLU\"\n bottom: \"fire9/expand1x1\"\n top: \"fire9/expand1x1\"\n}\nlayer {\n name: \"fire9/expand3x3\"\n type: \"Convolution\"\n bottom: \"fire9/squeeze1x1\"\n top: \"fire9/expand3x3\"\n convolution_param {\n num_output: 368\n kernel_size: 3\n pad: 1\n weight_filler {\n type: \"xavier\"\n }\n }\n}\nlayer {\n name: \"fire9/relu_expand3x3\"\n type: \"ReLU\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/expand3x3\"\n}\nlayer {\n name: \"fire9/concat\"\n type: \"Concat\"\n bottom: \"fire9/expand1x1\"\n bottom: \"fire9/expand3x3\"\n top: \"fire9/concat\"\n}\nlayer {\n name: \"drop9\"\n type: \"Dropout\"\n bottom: \"fire9/concat\"\n top: \"fire9/concat\"\n dropout_param {\n dropout_ratio: 0.5\n }\n}\n\nlayer {\n name: \"conv10/split1\"\n type: \"Convolution\"\n bottom: \"fire9/concat\"\n top: \"conv10/split1\"\n convolution_param {\n num_output: 512\n kernel_size: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0.0\n std: 0.01\n }\n }\n}\nlayer {\n name: \"conv10/split2\"\n type: \"Convolution\"\n bottom: \"fire9/concat\"\n top: \"conv10/split2\"\n convolution_param {\n num_output: 512\n kernel_size: 1\n weight_filler {\n type: \"gaussian\"\n mean: 0.0\n std: 0.01\n }\n }\n}\nlayer {\n name: \"conv10\"\n type: \"Concat\"\n bottom: \"conv10/split1\"\n bottom: \"conv10/split2\"\n top: \"conv10\"\n}\n\nlayer {\n name: \"pool10\"\n type: \"Pooling\"\n bottom: \"conv10\"\n top: \"pool10\"\n pooling_param {\n pool: AVE\n global_pooling: true\n }\n}\nlayer {\n name: \"loss\"\n type: \"SoftmaxWithLoss\"\n bottom: \"pool10\"\n top: \"loss\"\n include {\n phase: TRAIN\n }\n}" }, { "path": "quickstart.html", "content": "\n\n\n \n\t\n\t\n\t\n Quick Start — Netscope CNN Analyzer\n \n \n\n\n
\n

Netscope CNN Analyzer

\n

A web-based tool for visualizing and analyzing convolutional neural network architectures (or technically, any directed acyclic graph).\n\t\t Currently supports Caffe's prototxt format.

\n

Basis by ethereon. Extended for CNN Analysis by dgschwend.

\n
\n
\n

Gist Support

\n

If your .prototxt file is part of a GitHub Gist, you can visualize it by visiting this URL:

\n
http://dgschwend.github.io/netscope/#/gist/your-gist-id
\n

The Gist ID is the numeric suffix in the Gist's URL.

\n View Example\n
\n
\n

Editor

\n

You can use the inline editor to enter your network definition (currently limited to valid Caffe's prototext) and visualize the network.

\n

Press Shift+Enter in the editor to render your network.

\n Launch Editor\n
\n
\n

Presets

\n\t
\n ZynqNet CNN\n
David Gschwend (see the master thesis repository)
\n
\n
\n YOLO\n
Joseph Redmon, Ali Farhadi
\n
\n
\n SqueezeNet\n
Forrest Iandola, Matthew Moskewicz, Khalid Ashraf, Song Han, William Dally, Kurt Keutzer
\n
\n
\n SqueezeNet v1.1\n
Forrest Iandola, Matthew Moskewicz, Khalid Ashraf, Song Han, William Dally, Kurt Keutzer
\n
\n
\n Inception v4\n
Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, Zbigniew Wojna
\n\t\t
\n
\n Inception-ResNet-v2\n
Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, Zbigniew Wojna
\n\t\t
\n
\n Inception v3\n
Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, Zbigniew Wojna
\n
\n
\n
\n ResNet-50\n
Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
\n\t\t
\n
\n ResNet-152\n
Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun
\n\t\t
\n
\n AlexNet\n
Alex Krizhevsky, Ilya Sutskever, Geoffry Hinton
\n
\n
\n CaffeNet\n
Yangqing Jia, Evan Shelhamer, et. al.
\n
\n
\n FCN-16s: Fully Convolutional Network — Two Stream\n
Jonathan Long, et. al
\n
\n
\n\n
\n GoogLeNet\n
Christian Szegedy, et. al.
\n
\n
\n Network in Network\n
Min Lin, Qiang Chen, Shuicheng Yan
\n
\n
\n VGG 16 Layers\n
Karen Simonyan, Andrew Zisserman
\n
\n
\n\n\n" }, { "path": "required_nodejs_modules.txt", "content": "This project requires node.js and the following modules to be installed for successful compilation\r\ninstall with npm install [@]\r\n\r\ntablesorter@2.26\r\ncoffeeify\r\nwatchify\r\n" }, { "path": "scripts/watch.sh", "content": "#!/bin/sh\nwatchify -v -t coffeeify src/netscope.coffee -o assets/js/netscope.js\n" }, { "path": "src/analyzer.coffee", "content": "module.exports =\n class Analyzer\n constructor: ->\n\n analyze: (net) ->\n ## Add Input/Output Dimensions + Channels to each Node / Layer\n # shape.dim: ( N x K x W x H )\n # batch channel width height\n # chIn chOut wIn wOut\n\n for n in net.sortTopologically()\n\n layertype = n.type.toLowerCase()\n # Setup Default Values for Analysis\n d = n.analysis\n d.wIn = d.hIn = d.wOut = d.hOut = d.chIn = d.chOut = 0\n d.comp = {macc: 0, comp: 0, add: 0, div: 0, exp: 0}\n d.mem = {activation: 0, param: 0}\n d.variants = [];\n\n # Setup connection to parent layer\n parent = n.parents[0]?.analysis\n\n # Setup default channels + dimensions: inherited from parent\n d.batchOut = d.batchIn = parent?.batchOut\n d.wIn = parent?.wOut\n d.hIn = parent?.hOut\n d.chIn = parent?.chOut\n switch layertype\n when \"data\", \"input\"\n #dimensions\n if n.attribs.input_param?.shape?\n shape = n.attribs.input_param.shape\n d.batchIn = shape.dim[0]\n d.chIn = shape.dim[1]\n d.hIn = shape.dim[2]\n d.wIn = shape.dim[3]\n else if n.attribs.transform_param?.crop_size?\n d.wIn = d.hIn = n.attribs.transform_param.crop_size\n d.chIn = 3 # assume RGB\n d.batchOut = 1\n else\n onerror('Unknown Input Dimensions')\n debugger;\n # update output sizes\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n d.batchOut = d.batchIn\n #computation\n #-- none\n #memory\n #-- none\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"convolution\"\n #dimensions\n params = n.attribs.convolution_param\n kernel_w = params.kernel_w ? params.kernel_size\n kernel_h = params.kernel_h ? params.kernel_size\n stride_w = params.stride_w ? (params.stride ? 1)\n stride_h = params.stride_h ? (params.stride ? 1)\n pad_w = params.pad_w ? (params.pad ? 0)\n pad_h = params.pad_h ? (params.pad ? 0)\n numout = params.num_output\n group = params.group ? 1\n dilation = params.dilation ? 1\n has_bias = if (params.bias_term ? \"true\") == \"false\" then 0 else 1\n\n # according to http://caffe.berkeleyvision.org/tutorial/layers.html and https://github.com/BVLC/caffe/issues/3656\n kernel = dilation*(kernel_w-1)+1\n d.wOut = Math.floor((d.wIn + 2*pad_w - kernel) / stride_w) + 1\n kernel = dilation*(kernel_h-1)+1\n d.hOut = Math.floor((d.hIn + 2*pad_h - kernel) / stride_h) + 1\n\n d.chOut = numout\n #computation\n d.comp.macc = (kernel_w*kernel_h)*(d.wOut*d.hOut)*d.chIn*d.chOut*d.batchOut/group\n #memory\n d.mem.param = (kernel_w*kernel_h)*d.chIn*d.chOut/group + has_bias*d.chOut\n d.mem.activation = (d.wOut*d.hOut)*d.chOut*d.batchOut\n\n # CACHE AND BANDWIDTH for Implementation Variants\n if (do_variants_analysis)\n d.variants.push({\n name : \"complete outputs, input cache\"\n cache : d.chIn*kernel_h*d.wIn + # line buffers\n d.chIn*kernel_h*kernel_w # param cache\n readBW : d.chOut*d.chIn*(d.wIn*d.hIn)\n writeBW : d.chOut*(d.wOut*d.hOut) # ideal\n confBW : d.chOut*d.chIn*kernel_w*kernel_h # ideal\n })\n d.variants.push({\n name : \"complete inputs, input cache\"\n cache : kernel_h*d.wIn + # line buffers\n d.chIn*kernel_h*kernel_w # param cache\n readBW : d.chIn*((d.chOut+1)*(d.wIn*d.hIn))\n writeBW : d.chIn*((d.chOut)*(d.wOut*d.hOut))\n confBW : d.chOut*d.chIn*kernel_w*kernel_h # ideal\n })\n d.variants.push({\n name : \"complete inputs, input + output cache\"\n cache : kernel_h*d.wIn + # line buffers\n d.chIn*kernel_h*kernel_w + # param cache\n d.wIn*d.hIn*d.chOut # output cache\n readBW : d.chIn*(d.wIn*d.hIn) # ideal\n writeBW : d.chOut*(d.wOut*d.hOut) # ideal\n confBW : d.chOut*d.chIn*kernel_w*kernel_h # ideal\n })\n d.variants.push({\n name : \"streaming, input cache\"\n cache : d.chIn*kernel_h*d.wIn\n readBW : d.chIn*(d.wIn*d.hIn)\n writeBW : d.chOut*(d.wOut*d.hOut)\n confBW : d.hIn*(d.chIn*d.chOut*(kernel_w*kernel_h))\n })\n d.variants.push({\n name : \"streaming, input + config cache\"\n cache : d.chIn*kernel_h*d.wIn +\n d.chIn*d.chOut*(kernel_h*kernel_w)\n readBW : d.chIn*(d.wIn*d.hIn)\n writeBW : d.chOut*(d.wOut*d.hOut)\n confBW : d.chOut*d.chIn*kernel_w*kernel_h # ideal\n })\n d.variants.push({\n name : \"streaming, temp.\"\n img_cache : if kernel_h > 1 then d.chIn*kernel_h*d.wIn else d.chIn\n img_dim : d.chIn+\"ch ∙ \"+d.wIn+\" × \"+kernel_h+\" × 32b\"\n flt_cache : d.chIn*d.chOut*(kernel_h*kernel_w)\n squeeze_cache : if n.name.indexOf(\"squeeze\") > -1 then d.chOut*d.wOut*d.hOut else \"\"\n })\n\n\n when \"innerproduct\", \"inner_product\"\n #dimensions\n numout = n.attribs.inner_product_param.num_output\n has_bias = if (n.attribs.inner_product_param.bias_term ? \"true\") == \"false\" then 0 else 1\n d.wOut = 1\n d.hOut = 1\n d.chOut = numout\n #computation\n d.comp.macc = (d.wIn*d.hIn)*d.chIn*d.chOut*d.batchOut\n #memory\n d.mem.param = d.wIn*d.hIn*d.chIn*d.chOut + has_bias*d.chOut\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"pooling\"\n #dimensions\n params = n.attribs.pooling_param\n kernel_w = params.kernel_w ? params.kernel_size\n kernel_h = params.kernel_h ? params.kernel_size\n stride_w = params.stride_w ? (params.stride ? 1)\n stride_h = params.stride_h ? (params.stride ? 1)\n pad_w = params.pad_w ? (params.pad ? 0)\n pad_h = params.pad_h ? (params.pad ? 0)\n isglobal = params.global_pooling ? 0\n pooltype = (params.pool ? 'MAX').toUpperCase()\n d.chOut = d.chIn\n # according to http://caffe.berkeleyvision.org/tutorial/layers.html and https://github.com/BVLC/caffe/issues/3656\n d.wOut = Math.ceil((d.wIn + 2*pad_w - kernel_w) / stride_w) + 1\n d.hOut = Math.ceil((d.hIn + 2*pad_h - kernel_h) / stride_h) + 1\n if isglobal\n d.wOut = d.hOut = 1\n #computation\n num_ops = if isglobal then ((d.wIn*d.hIn)*d.chIn*d.batchOut) else ((d.wOut*d.hOut)*kernel_h*kernel_w*d.chOut*d.batchOut)\n if pooltype == 'MAX'\n d.comp.comp = num_ops\n else if pooltype == 'AVE'\n d.comp.add = num_ops\n #d.comp.div = (d.wOut*d.hOut*d.chOut) #divide by const.\n else\n onerror \"Unknown pooling type #{pooltype}\"\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"batchnorm\", \"bn\"\n #dimensions\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n #computation\n # BN: subtract mean, divide by variance for each channel\n # averages during training: over spatial dims + batch\n d.comp.add = d.wIn*d.hIn*d.chIn*d.batchOut\n d.comp.div = d.wIn*d.hIn*d.chIn*d.batchOut\n #memory\n d.mem.param = d.chIn*2\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"lrn\", \"normalize\"\n #dimensions\n #default mode: ACROSS_CHANNELS\n mode = n.attribs.lrn_param?.norm_region ? 'ACROSS_CHANNELS'\n size = n.attribs.lrn_param?.local_size ? 1\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n #computation\n # Each input value is divided by (1+(α/n)∑xi^2)^β\n num_inputs = d.wIn*d.hIn*d.chIn*d.batchOut\n d.comp.macc = num_inputs*size # (∑xi^2)\n d.comp.add = num_inputs # (1+...)\n d.comp.exp = num_inputs # (...)^β\n d.comp.div = num_inputs*2 # (α/n)*... + divide by sum\n #memory\n d.mem.param = 2 # alpha, beta\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"concat\"\n #dimensions\n d.wOut = d.wIn\n d.hOut = d.hIn\n # sum up channels from inputs\n d.chIn = 0\n d.chIn += p.analysis.chOut for p in n.parents\n d.chOut = d.chIn\n # check input dimensions\n failed = failed || (p.analysis.wOut != d.wIn || p.analysis.hOut != d.hIn) for p in n.parents\n window.onerror('CONCAT: input dimensions dont agree!') if failed\n #computation\n # --none\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n #relu/dropout use some memory, do some comparisons\n when \"relu\", \"relu6\", \"elu\", \"prelu\", \"dropout\"\n #dimensions\n d.wIn = parent.wOut\n d.hIn = parent.hOut\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn = parent.chOut\n #computation\n d.comp.comp = d.wIn*d.hIn*d.chIn*d.batchOut\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"softmax\", \"softmaxwithloss\", \"softmax_loss\"\n #dimensions\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n #computation\n d.comp.exp = d.wIn*d.hIn*d.chIn*d.batchOut\n d.comp.add = d.wIn*d.hIn*d.chIn*d.batchOut\n d.comp.div = d.wIn*d.hIn*d.chIn*d.batchOut\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"flatten\"\n #dimensions\n d.wOut = d.hOut = 1\n d.chOut = d.chIn * d.wIn * d.hIn\n #computation\n # --none\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"eltwise\"\n #dimensions\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n # check input dimensions\n failed = false\n for p in n.parents\n failed = failed or (d.wIn != p.analysis.wOut) or (d.hIn != p.analysis.hOut)\n onerror 'ELTWISE: input dimensions dont agree in '+n.name if failed\n #computation\n op = n.eltwise_param?.operation?.toUpperCase() ? 'SUM'\n if op == 'SUM'\n d.comp.add = d.wIn*d.hIn*d.chIn*d.batchOut\n else if op == 'MAX'\n d.comp.comp = d.wIn*d.hIn*d.chIn*d.batchOut\n else if op == 'PROD'\n d.comp.macc = d.wIn*d.hIn*d.chIn*d.batchOut\n else\n onerror 'ELTWISE: unknown operation '+op\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"deconvolution\"\n #dimensions\n params = n.attribs.convolution_param\n kernel_w = params.kernel_w ? params.kernel_size\n kernel_h = params.kernel_h ? params.kernel_size\n stride_w = params.stride_w ? (params.stride ? 1)\n stride_h = params.stride_h ? (params.stride ? 1)\n pad_w = params.pad_w ? (params.pad ? 0)\n pad_h = params.pad_h ? (params.pad ? 0)\n numout = params.num_output\n d.wOut = (stride_w*(d.wIn-1)+kernel_w-2*pad_w)\n d.hOut = (stride_h*(d.hIn-1)+kernel_h-2*pad_h)\n d.chOut = numout\n #computation\n d.comp.macc = d.chIn*d.chOut*d.wOut*d.hOut*(kernel_w/stride_w)*(kernel_h/stride_h)*d.batchOut\n #memory\n d.mem.param = kernel_w*kernel_h*d.chIn*d.chOut\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n when \"crop\"\n #dimensions\n ## crop to dims of 2nd parent\n parent2 = n.parents[1].analysis\n d.wOut = parent2.wOut\n d.hOut = parent2.hOut\n d.chOut = d.chIn\n #computation\n # --none\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n #scale layer use activation memory and does multiplies\n when \"scale\"\n #dimensions\n ## assume pass-through\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n #computation: scale = multiplication\n d.comp.macc = d.wOut*d.hOut*d.chOut*d.batchOut\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n #implicit layers use activation memory, but no computation\n when \"implicit\"\n #dimensions\n #fix potentially undefined inputs\n d.wIn = d.wIn ? \"?\"\n d.hIn = d.hIn ? \"?\"\n d.chIn = d.chIn ? \"?\"\n d.batchIn = d.batchIn ? \"?\"\n ## assume pass-through\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n d.batchOut = d.batchIn\n #computation\n # --none\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n d.mem.activation = 0 if isNaN(d.mem.activation)\n\n # accuracy layers just pass through\n when \"accuracy\"\n #dimensions\n ## assume pass-through\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n #computation\n # --none\n #memory\n # --none\n\n # power layers: computes outputs y = (shift + scale * x) ^ power\n when \"power\"\n params = n.attribs.power_param\n power = params.power ? 1\n scale = params.scale ? 1\n shift = params.shift ? 0\n #dimensions: pass-through\n d.wOut = d.wIn\n d.hOut = d.hIn\n d.chOut = d.chIn\n #computation\n n_elements = d.wOut * d.hOut * d.chOut\n d.comp.macc = if scale != 1 then n_elements else 0\n d.comp.add = if shift != 0 then n_elements else 0\n d.comp.exp = if power != 1 then n_elements else 0\n #memory\n d.mem.activation = n_elements\n\n # permute layers reorder the channels / dimensions\n when \"permute\"\n permutation = n.attribs.permute_param.order.slice(0) #copy array\n #dimension order: [batch, channels, height, width] according to http://caffe.berkeleyvision.org/tutorial/layers.html\n dim_in = [d.batchIn, d.chIn, d.hIn, d.wIn]\n d.batchOut = dim_in[permutation[0]];\n d.chOut = dim_in[permutation[1]];\n d.hOut = dim_in[permutation[2]];\n d.wOut = dim_in[permutation[3]];\n #computation\n # --none\n #memory\n # --none\n\n # generates prior boxes for SSD networks\n when \"priorbox\"\n settings = n.attribs.prior_box_param\n aspect_ratios = settings.aspect_ratio\n num_priors = settings.min_size * settings.aspect_ratio\n if settings.flip then num_priors *= 2\n\n d.batchOut = d.batchIn\n d.chOut = 2\n d.hOut = 4\n d.wOut = num_priors\n #computation\n # -- neglectable\n #memory\n # --neglectable\n\n # reshape layers just permute dimensions, assume on-the-fly operation\n when \"reshape\"\n #get reshape parameters\n newshape = n.attribs.reshape_param.shape.dim.slice(0) # copy array\n #debugger\n console.log(newshape);\n # 0 as dimension = inherit from input\n if (not newshape[0]) or (newshape[0] == 0) then newshape[0] = d.batchIn\n if (not newshape[1]) or (newshape[1] == 0) then newshape[1] = d.chIn\n if (not newshape[2]) or (newshape[2] == 0) then newshape[2] = d.hIn\n if (not newshape[3]) or (newshape[3] == 0) then newshape[3] = d.wIn\n # -1 as dimension = infer from other dimensions, allowed for at most 1 dimension\n prod_in_dims = d.batchIn * d.wIn * d.hIn * d.chIn\n prod_out_dims = newshape[0] * newshape[1] * newshape[2] * newshape[3] * (-1)# -1 compensates \"-1\" in newshape\n infered_dim = prod_in_dims / prod_out_dims\n if newshape[0] == -1 then newshape[0] = infered_dim\n if newshape[1] == -1 then newshape[1] = infered_dim\n if newshape[2] == -1 then newshape[2] = infered_dim\n if newshape[3] == -1 then newshape[3] = infered_dim\n # assign output dimensions\n d.batchOut = newshape[0]\n d.chOut = newshape[1]\n d.hOut = newshape[2]\n d.wOut = newshape[3]\n #computation\n # --none (some shifting-around only)\n #memory\n\n when \"python\"\n module = n.attribs.python_param.module\n\n if module == \"rpn.proposal_layer\"\n # ASSUME TEST.RPN_POST_NMS_TOP_N = 300\n num_region_proposals = 300 # see RPN_POST_NMS_TOP_N in lib/fast_rcnn/config.py\n\n #output dimensions:\n d.wOut = d.hOut = 1\n d.chOut = 5 # rectangle (x1, y1, x2, y2) (and image batch index n)\n d.batchOut = num_region_proposals\n\n #computation\n d.comp.div = (num_region_proposals*(num_region_proposals-1))/2\n d.comp.macc = d.batchIn * (4+4) * 9*(d.wIn*d.hIn) + 2*(d.comp.div)\n d.comp.add = d.batchIn * (8+2) * 9*(d.wIn*d.hIn) + 6*(d.comp.div)\n d.comp.comp = d.batchIn * (4+2) * 9*(d.wIn*d.hIn) + (9*(d.wIn*d.hIn))**2 + 7*(d.comp.div)\n d.comp.exp = d.batchIn * (2) * 9*(d.wIn*d.hIn)\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n else\n onerror('Unknown Python Layer: '+module)\n console.log(n)\n debugger;\n\n when \"roipooling\"\n # 2 parent layers: region proposals, feature vectors\n roi_proposals = if (n.parents[0].analysis.batchOut > 1) then n.parents[0].analysis else n.parents[1].analysis # parent with batchOut > 1 = region proposals\n feature_map = if (n.parents[0].analysis.batchOut > 1) then n.parents[1].analysis else n.parents[0].analysis # features = the other one\n # Input / Output dimensions\n d.chIn = d.chOut = feature_map.chIn\n d.hIn = feature_map.hIn\n d.wIn = feature_map.wIn\n d.hOut = n.attribs.roi_pooling_param.pooled_h\n d.wOut = n.attribs.roi_pooling_param.pooled_w\n d.batchIn = d.batchOut = roi_proposals.batchOut\n #spatial_scale = n.attribs.roi_pooling_param.spatial_scale\n #computation\n d.comp.add = d.batchOut\n d.comp.div = d.batchOut\n d.comp.macc = d.batchOut\n d.comp.comp = d.batchOut * d.chIn * d.wIn * d.hIn\n #memory\n d.mem.activation = d.wOut*d.hOut*d.chOut*d.batchOut\n\n else # unknown layer; print error message;\n onerror('Unknown Layer: '+layertype)\n console.log(n)\n debugger;\n\n # add dimensions to node attributes so they show in graph tooltips\n trivial_layers = [\"softmax\", \"softmaxwithloss\", \"softmax_loss\", \"dropout\", \"concat\", \"accuracy\"]\n if not ($.inArray(layertype, trivial_layers) >= 0)\n summary = {\n in: \"#{d.chIn}ch ⋅ #{d.wIn}×#{d.hIn} (×#{d.batchIn})\",\n out: \"#{d.chOut}ch ⋅ #{d.wOut}×#{d.hOut} (×#{d.batchOut})\" }\n # concat number of required operations into string\n ops = (val+'⋅'+key for key,val of d.comp when val isnt 0).join(', ')\n #debugger\n summary.ops = ops if ops != \"\"\n # concat memory requirements into string\n mem = (val+'⋅'+key for key,val of d.mem when val isnt 0).join(', ')\n summary.mem = mem if mem != \"\"\n # attach\n _.extend(n.attribs, {analysis: summary});\n\n return net\n" }, { "path": "src/app.coffee", "content": "Renderer = require './renderer.coffee'\nEditor = require './editor.coffee'\n\nmodule.exports =\nclass AppController\n constructor: ->\n @inProgress = false\n @$spinner = $('#net-spinner')\n @$netBox = $('#net-container')\n @$netError = $('#net-error')\n @svg = '#net-svg'\n @$tableBox = $('#table-container')\n @table = '#table-content'\n @setupErrorHandler()\n\n startLoading: (loaderFunc, loader, args...) ->\n if @inProgress\n return\n @$netError.hide()\n @$netBox.hide()\n @$tableBox.hide()\n @$spinner.show()\n loaderFunc args..., (net) => @completeLoading(net, loader)\n\n completeLoading: (net, loader) ->\n @$spinner.hide()\n\n $('#net-title').html(net.name.replace(/_/g, ' '))\n\n $('title').text(net.name.replace(/_/g, ' ')+' — Netscope CNN Analyzer')\n editlink = $(\"(edit)\").addClass(\"editlink\")\n editlink.appendTo $('#net-title')\n editlink.click( => @showEditor(loader))\n\n @$netBox.show()\n @$tableBox.show()\n $(@svg).empty()\n $('.qtip').remove()\n @renderer = new Renderer net, @svg, @table\n\n if not window.do_variants_analysis\n $(\"
\").appendTo @table \n extendlink = $('Excel-compatible Analysis Results (experimental)')\n extendlink.click( => \n window.do_variants_analysis = true\n @renderer.renderTable()\n )\n extendlink.appendTo @table \n\n @inProgress = false\n\n makeLoader: (loaderFunc, loader) ->\n (args...) =>\n @startLoading loaderFunc, loader, args...\n\n showEditor: (loader) ->\n # Display the editor by lazily loading CodeMirror.\n # loader is an instance of a Loader.\n if(_.isUndefined(window.CodeMirror))\n $.getScript 'assets/js/lib/codemirror.min.js', =>\n @netEditor = new Editor(@makeLoader(loader.load, loader), loader)\n else\n @netEditor.reload(loader.load, loader)\n\n setupErrorHandler: ->\n window.onerror = (message, filename, lineno, colno, e) =>\n msg = message\n if not (_.isUndefined(e) || _.isUndefined(e.line) || _.isUndefined(e.column))\n msg = _.template('Line ${line}, Column ${column}: ${message}')(e)\n @$spinner.hide()\n $('.msg', @$netError).html(msg);\n @$netError.show()\n @inProgress = false\n\n" }, { "path": "src/caffe/caffe.coffee", "content": "Parser = require './parser'\nNetwork = require '../network.coffee'\nAnalyzer = require '../analyzer.coffee'\n\ngenerateLayers = (descriptors, phase) ->\n phase ?= 'train'\n layers = []\n for entry in descriptors\n # Support the deprecated Caffe 'layers' key as well.\n layerDesc = entry.layer or entry.layers\n if layerDesc?\n layer = {}\n headerKeys = ['name', 'type', 'top', 'bottom']\n _.extend layer, _.pick(layerDesc, headerKeys)\n layer.attribs = _.omit layerDesc, headerKeys\n layers.push layer\n else\n console.log 'Unidentified entry ignored: ', entry\n layers = _.filter layers, (layer) ->\n layerPhase = layer.attribs.include?.phase\n not (layerPhase? and layerPhase!=phase)\n return layers\n\ngenerateNetwork = (layers, header) ->\n nodeTable = {}\n implicitLayers = []\n net = new Network header.name\n getSingleNode = (name) =>\n node = nodeTable[name]\n # Caffe allows top to be a layer which isn't explicitly\n # defined. Create an implicit layer if this is detected.\n if not node?\n debugger\n node = net.createNode name, 'implicit'\n nodeTable[name] = node\n return node\n getNodes = (names, exclude) =>\n names = [].concat names\n if exclude?\n _.pullAll names, exclude\n _.map names, getSingleNode\n # Build the node LUT.\n for layer in layers\n nodeTable[layer.name] = net.createNode layer.name, layer.type, layer.attribs, {}\n # Connect layers.\n inplaceTable = {}\n for layer in layers\n node = nodeTable[layer.name]\n if layer.top?\n if layer.top==layer.bottom\n # This is an inplace node. We will treat this specially.\n # Note that this would have otherwise introduced a cycle,\n # violating the requirements of a DAG.\n if not inplaceTable[layer.top]?\n inplaceTable[layer.top] = []\n inplaceTable[layer.top].push node\n continue\n else\n node.addChildren getNodes(layer.top, [layer.name])\n if layer.bottom?\n node.addParents getNodes(layer.bottom, [].concat layer.top)\n # Splice in the inplace nodes.\n for own k, inplaceOps of inplaceTable\n curNode = nodeTable[k]\n curNode.coalesce = inplaceOps\n children = curNode.detachChildren()\n for inplaceChild in inplaceOps\n inplaceChild.annotation = 'InPlace'\n curNode.addChild inplaceChild\n curNode = inplaceChild\n curNode.addChildren children\n # Patch in data layer parameters.\n if header?.input? and (header?.input_dim? or header?.input_shape?.dim?)\n inputs = [].concat header.input\n dims = header.input_dim or header.input_shape.dim\n if inputs.length==(dims.length*0.25)\n for input, i in inputs\n dataNode = nodeTable[input]\n dataNode.type = 'data'\n dataNode.attribs.input_param = {\n shape: {\n dim: dims.slice i*4, (i+1)*4\n }\n }\n else\n console.log 'Inconsistent input dimensions.'\n return net\n\nmodule.exports =\nclass CaffeParser\n @parse : (txt, phase) ->\n [header, layerDesc] = Parser.parse txt\n # if header is already a layer instead of a 'global header' with network name etc...:\n if header.layer\n layerDesc.unshift header\n header = {name: 'Unnamed Network'}\n # extract input_shape field from layerDesc to header\n if layerDesc[0].input_dim? or layerDesc[0].input_shape?\n _.extend(header,layerDesc[0])\n layers = generateLayers layerDesc, phase\n network = generateNetwork layers, header\n NetworkAnalyzer = new Analyzer()\n network = NetworkAnalyzer.analyze network\n return network\n" }, { "path": "src/caffe/grammar/proto.pegjs", "content": "Proto_text\n = wsc doc:doc wsc { return doc; }\n\nws \"whitespace\"\n = [ \\t\\n\\r]*\n\nwsc \"whitespace or comment\"\n = ws (comment)* ws\n\ndoc\n = first:value\n rest:(wsc v:value { return v; })*\n\nvalue\n = object\n / pairs\n\ncomment \"comment\"\n = ws \"#\" (!LineTerminator .)*\n\npairs\n = first:pair\n rest:(wsc m:pair { return m; })*\n {\n var result = {};\n var kvPairs = [first].concat(rest);\n for (var i = 0; i < kvPairs.length; i++)\n {\n var k = kvPairs[i].key;\n var v = kvPairs[i].value;\n if(k in result)\n {\n result[k] = [].concat(result[k]);\n result[k].push(v);\n }\n else\n {\n result[k] = v;\n }\n }\n return result;\n }\n\npair\n = key:key ws \":\" ws value:(string / number / key / list )\n {\n return {key: key, value: value};\n }\n\nlist\n = \"[\"\n entries:(ws v:(string / number) ws \",\"? { return v; })*\n ws \"]\"\n {\n return entries;\n }\n\nobject\n = key:key ws \":\"? ws \"{\" wsc\n first:(member)?\n rest:(wsc m:member { return m; })*\n wsc \"}\" wsc\n {\n var elems = [first].concat(rest);\n var merged = {};\n for (var i = 0; i < elems.length; ++i)\n {\n for(var k in elems[i])\n {\n merged[k] = elems[i][k];\n }\n }\n var result = {};\n result[key] = merged;\n return result;\n }\n\nmember\n = (comment / pairs / object )\n\nnumber \"number\"\n = minus? int frac? exp? { return parseFloat(text()); }\n\nexp\n = [eE] (minus / plus)? Digit+\n\nfrac\n = \".\" Digit+\n\nint\n = \"0\" / ([1-9] Digit*)\n\nminus\n = \"-\"\n\nplus\n = \"+\"\n\nstring\n = sstring\n / dstring\n\nsstring\n = \"'\" chars:schar* \"'\" { return chars.join(\"\"); }\n\ndstring\n = '\"' chars:dchar* '\"' { return chars.join(\"\"); }\n\nkey \"key\"\n = chars:[a-zA-Z0-9_-]+ { return chars.join(\"\").toLowerCase(); }\n\ndchar \"double-quoted string character\"\n = [\\x20-\\x21\\x23-\\x5B\\x5D-\\u10FFFF]\n / echar\n\nschar \"single-quoted string character\"\n = [\\x20-\\x26\\x28-\\x5B\\x5D-\\u10FFFF]\n / echar\n\nechar \"escaped character sequence\"\n = \"\\\\\"\n sequence:(\n '\"'\n / \"'\"\n / \"\\\\\"\n / \"/\"\n / \"b\" { return \"\\b\"; }\n / \"f\" { return \"\\f\"; }\n / \"n\" { return \"\\n\"; }\n / \"r\" { return \"\\r\"; }\n / \"t\" { return \"\\t\"; }\n / \"u\" digits:$(HexDigit HexDigit HexDigit HexDigit)\n {\n return String.fromCharCode(parseInt(digits, 16));\n }\n )\n {\n return sequence;\n }\n\nDigit\n = [0-9]\n\nHexDigit\n = [0-9a-f]i\n\nLineTerminator\n = [\\n\\r\\u2028\\u2029]" }, { "path": "src/caffe/parser.js", "content": "module.exports = (function() {\n \"use strict\";\n\n function peg$subclass(child, parent) {\n function ctor() { this.constructor = child; }\n ctor.prototype = parent.prototype;\n child.prototype = new ctor();\n }\n\n function peg$SyntaxError(message, expected, found, location) {\n this.message = message;\n this.expected = expected;\n this.found = found;\n this.location = location;\n this.name = \"SyntaxError\";\n\n if (typeof Error.captureStackTrace === \"function\") {\n Error.captureStackTrace(this, peg$SyntaxError);\n }\n }\n\n peg$subclass(peg$SyntaxError, Error);\n\n function peg$parse(input) {\n var options = arguments.length > 1 ? arguments[1] : {},\n parser = this,\n\n peg$FAILED = {},\n\n peg$startRuleFunctions = { Proto_text: peg$parseProto_text },\n peg$startRuleFunction = peg$parseProto_text,\n\n peg$c0 = function(doc) { return doc; },\n peg$c1 = { type: \"other\", description: \"whitespace\" },\n peg$c2 = /^[ \\t\\n\\r]/,\n peg$c3 = { type: \"class\", value: \"[ \\\\t\\\\n\\\\r]\", description: \"[ \\\\t\\\\n\\\\r]\" },\n peg$c4 = { type: \"other\", description: \"whitespace or comment\" },\n peg$c5 = function(first, v) { return v; },\n peg$c6 = { type: \"other\", description: \"comment\" },\n peg$c7 = \"#\",\n peg$c8 = { type: \"literal\", value: \"#\", description: \"\\\"#\\\"\" },\n peg$c9 = { type: \"any\", description: \"any character\" },\n peg$c10 = function(first, m) { return m; },\n peg$c11 = function(first, rest) {\n var result = {};\n var kvPairs = [first].concat(rest);\n for (var i = 0; i < kvPairs.length; i++)\n {\n var k = kvPairs[i].key;\n var v = kvPairs[i].value;\n if(k in result)\n {\n result[k] = [].concat(result[k]);\n result[k].push(v);\n }\n else\n {\n result[k] = v;\n }\n }\n return result;\n },\n peg$c12 = \":\",\n peg$c13 = { type: \"literal\", value: \":\", description: \"\\\":\\\"\" },\n peg$c14 = function(key, value) {\n return {key: key, value: value};\n },\n peg$c15 = \"[\",\n peg$c16 = { type: \"literal\", value: \"[\", description: \"\\\"[\\\"\" },\n peg$c17 = \",\",\n peg$c18 = { type: \"literal\", value: \",\", description: \"\\\",\\\"\" },\n peg$c19 = function(v) { return v; },\n peg$c20 = \"]\",\n peg$c21 = { type: \"literal\", value: \"]\", description: \"\\\"]\\\"\" },\n peg$c22 = function(entries) {\n return entries;\n },\n peg$c23 = \"{\",\n peg$c24 = { type: \"literal\", value: \"{\", description: \"\\\"{\\\"\" },\n peg$c25 = function(key, first, m) { return m; },\n peg$c26 = \"}\",\n peg$c27 = { type: \"literal\", value: \"}\", description: \"\\\"}\\\"\" },\n peg$c28 = function(key, first, rest) {\n var elems = [first].concat(rest);\n var merged = {};\n for (var i = 0; i < elems.length; ++i)\n {\n for(var k in elems[i])\n {\n merged[k] = elems[i][k];\n }\n }\n var result = {};\n result[key] = merged;\n return result;\n },\n peg$c29 = { type: \"other\", description: \"number\" },\n peg$c30 = function() { return parseFloat(text()); },\n peg$c31 = /^[eE]/,\n peg$c32 = { type: \"class\", value: \"[eE]\", description: \"[eE]\" },\n peg$c33 = \".\",\n peg$c34 = { type: \"literal\", value: \".\", description: \"\\\".\\\"\" },\n peg$c35 = \"0\",\n peg$c36 = { type: \"literal\", value: \"0\", description: \"\\\"0\\\"\" },\n peg$c37 = /^[1-9]/,\n peg$c38 = { type: \"class\", value: \"[1-9]\", description: \"[1-9]\" },\n peg$c39 = \"-\",\n peg$c40 = { type: \"literal\", value: \"-\", description: \"\\\"-\\\"\" },\n peg$c41 = \"+\",\n peg$c42 = { type: \"literal\", value: \"+\", description: \"\\\"+\\\"\" },\n peg$c43 = \"'\",\n peg$c44 = { type: \"literal\", value: \"'\", description: \"\\\"'\\\"\" },\n peg$c45 = function(chars) { return chars.join(\"\"); },\n peg$c46 = \"\\\"\",\n peg$c47 = { type: \"literal\", value: \"\\\"\", description: \"\\\"\\\\\\\"\\\"\" },\n peg$c48 = { type: \"other\", description: \"key\" },\n peg$c49 = /^[a-zA-Z0-9_\\-]/,\n peg$c50 = { type: \"class\", value: \"[a-zA-Z0-9_-]\", description: \"[a-zA-Z0-9_-]\" },\n peg$c51 = function(chars) { return chars.join(\"\").toLowerCase(); },\n peg$c52 = { type: \"other\", description: \"double-quoted string character\" },\n peg$c53 = /^[ -!#-[\\]-\\u10FFFF]/,\n peg$c54 = { type: \"class\", value: \"[\\\\x20-\\\\x21\\\\x23-\\\\x5B\\\\x5D-\\\\u10FFFF]\", description: \"[\\\\x20-\\\\x21\\\\x23-\\\\x5B\\\\x5D-\\\\u10FFFF]\" },\n peg$c55 = { type: \"other\", description: \"single-quoted string character\" },\n peg$c56 = /^[ -&(-[\\]-\\u10FFFF]/,\n peg$c57 = { type: \"class\", value: \"[\\\\x20-\\\\x26\\\\x28-\\\\x5B\\\\x5D-\\\\u10FFFF]\", description: \"[\\\\x20-\\\\x26\\\\x28-\\\\x5B\\\\x5D-\\\\u10FFFF]\" },\n peg$c58 = { type: \"other\", description: \"escaped character sequence\" },\n peg$c59 = \"\\\\\",\n peg$c60 = { type: \"literal\", value: \"\\\\\", description: \"\\\"\\\\\\\\\\\"\" },\n peg$c61 = \"/\",\n peg$c62 = { type: \"literal\", value: \"/\", description: \"\\\"/\\\"\" },\n peg$c63 = \"b\",\n peg$c64 = { type: \"literal\", value: \"b\", description: \"\\\"b\\\"\" },\n peg$c65 = function() { return \"\\b\"; },\n peg$c66 = \"f\",\n peg$c67 = { type: \"literal\", value: \"f\", description: \"\\\"f\\\"\" },\n peg$c68 = function() { return \"\\f\"; },\n peg$c69 = \"n\",\n peg$c70 = { type: \"literal\", value: \"n\", description: \"\\\"n\\\"\" },\n peg$c71 = function() { return \"\\n\"; },\n peg$c72 = \"r\",\n peg$c73 = { type: \"literal\", value: \"r\", description: \"\\\"r\\\"\" },\n peg$c74 = function() { return \"\\r\"; },\n peg$c75 = \"t\",\n peg$c76 = { type: \"literal\", value: \"t\", description: \"\\\"t\\\"\" },\n peg$c77 = function() { return \"\\t\"; },\n peg$c78 = \"u\",\n peg$c79 = { type: \"literal\", value: \"u\", description: \"\\\"u\\\"\" },\n peg$c80 = function(digits) {\n return String.fromCharCode(parseInt(digits, 16));\n },\n peg$c81 = function(sequence) {\n return sequence;\n },\n peg$c82 = /^[0-9]/,\n peg$c83 = { type: \"class\", value: \"[0-9]\", description: \"[0-9]\" },\n peg$c84 = /^[0-9a-f]/i,\n peg$c85 = { type: \"class\", value: \"[0-9a-f]i\", description: \"[0-9a-f]i\" },\n peg$c86 = /^[\\n\\r\\u2028\\u2029]/,\n peg$c87 = { type: \"class\", value: \"[\\\\n\\\\r\\\\u2028\\\\u2029]\", description: \"[\\\\n\\\\r\\\\u2028\\\\u2029]\" },\n\n peg$currPos = 0,\n peg$savedPos = 0,\n peg$posDetailsCache = [{ line: 1, column: 1, seenCR: false }],\n peg$maxFailPos = 0,\n peg$maxFailExpected = [],\n peg$silentFails = 0,\n\n peg$result;\n\n if (\"startRule\" in options) {\n if (!(options.startRule in peg$startRuleFunctions)) {\n throw new Error(\"Can't start parsing from rule \\\"\" + options.startRule + \"\\\".\");\n }\n\n peg$startRuleFunction = peg$startRuleFunctions[options.startRule];\n }\n\n function text() {\n return input.substring(peg$savedPos, peg$currPos);\n }\n\n function location() {\n return peg$computeLocation(peg$savedPos, peg$currPos);\n }\n\n function expected(description) {\n throw peg$buildException(\n null,\n [{ type: \"other\", description: description }],\n input.substring(peg$savedPos, peg$currPos),\n peg$computeLocation(peg$savedPos, peg$currPos)\n );\n }\n\n function error(message) {\n throw peg$buildException(\n message,\n null,\n input.substring(peg$savedPos, peg$currPos),\n peg$computeLocation(peg$savedPos, peg$currPos)\n );\n }\n\n function peg$computePosDetails(pos) {\n var details = peg$posDetailsCache[pos],\n p, ch;\n\n if (details) {\n return details;\n } else {\n p = pos - 1;\n while (!peg$posDetailsCache[p]) {\n p--;\n }\n\n details = peg$posDetailsCache[p];\n details = {\n line: details.line,\n column: details.column,\n seenCR: details.seenCR\n };\n\n while (p < pos) {\n ch = input.charAt(p);\n if (ch === \"\\n\") {\n if (!details.seenCR) { details.line++; }\n details.column = 1;\n details.seenCR = false;\n } else if (ch === \"\\r\" || ch === \"\\u2028\" || ch === \"\\u2029\") {\n details.line++;\n details.column = 1;\n details.seenCR = true;\n } else {\n details.column++;\n details.seenCR = false;\n }\n\n p++;\n }\n\n peg$posDetailsCache[pos] = details;\n return details;\n }\n }\n\n function peg$computeLocation(startPos, endPos) {\n var startPosDetails = peg$computePosDetails(startPos),\n endPosDetails = peg$computePosDetails(endPos);\n\n return {\n start: {\n offset: startPos,\n line: startPosDetails.line,\n column: startPosDetails.column\n },\n end: {\n offset: endPos,\n line: endPosDetails.line,\n column: endPosDetails.column\n }\n };\n }\n\n function peg$fail(expected) {\n if (peg$currPos < peg$maxFailPos) { return; }\n\n if (peg$currPos > peg$maxFailPos) {\n peg$maxFailPos = peg$currPos;\n peg$maxFailExpected = [];\n }\n\n peg$maxFailExpected.push(expected);\n }\n\n function peg$buildException(message, expected, found, location) {\n function cleanupExpected(expected) {\n var i = 1;\n\n expected.sort(function(a, b) {\n if (a.description < b.description) {\n return -1;\n } else if (a.description > b.description) {\n return 1;\n } else {\n return 0;\n }\n });\n\n while (i < expected.length) {\n if (expected[i - 1] === expected[i]) {\n expected.splice(i, 1);\n } else {\n i++;\n }\n }\n }\n\n function buildMessage(expected, found) {\n function stringEscape(s) {\n function hex(ch) { return ch.charCodeAt(0).toString(16).toUpperCase(); }\n\n return s\n .replace(/\\\\/g, '\\\\\\\\')\n .replace(/\"/g, '\\\\\"')\n .replace(/\\x08/g, '\\\\b')\n .replace(/\\t/g, '\\\\t')\n .replace(/\\n/g, '\\\\n')\n .replace(/\\f/g, '\\\\f')\n .replace(/\\r/g, '\\\\r')\n .replace(/[\\x00-\\x07\\x0B\\x0E\\x0F]/g, function(ch) { return '\\\\x0' + hex(ch); })\n .replace(/[\\x10-\\x1F\\x80-\\xFF]/g, function(ch) { return '\\\\x' + hex(ch); })\n .replace(/[\\u0100-\\u0FFF]/g, function(ch) { return '\\\\u0' + hex(ch); })\n .replace(/[\\u1000-\\uFFFF]/g, function(ch) { return '\\\\u' + hex(ch); });\n }\n\n var expectedDescs = new Array(expected.length),\n expectedDesc, foundDesc, i;\n\n for (i = 0; i < expected.length; i++) {\n expectedDescs[i] = expected[i].description;\n }\n\n expectedDesc = expected.length > 1\n ? expectedDescs.slice(0, -1).join(\", \")\n + \" or \"\n + expectedDescs[expected.length - 1]\n : expectedDescs[0];\n\n foundDesc = found ? \"\\\"\" + stringEscape(found) + \"\\\"\" : \"end of input\";\n\n return \"Expected \" + expectedDesc + \" but \" + foundDesc + \" found.\";\n }\n\n if (expected !== null) {\n cleanupExpected(expected);\n }\n\n return new peg$SyntaxError(\n message !== null ? message : buildMessage(expected, found),\n expected,\n found,\n location\n );\n }\n\n function peg$parseProto_text() {\n var s0, s1, s2, s3;\n\n s0 = peg$currPos;\n s1 = peg$parsewsc();\n if (s1 !== peg$FAILED) {\n s2 = peg$parsedoc();\n if (s2 !== peg$FAILED) {\n s3 = peg$parsewsc();\n if (s3 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c0(s2);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parsews() {\n var s0, s1;\n\n peg$silentFails++;\n s0 = [];\n if (peg$c2.test(input.charAt(peg$currPos))) {\n s1 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c3); }\n }\n while (s1 !== peg$FAILED) {\n s0.push(s1);\n if (peg$c2.test(input.charAt(peg$currPos))) {\n s1 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c3); }\n }\n }\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c1); }\n }\n\n return s0;\n }\n\n function peg$parsewsc() {\n var s0, s1, s2, s3;\n\n peg$silentFails++;\n s0 = peg$currPos;\n s1 = peg$parsews();\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$parsecomment();\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$parsecomment();\n }\n if (s2 !== peg$FAILED) {\n s3 = peg$parsews();\n if (s3 !== peg$FAILED) {\n s1 = [s1, s2, s3];\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c4); }\n }\n\n return s0;\n }\n\n function peg$parsedoc() {\n var s0, s1, s2, s3, s4, s5;\n\n s0 = peg$currPos;\n s1 = peg$parsevalue();\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$currPos;\n s4 = peg$parsewsc();\n if (s4 !== peg$FAILED) {\n s5 = peg$parsevalue();\n if (s5 !== peg$FAILED) {\n peg$savedPos = s3;\n s4 = peg$c5(s1, s5);\n s3 = s4;\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$currPos;\n s4 = peg$parsewsc();\n if (s4 !== peg$FAILED) {\n s5 = peg$parsevalue();\n if (s5 !== peg$FAILED) {\n peg$savedPos = s3;\n s4 = peg$c5(s1, s5);\n s3 = s4;\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n }\n if (s2 !== peg$FAILED) {\n s1 = [s1, s2];\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parsevalue() {\n var s0;\n\n s0 = peg$parseobject();\n if (s0 === peg$FAILED) {\n s0 = peg$parsepairs();\n }\n\n return s0;\n }\n\n function peg$parsecomment() {\n var s0, s1, s2, s3, s4, s5, s6;\n\n peg$silentFails++;\n s0 = peg$currPos;\n s1 = peg$parsews();\n if (s1 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 35) {\n s2 = peg$c7;\n peg$currPos++;\n } else {\n s2 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c8); }\n }\n if (s2 !== peg$FAILED) {\n s3 = [];\n s4 = peg$currPos;\n s5 = peg$currPos;\n peg$silentFails++;\n s6 = peg$parseLineTerminator();\n peg$silentFails--;\n if (s6 === peg$FAILED) {\n s5 = void 0;\n } else {\n peg$currPos = s5;\n s5 = peg$FAILED;\n }\n if (s5 !== peg$FAILED) {\n if (input.length > peg$currPos) {\n s6 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s6 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c9); }\n }\n if (s6 !== peg$FAILED) {\n s5 = [s5, s6];\n s4 = s5;\n } else {\n peg$currPos = s4;\n s4 = peg$FAILED;\n }\n } else {\n peg$currPos = s4;\n s4 = peg$FAILED;\n }\n while (s4 !== peg$FAILED) {\n s3.push(s4);\n s4 = peg$currPos;\n s5 = peg$currPos;\n peg$silentFails++;\n s6 = peg$parseLineTerminator();\n peg$silentFails--;\n if (s6 === peg$FAILED) {\n s5 = void 0;\n } else {\n peg$currPos = s5;\n s5 = peg$FAILED;\n }\n if (s5 !== peg$FAILED) {\n if (input.length > peg$currPos) {\n s6 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s6 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c9); }\n }\n if (s6 !== peg$FAILED) {\n s5 = [s5, s6];\n s4 = s5;\n } else {\n peg$currPos = s4;\n s4 = peg$FAILED;\n }\n } else {\n peg$currPos = s4;\n s4 = peg$FAILED;\n }\n }\n if (s3 !== peg$FAILED) {\n s1 = [s1, s2, s3];\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c6); }\n }\n\n return s0;\n }\n\n function peg$parsepairs() {\n var s0, s1, s2, s3, s4, s5;\n\n s0 = peg$currPos;\n s1 = peg$parsepair();\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$currPos;\n s4 = peg$parsewsc();\n if (s4 !== peg$FAILED) {\n s5 = peg$parsepair();\n if (s5 !== peg$FAILED) {\n peg$savedPos = s3;\n s4 = peg$c10(s1, s5);\n s3 = s4;\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$currPos;\n s4 = peg$parsewsc();\n if (s4 !== peg$FAILED) {\n s5 = peg$parsepair();\n if (s5 !== peg$FAILED) {\n peg$savedPos = s3;\n s4 = peg$c10(s1, s5);\n s3 = s4;\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n }\n if (s2 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c11(s1, s2);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parsepair() {\n var s0, s1, s2, s3, s4, s5;\n\n s0 = peg$currPos;\n s1 = peg$parsekey();\n if (s1 !== peg$FAILED) {\n s2 = peg$parsews();\n if (s2 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 58) {\n s3 = peg$c12;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c13); }\n }\n if (s3 !== peg$FAILED) {\n s4 = peg$parsews();\n if (s4 !== peg$FAILED) {\n s5 = peg$parsestring();\n if (s5 === peg$FAILED) {\n s5 = peg$parsenumber();\n if (s5 === peg$FAILED) {\n s5 = peg$parsekey();\n if (s5 === peg$FAILED) {\n s5 = peg$parselist();\n }\n }\n }\n if (s5 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c14(s1, s5);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parselist() {\n var s0, s1, s2, s3, s4, s5, s6, s7;\n\n s0 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 91) {\n s1 = peg$c15;\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c16); }\n }\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$currPos;\n s4 = peg$parsews();\n if (s4 !== peg$FAILED) {\n s5 = peg$parsestring();\n if (s5 === peg$FAILED) {\n s5 = peg$parsenumber();\n }\n if (s5 !== peg$FAILED) {\n s6 = peg$parsews();\n if (s6 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 44) {\n s7 = peg$c17;\n peg$currPos++;\n } else {\n s7 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c18); }\n }\n if (s7 === peg$FAILED) {\n s7 = null;\n }\n if (s7 !== peg$FAILED) {\n peg$savedPos = s3;\n s4 = peg$c19(s5);\n s3 = s4;\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$currPos;\n s4 = peg$parsews();\n if (s4 !== peg$FAILED) {\n s5 = peg$parsestring();\n if (s5 === peg$FAILED) {\n s5 = peg$parsenumber();\n }\n if (s5 !== peg$FAILED) {\n s6 = peg$parsews();\n if (s6 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 44) {\n s7 = peg$c17;\n peg$currPos++;\n } else {\n s7 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c18); }\n }\n if (s7 === peg$FAILED) {\n s7 = null;\n }\n if (s7 !== peg$FAILED) {\n peg$savedPos = s3;\n s4 = peg$c19(s5);\n s3 = s4;\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n } else {\n peg$currPos = s3;\n s3 = peg$FAILED;\n }\n }\n if (s2 !== peg$FAILED) {\n s3 = peg$parsews();\n if (s3 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 93) {\n s4 = peg$c20;\n peg$currPos++;\n } else {\n s4 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c21); }\n }\n if (s4 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c22(s2);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parseobject() {\n var s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11;\n\n s0 = peg$currPos;\n s1 = peg$parsekey();\n if (s1 !== peg$FAILED) {\n s2 = peg$parsews();\n if (s2 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 58) {\n s3 = peg$c12;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c13); }\n }\n if (s3 === peg$FAILED) {\n s3 = null;\n }\n if (s3 !== peg$FAILED) {\n s4 = peg$parsews();\n if (s4 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 123) {\n s5 = peg$c23;\n peg$currPos++;\n } else {\n s5 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c24); }\n }\n if (s5 !== peg$FAILED) {\n s6 = peg$parsewsc();\n if (s6 !== peg$FAILED) {\n s7 = peg$parsemember();\n if (s7 === peg$FAILED) {\n s7 = null;\n }\n if (s7 !== peg$FAILED) {\n s8 = [];\n s9 = peg$currPos;\n s10 = peg$parsewsc();\n if (s10 !== peg$FAILED) {\n s11 = peg$parsemember();\n if (s11 !== peg$FAILED) {\n peg$savedPos = s9;\n s10 = peg$c25(s1, s7, s11);\n s9 = s10;\n } else {\n peg$currPos = s9;\n s9 = peg$FAILED;\n }\n } else {\n peg$currPos = s9;\n s9 = peg$FAILED;\n }\n while (s9 !== peg$FAILED) {\n s8.push(s9);\n s9 = peg$currPos;\n s10 = peg$parsewsc();\n if (s10 !== peg$FAILED) {\n s11 = peg$parsemember();\n if (s11 !== peg$FAILED) {\n peg$savedPos = s9;\n s10 = peg$c25(s1, s7, s11);\n s9 = s10;\n } else {\n peg$currPos = s9;\n s9 = peg$FAILED;\n }\n } else {\n peg$currPos = s9;\n s9 = peg$FAILED;\n }\n }\n if (s8 !== peg$FAILED) {\n s9 = peg$parsewsc();\n if (s9 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 125) {\n s10 = peg$c26;\n peg$currPos++;\n } else {\n s10 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c27); }\n }\n if (s10 !== peg$FAILED) {\n s11 = peg$parsewsc();\n if (s11 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c28(s1, s7, s8);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parsemember() {\n var s0;\n\n s0 = peg$parsecomment();\n if (s0 === peg$FAILED) {\n s0 = peg$parsepairs();\n if (s0 === peg$FAILED) {\n s0 = peg$parseobject();\n }\n }\n\n return s0;\n }\n\n function peg$parsenumber() {\n var s0, s1, s2, s3, s4;\n\n peg$silentFails++;\n s0 = peg$currPos;\n s1 = peg$parseminus();\n if (s1 === peg$FAILED) {\n s1 = null;\n }\n if (s1 !== peg$FAILED) {\n s2 = peg$parseint();\n if (s2 !== peg$FAILED) {\n s3 = peg$parsefrac();\n if (s3 === peg$FAILED) {\n s3 = null;\n }\n if (s3 !== peg$FAILED) {\n s4 = peg$parseexp();\n if (s4 === peg$FAILED) {\n s4 = null;\n }\n if (s4 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c30();\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c29); }\n }\n\n return s0;\n }\n\n function peg$parseexp() {\n var s0, s1, s2, s3, s4;\n\n s0 = peg$currPos;\n if (peg$c31.test(input.charAt(peg$currPos))) {\n s1 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c32); }\n }\n if (s1 !== peg$FAILED) {\n s2 = peg$parseminus();\n if (s2 === peg$FAILED) {\n s2 = peg$parseplus();\n }\n if (s2 === peg$FAILED) {\n s2 = null;\n }\n if (s2 !== peg$FAILED) {\n s3 = [];\n s4 = peg$parseDigit();\n if (s4 !== peg$FAILED) {\n while (s4 !== peg$FAILED) {\n s3.push(s4);\n s4 = peg$parseDigit();\n }\n } else {\n s3 = peg$FAILED;\n }\n if (s3 !== peg$FAILED) {\n s1 = [s1, s2, s3];\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parsefrac() {\n var s0, s1, s2, s3;\n\n s0 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 46) {\n s1 = peg$c33;\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c34); }\n }\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$parseDigit();\n if (s3 !== peg$FAILED) {\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$parseDigit();\n }\n } else {\n s2 = peg$FAILED;\n }\n if (s2 !== peg$FAILED) {\n s1 = [s1, s2];\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parseint() {\n var s0, s1, s2, s3;\n\n if (input.charCodeAt(peg$currPos) === 48) {\n s0 = peg$c35;\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c36); }\n }\n if (s0 === peg$FAILED) {\n s0 = peg$currPos;\n if (peg$c37.test(input.charAt(peg$currPos))) {\n s1 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c38); }\n }\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$parseDigit();\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$parseDigit();\n }\n if (s2 !== peg$FAILED) {\n s1 = [s1, s2];\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n }\n\n return s0;\n }\n\n function peg$parseminus() {\n var s0;\n\n if (input.charCodeAt(peg$currPos) === 45) {\n s0 = peg$c39;\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c40); }\n }\n\n return s0;\n }\n\n function peg$parseplus() {\n var s0;\n\n if (input.charCodeAt(peg$currPos) === 43) {\n s0 = peg$c41;\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c42); }\n }\n\n return s0;\n }\n\n function peg$parsestring() {\n var s0;\n\n s0 = peg$parsesstring();\n if (s0 === peg$FAILED) {\n s0 = peg$parsedstring();\n }\n\n return s0;\n }\n\n function peg$parsesstring() {\n var s0, s1, s2, s3;\n\n s0 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 39) {\n s1 = peg$c43;\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c44); }\n }\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$parseschar();\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$parseschar();\n }\n if (s2 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 39) {\n s3 = peg$c43;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c44); }\n }\n if (s3 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c45(s2);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parsedstring() {\n var s0, s1, s2, s3;\n\n s0 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 34) {\n s1 = peg$c46;\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c47); }\n }\n if (s1 !== peg$FAILED) {\n s2 = [];\n s3 = peg$parsedchar();\n while (s3 !== peg$FAILED) {\n s2.push(s3);\n s3 = peg$parsedchar();\n }\n if (s2 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 34) {\n s3 = peg$c46;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c47); }\n }\n if (s3 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c45(s2);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n\n return s0;\n }\n\n function peg$parsekey() {\n var s0, s1, s2;\n\n peg$silentFails++;\n s0 = peg$currPos;\n s1 = [];\n if (peg$c49.test(input.charAt(peg$currPos))) {\n s2 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s2 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c50); }\n }\n if (s2 !== peg$FAILED) {\n while (s2 !== peg$FAILED) {\n s1.push(s2);\n if (peg$c49.test(input.charAt(peg$currPos))) {\n s2 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s2 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c50); }\n }\n }\n } else {\n s1 = peg$FAILED;\n }\n if (s1 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c51(s1);\n }\n s0 = s1;\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c48); }\n }\n\n return s0;\n }\n\n function peg$parsedchar() {\n var s0, s1;\n\n peg$silentFails++;\n if (peg$c53.test(input.charAt(peg$currPos))) {\n s0 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c54); }\n }\n if (s0 === peg$FAILED) {\n s0 = peg$parseechar();\n }\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c52); }\n }\n\n return s0;\n }\n\n function peg$parseschar() {\n var s0, s1;\n\n peg$silentFails++;\n if (peg$c56.test(input.charAt(peg$currPos))) {\n s0 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c57); }\n }\n if (s0 === peg$FAILED) {\n s0 = peg$parseechar();\n }\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c55); }\n }\n\n return s0;\n }\n\n function peg$parseechar() {\n var s0, s1, s2, s3, s4, s5, s6, s7, s8, s9;\n\n peg$silentFails++;\n s0 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 92) {\n s1 = peg$c59;\n peg$currPos++;\n } else {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c60); }\n }\n if (s1 !== peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 34) {\n s2 = peg$c46;\n peg$currPos++;\n } else {\n s2 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c47); }\n }\n if (s2 === peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 39) {\n s2 = peg$c43;\n peg$currPos++;\n } else {\n s2 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c44); }\n }\n if (s2 === peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 92) {\n s2 = peg$c59;\n peg$currPos++;\n } else {\n s2 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c60); }\n }\n if (s2 === peg$FAILED) {\n if (input.charCodeAt(peg$currPos) === 47) {\n s2 = peg$c61;\n peg$currPos++;\n } else {\n s2 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c62); }\n }\n if (s2 === peg$FAILED) {\n s2 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 98) {\n s3 = peg$c63;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c64); }\n }\n if (s3 !== peg$FAILED) {\n peg$savedPos = s2;\n s3 = peg$c65();\n }\n s2 = s3;\n if (s2 === peg$FAILED) {\n s2 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 102) {\n s3 = peg$c66;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c67); }\n }\n if (s3 !== peg$FAILED) {\n peg$savedPos = s2;\n s3 = peg$c68();\n }\n s2 = s3;\n if (s2 === peg$FAILED) {\n s2 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 110) {\n s3 = peg$c69;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c70); }\n }\n if (s3 !== peg$FAILED) {\n peg$savedPos = s2;\n s3 = peg$c71();\n }\n s2 = s3;\n if (s2 === peg$FAILED) {\n s2 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 114) {\n s3 = peg$c72;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c73); }\n }\n if (s3 !== peg$FAILED) {\n peg$savedPos = s2;\n s3 = peg$c74();\n }\n s2 = s3;\n if (s2 === peg$FAILED) {\n s2 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 116) {\n s3 = peg$c75;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c76); }\n }\n if (s3 !== peg$FAILED) {\n peg$savedPos = s2;\n s3 = peg$c77();\n }\n s2 = s3;\n if (s2 === peg$FAILED) {\n s2 = peg$currPos;\n if (input.charCodeAt(peg$currPos) === 117) {\n s3 = peg$c78;\n peg$currPos++;\n } else {\n s3 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c79); }\n }\n if (s3 !== peg$FAILED) {\n s4 = peg$currPos;\n s5 = peg$currPos;\n s6 = peg$parseHexDigit();\n if (s6 !== peg$FAILED) {\n s7 = peg$parseHexDigit();\n if (s7 !== peg$FAILED) {\n s8 = peg$parseHexDigit();\n if (s8 !== peg$FAILED) {\n s9 = peg$parseHexDigit();\n if (s9 !== peg$FAILED) {\n s6 = [s6, s7, s8, s9];\n s5 = s6;\n } else {\n peg$currPos = s5;\n s5 = peg$FAILED;\n }\n } else {\n peg$currPos = s5;\n s5 = peg$FAILED;\n }\n } else {\n peg$currPos = s5;\n s5 = peg$FAILED;\n }\n } else {\n peg$currPos = s5;\n s5 = peg$FAILED;\n }\n if (s5 !== peg$FAILED) {\n s4 = input.substring(s4, peg$currPos);\n } else {\n s4 = s5;\n }\n if (s4 !== peg$FAILED) {\n peg$savedPos = s2;\n s3 = peg$c80(s4);\n s2 = s3;\n } else {\n peg$currPos = s2;\n s2 = peg$FAILED;\n }\n } else {\n peg$currPos = s2;\n s2 = peg$FAILED;\n }\n }\n }\n }\n }\n }\n }\n }\n }\n }\n if (s2 !== peg$FAILED) {\n peg$savedPos = s0;\n s1 = peg$c81(s2);\n s0 = s1;\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n } else {\n peg$currPos = s0;\n s0 = peg$FAILED;\n }\n peg$silentFails--;\n if (s0 === peg$FAILED) {\n s1 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c58); }\n }\n\n return s0;\n }\n\n function peg$parseDigit() {\n var s0;\n\n if (peg$c82.test(input.charAt(peg$currPos))) {\n s0 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c83); }\n }\n\n return s0;\n }\n\n function peg$parseHexDigit() {\n var s0;\n\n if (peg$c84.test(input.charAt(peg$currPos))) {\n s0 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c85); }\n }\n\n return s0;\n }\n\n function peg$parseLineTerminator() {\n var s0;\n\n if (peg$c86.test(input.charAt(peg$currPos))) {\n s0 = input.charAt(peg$currPos);\n peg$currPos++;\n } else {\n s0 = peg$FAILED;\n if (peg$silentFails === 0) { peg$fail(peg$c87); }\n }\n\n return s0;\n }\n\n peg$result = peg$startRuleFunction();\n\n if (peg$result !== peg$FAILED && peg$currPos === input.length) {\n return peg$result;\n } else {\n if (peg$result !== peg$FAILED && peg$currPos < input.length) {\n peg$fail({ type: \"end\", description: \"end of input\" });\n }\n\n throw peg$buildException(\n null,\n peg$maxFailExpected,\n peg$maxFailPos < input.length ? input.charAt(peg$maxFailPos) : null,\n peg$maxFailPos < input.length\n ? peg$computeLocation(peg$maxFailPos, peg$maxFailPos + 1)\n : peg$computeLocation(peg$maxFailPos, peg$maxFailPos)\n );\n }\n }\n\n return {\n SyntaxError: peg$SyntaxError,\n parse: peg$parse\n };\n})();\n" }, { "path": "src/editor.coffee", "content": "module.exports =\nclass Editor\n constructor: (@loaderFunc, loader) ->\n editorWidthPercentage = 30;\n $editorBox = $($.parseHTML '
')\n $editorBox.width(editorWidthPercentage+'%')\n $('#net-column').width((100-editorWidthPercentage)+'%')\n $('#master-container').prepend $editorBox\n preset = loader.dataLoaded ? '# Enter your network definition here.\\n# Use Shift+Enter to update the visualization.'\n @editor = CodeMirror $editorBox[0],\n value: preset\n lineNumbers : true\n lineWrapping : true\n @editor.on 'keydown', (cm, e) => @onKeyDown(e)\n\n reload: (@loaderFunc, loader) ->\n preset = loader.dataLoaded ? '# Enter your network definition here.\\n# Use Shift+Enter to update the visualization.'\n @editor.setValue(preset)\n #alert(preset)\n\n onKeyDown: (e) ->\n if (e.shiftKey && e.keyCode==13)\n # Using onKeyDown lets us prevent the default action,\n # even if an error is encountered (say, due to parsing).\n # This would not be possible with keymaps.\n e.preventDefault()\n @loaderFunc @editor.getValue()\n" }, { "path": "src/loader.coffee", "content": "module.exports =\nclass Loader\n constructor: (@parser) ->\n # The parser is a unary function that accepts the network source\n # and outputs a Network instance.\n @dataLoaded = null;\n\n fromGist: (gistID, callback) =>\n # Load the model with the given Gist ID.\n url = 'https://api.github.com/gists/'+gistID\n $.getJSON url, (data) =>\n fileSet = data['files']\n isSolitaryFile = Object.keys(fileSet).length==1\n for fileKey of fileSet\n fileInfo = fileSet[fileKey]\n filename = fileInfo['filename'].toLowerCase()\n isProto = _.endsWith filename, '.prototxt'\n isSolver = _.startsWith filename, 'solver'\n if (isProto and not isSolver) or isSolitaryFile\n @load fileInfo['content'], callback\n return\n console.log 'No prototxt found in the given GIST.'\n\n fromURL: (url, callback) =>\n # Load the model from the given URL.\n # This may fail due to same-origin policy.\n $.ajax\n url: url\n success: => @load data, callback\n\n fromPreset: (name, callback) =>\n # Load a preset model. Caffe Only.\n $.get './presets/'+name+'.prototxt', (data) =>\n @load data, callback\n\n load: (data, callback) =>\n @dataLoaded = data\n net = @parser.parse data\n if not _.isUndefined(callback)\n callback net\n return net\n" }, { "path": "src/netscope.coffee", "content": "AppController = require './app.coffee'\nCaffeNetwork = require './caffe/caffe.coffee'\nLoader = require './loader.coffee'\n\nwindow.do_variants_analysis = false\n\nshowDocumentation = ->\n window.location.href = 'quickstart.html'\n\n$(document).ready ->\n app = new AppController()\n # Setup Caffe model loader.\n # This can be replaced with any arbitrary parser to support\n # formats other than Caffe.\n loader = new Loader(CaffeNetwork)\n # Helper function for wrapping the load calls.\n makeLoader = (loadingFunc, loader) ->\n (args...) ->\n app.startLoading loadingFunc, loader, args...\n\n # Register routes\n routes =\n '/gist/:gistID' : makeLoader loader.fromGist, loader\n '/url/(.+)' : makeLoader loader.fromURL, loader\n '/preset/:name' : makeLoader loader.fromPreset, loader\n '/editor(/?)' : => app.showEditor loader\n '/doc' : => showDocumentation()\n router = Router(routes)\n router.init '/doc'" }, { "path": "src/network.coffee", "content": "class Node\n constructor: (@name, @type, @attribs={}, @analysis={}) ->\n @parents = []\n @children = []\n # Nodes to be coalesced (by the renderer) with the current one.\n # For instance, this can be used for grouping in-place operations.\n # Note that this assumes the nodes to be coalesced and the current\n # node form a simple chain structure.\n @coalesce = []\n\n addChild: (child) =>\n if child not in @children\n @children.push child\n if @ not in child.parents\n child.parents.push @\n\n addChildren: (children) =>\n _.forEach children, (c) => @addChild c\n\n addParent: (parent) =>\n parent.addChild @\n\n addParents: (parents) =>\n _.forEach parents, (p) => @addParent p\n\n detachChild: (child) =>\n _.pull @children, child\n _.pull child.parents, @\n\n detachChildren: =>\n children = _.clone @children\n _.forEach children, (c) => @detachChild c\n return children\n\nmodule.exports =\nclass Network\n constructor: (@name='Untitled Network') ->\n @nodes = []\n\n createNode: (label, type, attribs, analysis) ->\n node = new Node label, type, attribs, analysis\n @nodes.push node\n return node\n\n sortTopologically: =>\n sortedNodes = []\n unsortedNodes = _.clone @nodes\n for node in unsortedNodes\n node.sort_ = {temp:false, perm: false}\n visit = (node) ->\n if node.sort_.temp==true\n throw \"Graph is not a DAG.\"\n if node.sort_.perm\n return\n node.sort_.temp = true\n for child in node.children\n visit child\n node.sort_.perm = true\n node.sort_.temp = false\n sortedNodes.unshift node\n while unsortedNodes.length!=0\n visit unsortedNodes.pop()\n for node in sortedNodes\n delete node.sort_\n return sortedNodes\n" }, { "path": "src/renderer.coffee", "content": "Tableify = require('tableify')\nrequire('tablesorter')\n\nmodule.exports =\nclass Renderer\n constructor: (@net, @parent, @table) ->\n @iconify = false\n @layoutDirection = 'tb'\n @generateGraph()\n @renderTable()\n\n setupGraph: ->\n @graph = new dagreD3.graphlib.Graph()\n @graph.setDefaultEdgeLabel ( -> {} )\n @graph.setGraph\n rankdir: @layoutDirection\n ranksep: 10, # Vertical node separation\n nodesep: 5, # Horizontal node separation\n edgesep: 10, # Horizontal edge separation\n marginx: 0, # Horizontal graph margin\n marginy: 0 # Vertical graph margin\n\n generateGraph: ->\n @setupGraph()\n nodes = @net.sortTopologically()\n for node in nodes\n if node.isInGraph\n continue\n layers = [node].concat node.coalesce\n if layers.length>1\n # Rewire the node following the last coalesced node to this one\n lastCoalesed = layers[layers.length-1]\n for child in lastCoalesed.children\n uberParents = _.clone child.parents\n uberParents[uberParents.indexOf lastCoalesed] = node\n child.parents = uberParents\n @insertNode layers\n\n for parent in node.parents\n @insertLink parent, node\n for source in @graph.sources()\n (@graph.node source).class = 'node-type-source'\n for sink in @graph.sinks()\n (@graph.node sink).class = 'node-type-sink'\n @render()\n\n generateTable: ->\n entry = {name: 'start'}\n tbl = []\n id = 0\n worstcasepervariant = null\n\n # Build up Layer Table\n for n in @net.sortTopologically()\n\n # summarize Values in Variant Implementations\n if (do_variants_analysis)\n if (n.analysis.variants.length > 0)\n if not worstcasepervariant # initial copy\n worstcasepervariant = _.cloneDeep(n.analysis.variants)\n variantcopy = _.extend([],n.analysis.variants)\n for variant,idx in variantcopy\n worstcasepervariant[idx][key] = val for key,val of variant when worstcasepervariant[idx][key] < val\n variant[key] = @toSuffixForm(val) for key,val of variant when val > 0\n\n id++\n entry = {\n ID: id\n name: n.name\n type: n.type\n batch: n.analysis.batchIn\n ch_in: n.analysis.chIn\n dim_in: n.analysis.wIn+'x'+n.analysis.hIn\n ch_out: n.analysis.chOut\n dim_out: n.analysis.wOut+'x'+n.analysis.hOut\n ops_raw: n.analysis.comp\n mem_raw: n.analysis.mem\n }\n if (do_variants_analysis) then entry.implementations = n.analysis.variants;\n tbl.push(entry)\n\n if (do_variants_analysis and worstcasepervariant)\n # worst case variant summary\n for variant in worstcasepervariant\n variant[key] = @toSuffixForm(val) for key,val of variant when val > 0\n entry = {\n ID: 999\n name: \"Worst-Case Requirements\"\n implementations: worstcasepervariant\n }\n tbl.push(entry)\n return tbl\n\n toSuffixForm: (num, decimals = 2) ->\n exponents = [12, 9, 6, 3]\n suffices = [\"T\",\"G\",\"M\",\"k\"]\n decimals = Math.pow(10, decimals)\n #debugger\n for exponent,i in exponents\n suffix = suffices[i]\n factor = Math.pow(10, exponent)\n if (num > factor)\n return Math.round(num/factor*decimals)/decimals+suffix\n # too small, no suffix\n return num\n\n summarizeTable: (tbl) ->\n entry = {name: 'start'}\n summary = []\n num_subs = 0\n for n in tbl\n slashindex = n.name.indexOf('/')\n if (slashindex>0 and entry.name.substring(0,slashindex) == n.name.substring(0,slashindex)) # layer has same prefix as current summary item\n num_subs++\n entry.name = n.name.substring(0,slashindex)\n entry.type = 'submodule('+num_subs+')'\n entry.ch_out = n.ch_out\n entry.dim_out = n.dim_out\n entry.ops_raw[key] += n.ops_raw[key] for key of entry.ops_raw\n entry.mem_raw[key] += n.mem_raw[key] for key of entry.mem_raw\n entry.ops[key] = @toSuffixForm(val) for key,val of entry.ops_raw when val > 0\n entry.mem[key] = @toSuffixForm(val) for key,val of entry.mem_raw when val > 0\n #debugger\n summary.pop()\n summary.push(entry)\n else\n num_subs = 0\n entry = {\n ID: n.ID\n name: n.name\n type: n.type\n batch: n.batchIn\n ch_in: n.ch_in\n dim_in: n.dim_in\n ch_out: n.ch_out\n dim_out: n.dim_out\n ops_raw: _.extend({}, n.ops_raw)\n mem_raw: _.extend({}, n.mem_raw)\n ops: {}\n mem: {}\n }\n entry.ops[key] = @toSuffixForm(val) for key,val of entry.ops_raw when val > 0\n entry.mem[key] = @toSuffixForm(val) for key,val of entry.mem_raw when val > 0\n summary.push(entry)\n\n # initialize TOTAL row\n total = {name: 'TOTAL', ops_raw: {}, mem_raw: {}, ops: {}, mem: {}}\n _.extend(total.ops_raw, summary[0].ops_raw) # copy zeros from data layer\n _.extend(total.mem_raw, summary[0].mem_raw) # idem\n total.mem_raw.activation = 0 # data layer already uses activation --> set to zero\n for entry in summary\n #debugger\n total.ops_raw[key] += entry.ops_raw[key] for key of entry.ops_raw\n total.mem_raw[key] += entry.mem_raw[key] for key of entry.mem_raw\n total.ops[key] = @toSuffixForm(val) for key,val of total.ops_raw\n total.mem[key] = @toSuffixForm(val) for key,val of total.mem_raw\n summary.push(total)\n summary_without_raw = (_.omit(entry, ['ops_raw','mem_raw']) for entry in summary)\n return summary_without_raw\n\n renderTable: ->\n # Generate Detail Table and Summary\n detail = @generateTable()\n summary = @summarizeTable(detail)\n $(@table).html('

Summary:

'+Tableify(summary)+\n '

Details:

'+Tableify(detail));\n\n # Add Sorting Headers\n $(@table+' table').tablesorter()\n\n # Add Click-to-Scroll Handlers\n # Closure Function that executes scroll:\n scroll_to = (el) ->\n return () ->\n top_coord = $(el).offset().top-200;\n $(\"body,html\").animate({ scrollTop: top_coord }, 200);\n $(el).addClass 'node-highlight'\n removeHighlight = (node) ->\n return () -> $(node).removeClass 'node-highlight'\n window.setTimeout removeHighlight(el), 4000\n\n # Add Click-to-Scroll to all summary rows, except last\n summary_table = $(@table+' table')[0]\n summary_body = summary_table.children[1]\n row_array = Array.prototype.slice.call(summary_body.children)\n for row in row_array.slice(0,-1)\n # Add Link between Node and Table Element -> both directions work\n $table_elem = $(row.children[1])\n $node_elem = $('div[id^=\"node-'+$table_elem.text()+'\"]')\n $table_elem.click( scroll_to $node_elem )\n $node_elem.click( scroll_to $table_elem )\n\n if do_variants_analysis\n # Calculate Per-Layer Statistics\n areatbl = []\n for entry in detail when (entry.type == \"Convolution\" or entry.type == \"Concat\" or entry.type == \"SoftmaxWithLoss\" or entry.type == \"innerproduct\")\n # extract input dimension:\n dim_in = entry.dim_in?.split(\"x\").pop()\n # add entry\n suffix = \" \" + @net.name\n line = {}\n line[\"layer\"] = entry.name;\n line[\"capacity\"+suffix] = if entry.mem_raw?.activation > 0 then entry.mem_raw.activation else \"\"\n line[\"macc \"+suffix] = if entry.ops_raw?.macc > 0 then entry.ops_raw.macc else \"\"\n line[\"param \"+suffix] = if entry.mem_raw?.param > 0 then entry.mem_raw.param else \"\"\n line[\"ch_out \"+suffix] = entry.ch_out\n line[\"width \"+suffix] = dim_in\n areatbl.push(line)\n $(Tableify(areatbl)).appendTo(@table)\n return null\n\n insertNode: (layers) ->\n baseNode = layers[0]\n nodeClass = 'node-type-'+baseNode.type.replace(/_/g, '-').toLowerCase()\n nodeLabel = ''\n for layer in layers\n layer.isInGraph = true\n nodeLabel += @generateLabel layer\n nodeDesc =\n labelType : 'html'\n label : nodeLabel\n class : nodeClass\n layers : layers\n rx : 5\n ry : 5\n if @iconify\n _.extend nodeDesc,\n shape: 'circle'\n @graph.setNode baseNode.name, nodeDesc\n\n generateLabel: (layer) ->\n if not @iconify\n '
'+layer.name+'
'\n else\n ''\n\n insertLink: (src, dst) ->\n if not @iconify\n ch = src.analysis.chOut ? \"?\"\n w = src.analysis.wOut ? \"?\"\n h = src.analysis.hOut ? \"?\"\n b = src.analysis.batchOut ? \"?\"\n lbl = ch+'ch ⋅ '+w+'×'+h\n lbl += ' (×'+b+')' if b > 1\n else\n lbl = ''\n @graph.setEdge(src.name, dst.name, { arrowhead: 'vee', label: lbl } );\n\n renderKey:(key) ->\n key.replace(/_/g, ' ')\n\n renderValue: (value) ->\n if Array.isArray value\n return value.join(', ')\n return value\n\n renderSection: (section) ->\n s = ''\n for own key of section\n val = section[key]\n isSection = (typeof val is 'object') and not Array.isArray(val)\n if isSection\n s += '
'+@renderKey(key)+'
'\n s += '
'\n s+= @renderSection val\n else\n s += '
'\n s += ''+@renderKey(key)+': '\n s += ''+@renderValue(val)+''\n s += '
'\n return s\n\n tipForNode: (nodeKey) ->\n node = @graph.node nodeKey\n s = ''\n for layer in node.layers\n s += '
'\n s += '
'\n s += ''+layer.name+''\n s += ' · '\n s += ''+@renderKey(layer.type)+''\n if layer.annotation?\n s += ' · '+layer.annotation+''\n s += '
'\n s += @renderSection layer.attribs\n return s\n\n render: ->\n svg = d3.select(@parent)\n svgGroup = svg.append('g')\n graphRender = new dagreD3.render()\n graphRender svgGroup, @graph\n\n # Size to fit.\n # getBBox appears to do the right thing on Chrome,\n # but not on Firefox. getBoundingClientRect works on both.\n bbox = svgGroup.node().getBoundingClientRect()\n svg.attr('width', bbox.width)\n svg.attr('height', bbox.height)\n\n # Configure Tooltips.\n tipPositions =\n tb:\n my: 'left center'\n at: 'right center'\n lr:\n my: 'top center'\n at: 'bottom center'\n that = @\n svgGroup.selectAll(\"g.node\").each (nodeKey) ->\n position = tipPositions[that.layoutDirection]\n position.viewport = $(window)\n $(this).qtip\n content:\n text: that.tipForNode nodeKey\n position: position\n show:\n delay: 0\n effect: false\n hide:\n effect: false\n" } ]