[
{
"path": "LICENSE",
"content": "\n Apache License\n Version 2.0, January 2004\n http://www.apache.org/licenses/\n\n TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION\n\n 1. Definitions.\n\n \"License\" shall mean the terms and conditions for use, reproduction,\n and distribution as defined by Sections 1 through 9 of this document.\n\n \"Licensor\" shall mean the copyright owner or entity authorized by\n the copyright owner that is granting the License.\n\n \"Legal Entity\" shall mean the union of the acting entity and all\n other entities that control, are controlled by, or are under common\n control with that entity. For the purposes of this definition,\n \"control\" means (i) the power, direct or indirect, to cause the\n direction or management of such entity, whether by contract or\n otherwise, or (ii) ownership of fifty percent (50%) or more of the\n outstanding shares, or (iii) beneficial ownership of such entity.\n\n \"You\" (or \"Your\") shall mean an individual or Legal Entity\n exercising permissions granted by this License.\n\n \"Source\" form shall mean the preferred form for making modifications,\n including but not limited to software source code, documentation\n source, and configuration files.\n\n \"Object\" form shall mean any form resulting from mechanical\n transformation or translation of a Source form, including but\n not limited to compiled object code, generated documentation,\n and conversions to other media types.\n\n \"Work\" shall mean the work of authorship, whether in Source or\n Object form, made available under the License, as indicated by a\n copyright notice that is included in or attached to the work\n (an example is provided in the Appendix below).\n\n \"Derivative Works\" shall mean any work, whether in Source or Object\n form, that is based on (or derived from) the Work and for which the\n editorial revisions, annotations, elaborations, or other modifications\n represent, as a whole, an original work of authorship. For the purposes\n of this License, Derivative Works shall not include works that remain\n separable from, or merely link (or bind by name) to the interfaces of,\n the Work and Derivative Works thereof.\n\n \"Contribution\" shall mean any work of authorship, including\n the original version of the Work and any modifications or additions\n to that Work or Derivative Works thereof, that is intentionally\n submitted to Licensor for inclusion in the Work by the copyright owner\n or by an individual or Legal Entity authorized to submit on behalf of\n the copyright owner. For the purposes of this definition, \"submitted\"\n means any form of electronic, verbal, or written communication sent\n to the Licensor or its representatives, including but not limited to\n communication on electronic mailing lists, source code control systems,\n and issue tracking systems that are managed by, or on behalf of, the\n Licensor for the purpose of discussing and improving the Work, but\n excluding communication that is conspicuously marked or otherwise\n designated in writing by the copyright owner as \"Not a Contribution.\"\n\n \"Contributor\" shall mean Licensor and any individual or Legal Entity\n on behalf of whom a Contribution has been received by Licensor and\n subsequently incorporated within the Work.\n\n 2. Grant of Copyright License. Subject to the terms and conditions of\n this License, each Contributor hereby grants to You a perpetual,\n worldwide, non-exclusive, no-charge, royalty-free, irrevocable\n copyright license to reproduce, prepare Derivative Works of,\n publicly display, publicly perform, sublicense, and distribute the\n Work and such Derivative Works in Source or Object form.\n\n 3. Grant of Patent License. Subject to the terms and conditions of\n this License, each Contributor hereby grants to You a perpetual,\n worldwide, non-exclusive, no-charge, royalty-free, irrevocable\n (except as stated in this section) patent license to make, have made,\n use, offer to sell, sell, import, and otherwise transfer the Work,\n where such license applies only to those patent claims licensable\n by such Contributor that are necessarily infringed by their\n Contribution(s) alone or by combination of their Contribution(s)\n with the Work to which such Contribution(s) was submitted. If You\n institute patent litigation against any entity (including a\n cross-claim or counterclaim in a lawsuit) alleging that the Work\n or a Contribution incorporated within the Work constitutes direct\n or contributory patent infringement, then any patent licenses\n granted to You under this License for that Work shall terminate\n as of the date such litigation is filed.\n\n 4. Redistribution. You may reproduce and distribute copies of the\n Work or Derivative Works thereof in any medium, with or without\n modifications, and in Source or Object form, provided that You\n meet the following conditions:\n\n (a) You must give any other recipients of the Work or\n Derivative Works a copy of this License; and\n\n (b) You must cause any modified files to carry prominent notices\n stating that You changed the files; and\n\n (c) You must retain, in the Source form of any Derivative Works\n that You distribute, all copyright, patent, trademark, and\n attribution notices from the Source form of the Work,\n excluding those notices that do not pertain to any part of\n the Derivative Works; and\n\n (d) If the Work includes a \"NOTICE\" text file as part of its\n distribution, then any Derivative Works that You distribute must\n include a readable copy of the attribution notices contained\n within such NOTICE file, excluding those notices that do not\n pertain to any part of the Derivative Works, in at least one\n of the following places: within a NOTICE text file distributed\n as part of the Derivative Works; within the Source form or\n documentation, if provided along with the Derivative Works; or,\n within a display generated by the Derivative Works, if and\n wherever such third-party notices normally appear. The contents\n of the NOTICE file are for informational purposes only and\n do not modify the License. You may add Your own attribution\n notices within Derivative Works that You distribute, alongside\n or as an addendum to the NOTICE text from the Work, provided\n that such additional attribution notices cannot be construed\n as modifying the License.\n\n You may add Your own copyright statement to Your modifications and\n may provide additional or different license terms and conditions\n for use, reproduction, or distribution of Your modifications, or\n for any such Derivative Works as a whole, provided Your use,\n reproduction, and distribution of the Work otherwise complies with\n the conditions stated in this License.\n\n 5. Submission of Contributions. Unless You explicitly state otherwise,\n any Contribution intentionally submitted for inclusion in the Work\n by You to the Licensor shall be under the terms and conditions of\n this License, without any additional terms or conditions.\n Notwithstanding the above, nothing herein shall supersede or modify\n the terms of any separate license agreement you may have executed\n with Licensor regarding such Contributions.\n\n 6. Trademarks. This License does not grant permission to use the trade\n names, trademarks, service marks, or product names of the Licensor,\n except as required for reasonable and customary use in describing the\n origin of the Work and reproducing the content of the NOTICE file.\n\n 7. Disclaimer of Warranty. Unless required by applicable law or\n agreed to in writing, Licensor provides the Work (and each\n Contributor provides its Contributions) on an \"AS IS\" BASIS,\n WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or\n implied, including, without limitation, any warranties or conditions\n of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A\n PARTICULAR PURPOSE. You are solely responsible for determining the\n appropriateness of using or redistributing the Work and assume any\n risks associated with Your exercise of permissions under this License.\n\n 8. Limitation of Liability. In no event and under no legal theory,\n whether in tort (including negligence), contract, or otherwise,\n unless required by applicable law (such as deliberate and grossly\n negligent acts) or agreed to in writing, shall any Contributor be\n liable to You for damages, including any direct, indirect, special,\n incidental, or consequential damages of any character arising as a\n result of this License or out of the use or inability to use the\n Work (including but not limited to damages for loss of goodwill,\n work stoppage, computer failure or malfunction, or any and all\n other commercial damages or losses), even if such Contributor\n has been advised of the possibility of such damages.\n\n 9. Accepting Warranty or Additional Liability. While redistributing\n the Work or Derivative Works thereof, You may choose to offer,\n and charge a fee for, acceptance of support, warranty, indemnity,\n or other liability obligations and/or rights consistent with this\n License. However, in accepting such obligations, You may act only\n on Your own behalf and on Your sole responsibility, not on behalf\n of any other Contributor, and only if You agree to indemnify,\n defend, and hold each Contributor harmless for any liability\n incurred by, or claims asserted against, such Contributor by reason\n of your accepting any such warranty or additional liability.\n\n END OF TERMS AND CONDITIONS\n\n APPENDIX: How to apply the Apache License to your work.\n\n To apply the Apache License to your work, attach the following\n boilerplate notice, with the fields enclosed by brackets \"[]\"\n replaced with your own identifying information. (Don't include\n the brackets!) The text should be enclosed in the appropriate\n comment syntax for the file format. We also recommend that a\n file or class name and description of purpose be included on the\n same \"printed page\" as the copyright notice for easier\n identification within third-party archives.\n\n Copyright 2017 Yinpeng Dong\n\n Licensed under the Apache License, Version 2.0 (the \"License\");\n you may not use this file except in compliance with the License.\n You may obtain a copy of the License at\n\n http://www.apache.org/licenses/LICENSE-2.0\n\n Unless required by applicable law or agreed to in writing, software\n distributed under the License is distributed on an \"AS IS\" BASIS,\n WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n See the License for the specific language governing permissions and\n limitations under the License.\n"
},
{
"path": "README.md",
"content": "# Translation-Invariant Attacks\n\n## Introduction\nThis repository contains the code for [Evading Defenses to Transferable Adversarial Examples by Translation-Invariant Attacks](https://arxiv.org/pdf/1904.02884.pdf) (CVPR 2019 Oral).\n\n## Method\nWe proposed a translation-invariant (TI) attack method to generate more transferable adversarial examples. This method is implemented by convolving the gradient with a pre-defined kernel in each attack iteration, and can be integrated into any gradient-based attack method. \n\n## Run the code\nFirst download the [models](#Models). You can also use other models by changing the model definition part in the code.\nThen run the following command\n```\nbash run_attack.sh input_dir output_dir 16\n```\nwhere original images are stored in ``input_dir`` with ``.png`` format, and the generated adversarial images are saved in ``output_dir``.\nWe used the Python 2.7 and Tensorflow 1.12 versions. \n\n## Results\n\nWe consider eight SOTA defense models on ImageNet:\n\n* Inc-v3ens3, Inc-v3ens4, IncRes-v2ens trained by [Ensemble Adversarial Training](https://arxiv.org/abs/1705.07204);\n* [High-level representation guided denoiser](https://arxiv.org/abs/1712.02976) (HGD, top-1 submission in the NIPS 2017 defense competition);\n* [Input transformation through random resizing and padding](https://arxiv.org/abs/1711.01991) (R&P, rank-2 submission in the NIPS 2017 defense competition);\n* [Input transformation through JPEG compression or total variance minimization (TVM)](https://openreview.net/pdf?id=SyJ7ClWCb);\n* [Rank-3 submission in the NIPS 2017 defense competition (NIPS-r3)](https://github.com/anlthms/nips-2017/tree/master/mmd);\n\nWe attacked these models by the [fast gradient sign method](https://arxiv.org/abs/1412.6572) (FGSM), [momentum iterative fast gradient sign method](https://arxiv.org/abs/1710.06081) (MI-FGSM), [diverse input method](https://arxiv.org/abs/1803.06978) (DIM), and their translation-invariant versions as TI-FGSM, TI-MI-FGSM, and TI-DIM. We generated adversarial examples for the ensemble of Inception V3, Inception V4, Inception ResNet V2, and ResNet V2 152 with epsilon 16. The success rates against the eight defenses are:\n\n![](\"https://github.com/dongyp13/Translation-Invariant-Attacks/blob/master/results.png\")
\n\n### Citation\nIf you use our method for attacks in your research, please consider citing\n\n @inproceedings{dong2019evading,\n title={Evading Defenses to Transferable Adversarial Examples by Translation-Invariant Attacks},\n author={Dong, Yinpeng and Pang, Tianyu and Su, Hang and Zhu, Jun},\n booktitle={Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition},\n year={2019}\n }\n\n## Implementation\n\n### Models\nThe models can be downloaded at [Inception V3](http://ml.cs.tsinghua.edu.cn/~yinpeng/downloads/inception_v3.ckpt), [Inception V4](http://ml.cs.tsinghua.edu.cn/~yinpeng/downloads/inception_v4.ckpt), [Inception ResNet V2](http://ml.cs.tsinghua.edu.cn/~yinpeng/downloads/inception_resnet_v2_2016_08_30.ckpt), and [ResNet V2 152](http://ml.cs.tsinghua.edu.cn/~yinpeng/downloads/resnet_v2_152.ckpt).\n\nIf you want to attack other models, you can replace the model definition part to your own models.\n\n### Hyper-parameters\n* For TI-FGSM, set ``num_iter=1``, ``momentum=0.0``, ``prob=0.0``;\n* For TI-MI-FGSM, set ``num_iter=10``, ``momentum=1.0``, ``prob=0.0``;\n* For TI-DIM, set ``num_iter=10``, ``momentum=1.0``, ``prob=0.7``;\n"
},
{
"path": "attack_iter.py",
"content": "\"\"\"Implementation of sample attack.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport os\nimport time\nstart_time = time.time()\n\nimport numpy as np\nfrom scipy.misc import imread\nfrom scipy.misc import imsave\nfrom scipy.misc import imresize\n\nimport tensorflow as tf\n\nfrom nets import inception_v3, inception_v4, inception_resnet_v2, resnet_v2\n\nslim = tf.contrib.slim\n\n\ntf.flags.DEFINE_string('master', '', 'The address of the TensorFlow master to use.')\n\ntf.flags.DEFINE_string('checkpoint_path_inception_v3', '', 'Path to checkpoint for inception network.')\n\ntf.flags.DEFINE_string('checkpoint_path_inception_v4', '', 'Path to checkpoint for inception network.')\n\ntf.flags.DEFINE_string('checkpoint_path_inception_resnet_v2', '', 'Path to checkpoint for inception network.')\n\ntf.flags.DEFINE_string('checkpoint_path_resnet', '', 'Path to checkpoint for inception network.')\n\ntf.flags.DEFINE_string('input_dir', '', 'Input directory with images.')\n\ntf.flags.DEFINE_string('output_dir', '', 'Output directory with images.')\n\ntf.flags.DEFINE_float('max_epsilon', 32.0, 'Maximum size of adversarial perturbation.')\n\ntf.flags.DEFINE_integer('num_iter', 10, 'Number of iterations.')\n\ntf.flags.DEFINE_integer('image_width', 299, 'Width of each input images.')\n\ntf.flags.DEFINE_integer('image_height', 299, 'Height of each input images.')\n\ntf.flags.DEFINE_integer('image_resize', 330, 'Height of each input images.')\n\ntf.flags.DEFINE_integer('batch_size', 10, 'How many images process at one time.')\n\ntf.flags.DEFINE_float('momentum', 1.0, 'Momentum.')\n\ntf.flags.DEFINE_float('prob', 0.4, 'probability of using diverse inputs.')\n\nFLAGS = tf.flags.FLAGS\n\ndef gkern(kernlen=21, nsig=3):\n \"\"\"Returns a 2D Gaussian kernel array.\"\"\"\n import scipy.stats as st\n\n x = np.linspace(-nsig, nsig, kernlen)\n kern1d = st.norm.pdf(x)\n kernel_raw = np.outer(kern1d, kern1d)\n kernel = kernel_raw / kernel_raw.sum()\n return kernel\n\nkernel = gkern(15, 3).astype(np.float32)\nstack_kernel = np.stack([kernel, kernel, kernel]).swapaxes(2, 0)\nstack_kernel = np.expand_dims(stack_kernel, 3)\n\ndef load_images(input_dir, batch_shape):\n \"\"\"Read png images from input directory in batches.\n\n Args:\n input_dir: input directory\n batch_shape: shape of minibatch array, i.e. [batch_size, height, width, 3]\n\n Yields:\n filenames: list file names without path of each image\n Lenght of this list could be less than batch_size, in this case only\n first few images of the result are elements of the minibatch.\n images: array with all images from this batch\n \"\"\"\n images = np.zeros(batch_shape)\n filenames = []\n idx = 0\n batch_size = batch_shape[0]\n for filepath in tf.gfile.Glob(os.path.join(input_dir, '*.png')):\n with tf.gfile.Open(filepath) as f:\n image = imresize(imread(f, mode='RGB'), [FLAGS.image_height, FLAGS.image_width]).astype(np.float) / 255.0\n # Images for inception classifier are normalized to be in [-1, 1] interval.\n images[idx, :, :, :] = image * 2.0 - 1.0\n filenames.append(os.path.basename(filepath))\n idx += 1\n if idx == batch_size:\n yield filenames, images\n filenames = []\n images = np.zeros(batch_shape)\n idx = 0\n if idx > 0:\n yield filenames, images\n\n\ndef save_images(images, filenames, output_dir):\n \"\"\"Saves images to the output directory.\n\n Args:\n images: array with minibatch of images\n filenames: list of filenames without path\n If number of file names in this list less than number of images in\n the minibatch then only first len(filenames) images will be saved.\n output_dir: directory where to save images\n \"\"\"\n for i, filename in enumerate(filenames):\n # Images for inception classifier are normalized to be in [-1, 1] interval,\n # so rescale them back to [0, 1].\n with tf.gfile.Open(os.path.join(output_dir, filename), 'w') as f:\n imsave(f, (images[i, :, :, :] + 1.0) * 0.5, format='png')\n\n\ndef graph(x, y, i, x_max, x_min, grad):\n eps = 2.0 * FLAGS.max_epsilon / 255.0\n num_iter = FLAGS.num_iter\n alpha = eps / num_iter\n momentum = FLAGS.momentum\n num_classes = 1001\n\n # should keep original x here for output\n\n with slim.arg_scope(inception_v3.inception_v3_arg_scope()):\n logits_v3, end_points_v3 = inception_v3.inception_v3(\n input_diversity(x), num_classes=num_classes, is_training=False)\n\n with slim.arg_scope(inception_v4.inception_v4_arg_scope()):\n logits_v4, end_points_v4 = inception_v4.inception_v4(\n input_diversity(x), num_classes=num_classes, is_training=False)\n\n with slim.arg_scope(inception_resnet_v2.inception_resnet_v2_arg_scope()):\n logits_res_v2, end_points_res_v2 = inception_resnet_v2.inception_resnet_v2(\n input_diversity(x), num_classes=num_classes, is_training=False, reuse=True)\n\n with slim.arg_scope(resnet_v2.resnet_arg_scope()):\n logits_resnet, end_points_resnet = resnet_v2.resnet_v2_152(\n input_diversity(x), num_classes=num_classes, is_training=False)\n\n logits = (logits_v3 + logits_v4 + logits_res_v2 + logits_resnet) / 4\n auxlogits = (end_points_v3['AuxLogits'] + end_points_v4['AuxLogits'] + end_points_res_v2['AuxLogits']) / 3\n cross_entropy = tf.losses.softmax_cross_entropy(y,\n logits,\n label_smoothing=0.0,\n weights=1.0)\n cross_entropy += tf.losses.softmax_cross_entropy(y,\n auxlogits,\n label_smoothing=0.0,\n weights=0.4)\n noise = tf.gradients(cross_entropy, x)[0]\n noise = tf.nn.depthwise_conv2d(noise, stack_kernel, strides=[1, 1, 1, 1], padding='SAME')\n noise = noise / tf.reduce_mean(tf.abs(noise), [1, 2, 3], keep_dims=True)\n noise = momentum * grad + noise\n x = x + alpha * tf.sign(noise)\n x = tf.clip_by_value(x, x_min, x_max)\n i = tf.add(i, 1)\n return x, y, i, x_max, x_min, noise\n\n\ndef stop(x, y, i, x_max, x_min, grad):\n num_iter = FLAGS.num_iter\n return tf.less(i, num_iter)\n\n\ndef input_diversity(input_tensor):\n \n rnd = tf.random_uniform((), FLAGS.image_width, FLAGS.image_resize, dtype=tf.int32)\n rescaled = tf.image.resize_images(input_tensor, [rnd, rnd], method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)\n h_rem = FLAGS.image_resize - rnd\n w_rem = FLAGS.image_resize - rnd\n pad_top = tf.random_uniform((), 0, h_rem, dtype=tf.int32)\n pad_bottom = h_rem - pad_top\n pad_left = tf.random_uniform((), 0, w_rem, dtype=tf.int32)\n pad_right = w_rem - pad_left\n padded = tf.pad(rescaled, [[0, 0], [pad_top, pad_bottom], [pad_left, pad_right], [0, 0]], constant_values=0.)\n padded.set_shape((input_tensor.shape[0], FLAGS.image_resize, FLAGS.image_resize, 3))\n return tf.cond(tf.random_uniform(shape=[1])[0] < tf.constant(FLAGS.prob), lambda: padded, lambda: input_tensor)\n\n\ndef main(_):\n # Images for inception classifier are normalized to be in [-1, 1] interval,\n # eps is a difference between pixels so it should be in [0, 2] interval.\n # Renormalizing epsilon from [0, 255] to [0, 2].\n eps = 2.0 * FLAGS.max_epsilon / 255.0\n num_classes = 1001\n batch_shape = [FLAGS.batch_size, FLAGS.image_height, FLAGS.image_width, 3]\n\n tf.logging.set_verbosity(tf.logging.INFO)\n\n print(time.time() - start_time)\n\n with tf.Graph().as_default():\n # Prepare graph\n x_input = tf.placeholder(tf.float32, shape=batch_shape)\n x_max = tf.clip_by_value(x_input + eps, -1.0, 1.0)\n x_min = tf.clip_by_value(x_input - eps, -1.0, 1.0)\n\n with slim.arg_scope(inception_resnet_v2.inception_resnet_v2_arg_scope()):\n _, end_points = inception_resnet_v2.inception_resnet_v2(\n x_input, num_classes=num_classes, is_training=False)\n\n predicted_labels = tf.argmax(end_points['Predictions'], 1)\n y = tf.one_hot(predicted_labels, num_classes)\n\n i = tf.constant(0)\n grad = tf.zeros(shape=batch_shape)\n x_adv, _, _, _, _, _ = tf.while_loop(stop, graph, [x_input, y, i, x_max, x_min, grad])\n\n # Run computation\n s1 = tf.train.Saver(slim.get_model_variables(scope='InceptionV3'))\n s5 = tf.train.Saver(slim.get_model_variables(scope='InceptionV4'))\n s6 = tf.train.Saver(slim.get_model_variables(scope='InceptionResnetV2'))\n s8 = tf.train.Saver(slim.get_model_variables(scope='resnet_v2'))\n\n with tf.Session() as sess:\n s1.restore(sess, FLAGS.checkpoint_path_inception_v3)\n s5.restore(sess, FLAGS.checkpoint_path_inception_v4)\n s6.restore(sess, FLAGS.checkpoint_path_inception_resnet_v2)\n s8.restore(sess, FLAGS.checkpoint_path_resnet)\n print(time.time() - start_time)\n\n for filenames, images in load_images(FLAGS.input_dir, batch_shape):\n adv_images = sess.run(x_adv, feed_dict={x_input: images})\n save_images(adv_images, filenames, FLAGS.output_dir)\n\n print(time.time() - start_time)\n\n\nif __name__ == '__main__':\n tf.app.run()\n"
},
{
"path": "attack_vgg16_mim.py",
"content": "\"\"\"Implementation of sample attack.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport os\nimport time\n\nimport numpy as np\nfrom scipy.misc import imread\nfrom scipy.misc import imsave, imresize\n\nimport tensorflow as tf\nfrom nets import vgg\n\nslim = tf.contrib.slim\n\n_R_MEAN = 123.68\n_G_MEAN = 116.78\n_B_MEAN = 103.94\n\n\ntf.flags.DEFINE_string(\n 'master', '', 'The address of the TensorFlow master to use.')\n\ntf.flags.DEFINE_string(\n 'checkpoint_path_vgg', '', 'Path to checkpoint for inception network.')\n\ntf.flags.DEFINE_string(\n 'input_dir', '', 'Input directory with images.')\n\ntf.flags.DEFINE_string(\n 'output_dir', '', 'Output directory with images.')\n\ntf.flags.DEFINE_float(\n 'max_epsilon', 16.0, 'Maximum size of adversarial perturbation.')\n\ntf.flags.DEFINE_integer(\n 'num_iter', 10, 'Number of iterations.')\n\ntf.flags.DEFINE_integer(\n 'image_width', 224, 'Width of each input images.')\n\ntf.flags.DEFINE_integer(\n 'image_height', 224, 'Height of each input images.')\n\ntf.flags.DEFINE_integer(\n 'batch_size', 16, 'How many images process at one time.')\n\ntf.flags.DEFINE_float(\n 'momentum', 0.0, 'Momentum.')\n\nFLAGS = tf.flags.FLAGS\n\ndef gkern(kernlen=21, nsig=3):\n \"\"\"Returns a 2D Gaussian kernel array.\"\"\"\n import scipy.stats as st\n\n x = np.linspace(-nsig, nsig, kernlen)\n kern1d = st.norm.pdf(x)\n kernel_raw = np.outer(kern1d, kern1d)\n kernel = kernel_raw / kernel_raw.sum()\n return kernel\n\nkernel = gkern(15, 3).astype(np.float32)\nstack_kernel = np.stack([kernel, kernel, kernel]).swapaxes(2, 0)\nstack_kernel = np.expand_dims(stack_kernel, 3)\n\n\n\ndef load_images(input_dir, batch_shape):\n \"\"\"Read png images from input directory in batches.\n\n Args:\n input_dir: input directory\n batch_shape: shape of minibatch array, i.e. [batch_size, height, width, 3]\n\n Yields:\n filenames: list file names without path of each image\n Lenght of this list could be less than batch_size, in this case only\n first few images of the result are elements of the minibatch.\n images: array with all images from this batch\n \"\"\"\n images = np.zeros(batch_shape)\n filenames = []\n idx = 0\n batch_size = batch_shape[0]\n for filepath in tf.gfile.Glob(os.path.join(input_dir, '*.png')):\n with tf.gfile.Open(filepath) as f:\n image = imread(f, mode='RGB')\n image = imresize(image, [FLAGS.image_height, FLAGS.image_width]).astype(np.float)\n image[:,:,0] -= _R_MEAN\n image[:,:,1] -= _G_MEAN\n image[:,:,2] -= _B_MEAN\n images[idx, :, :, :] = image\n filenames.append(os.path.basename(filepath))\n idx += 1\n if idx == batch_size:\n yield filenames, images\n filenames = []\n images = np.zeros(batch_shape)\n idx = 0\n if idx > 0:\n yield filenames, images\n\n\ndef save_images(images, filenames, output_dir):\n \"\"\"Saves images to the output directory.\n\n Args:\n images: array with minibatch of images\n filenames: list of filenames without path\n If number of file names in this list less than number of images in\n the minibatch then only first len(filenames) images will be saved.\n output_dir: directory where to save images\n \"\"\"\n for i, filename in enumerate(filenames):\n # Images for inception classifier are normalized to be in [-1, 1] interval,\n # so rescale them back to [0, 1].\n with tf.gfile.Open(os.path.join(output_dir, filename), 'w') as f:\n image = images[i, :, :, :]\n image[:,:,0] += _R_MEAN\n image[:,:,1] += _G_MEAN\n image[:,:,2] += _B_MEAN\n image = imresize(image, [299, 299])\n imsave(f, image, format='png')\n\n\ndef graph(x, y, i, x_max, x_min, grad):\n eps = FLAGS.max_epsilon\n num_iter = FLAGS.num_iter\n alpha = eps / num_iter\n momentum = FLAGS.momentum\n num_classes = 1000\n\n with slim.arg_scope(vgg.vgg_arg_scope()):\n logits, end_points = vgg.vgg_16(\n x, num_classes=num_classes, is_training=False)\n \n pred = tf.argmax(logits, 1)\n\n first_round = tf.cast(tf.equal(i, 0), tf.int64)\n y = first_round * pred + (1 - first_round) * y\n one_hot = tf.one_hot(y, num_classes)\n\n cross_entropy = tf.losses.softmax_cross_entropy(one_hot,\n logits,\n label_smoothing=0.0,\n weights=1.0)\n noise = tf.gradients(cross_entropy, x)[0]\n noise = tf.nn.depthwise_conv2d(noise, stack_kernel, strides=[1, 1, 1, 1], padding='SAME')\n noise = noise / tf.reduce_mean(tf.abs(noise), [1,2,3], keep_dims=True)\n noise = momentum * grad + noise\n x = x + alpha * tf.sign(noise)\n x = tf.clip_by_value(x, x_min, x_max)\n i = tf.add(i, 1)\n return x, y, i, x_max, x_min, noise\n\ndef stop(x, y, i, x_max, x_min, grad):\n num_iter = FLAGS.num_iter\n return tf.less(i, num_iter)\n\n\ndef main(_):\n # Images for inception classifier are normalized to be in [-1, 1] interval,\n # eps is a difference between pixels so it should be in [0, 2] interval.\n # Renormalizing epsilon from [0, 255] to [0, 2].\n start_time = time.time()\n\n eps = FLAGS.max_epsilon\n\n batch_shape = [FLAGS.batch_size, FLAGS.image_height, FLAGS.image_width, 3]\n\n tf.logging.set_verbosity(tf.logging.INFO)\n\n with tf.Graph().as_default():\n # Prepare graph\n x_input = tf.placeholder(tf.float32, shape=batch_shape)\n x_max = x_input + eps\n x_min = x_input - eps\n\n y = tf.constant(np.zeros([FLAGS.batch_size]), tf.int64)\n i = tf.constant(0)\n grad = tf.zeros(shape=batch_shape)\n x_adv, _, _, _, _, _ = tf.while_loop(stop, graph, [x_input, y, i, x_max, x_min, grad])\n\n # Run computation\n s1 = tf.train.Saver(slim.get_model_variables())\n\n print('Building Graph Done', time.time() - start_time)\n\n with tf.Session() as sess:\n s1.restore(sess, FLAGS.checkpoint_path_vgg)\n\n print('Load Parameters Done', time.time() - start_time)\n tot_images = 0\n\n for filenames, images in load_images(FLAGS.input_dir, batch_shape):\n adv_images = sess.run(x_adv, feed_dict={x_input: images})\n save_images(adv_images, filenames, FLAGS.output_dir)\n\n tot_images += len(filenames)\n print(tot_images, time.time() - start_time)\n\n print(time.time() - start_time)\n\n\nif __name__ == '__main__':\n tf.app.run()\n"
},
{
"path": "dataset/README.md",
"content": "\n# Dataset for adversarial competition.\n\nTwo datasets will be used for the competition:\n\n* **DEV** dataset which is available here for development and experimenting.\n* **TEST** dataset which will be kept secret until after the competition\n and will be used for final scoring.\n\nBoth datasets are composed from publicly available images which were posted\nonline under CC-BY license.\n\n## Dataset format\n\nDEV dataset is defined by `dev_dataset.csv`\nwhich contains URLs of the images along with bounding boxes\nand classification labels.\n\n`dev_dataset.csv` is a table in\n[CSV](https://en.wikipedia.org/wiki/Comma-separated_values)\nformat with the following columns:\n\n* **ImageId** - id of the image.\n* **URL** - URL of the image.\n* **x1**, **y1**, **x2**, **y2** - bounding box of the area of interest in\n the image. Bounding box is relative, which means that all coordinates are\n between 0 and 1.\n* **TrueLabel** - true label of the image.\n* **TargetClass** - label for targeted adversarial attack.\n* **OriginalLandingURL** - original landing page where this image was found.\n* **License** - licence under which image was distributed by author.\n* **Author** - author of the image.\n* **AuthorProfileURL** - URL of the author's profile.\n\nDataset is labelled with\n[ImageNet](http://www.image-net.org/challenges/LSVRC/2012/) labels.\nSpecific values of labels are compatible with pre-trained Inception models,\nwhich are available as a part of\n[TF-Slim](https://github.com/tensorflow/models/tree/master/slim).\nIn particular pre-trained Inception v3 and InceptionResnet v2 could be used\nto classify dataset with high accuracy.\n\n## Downloading images\n\n`dev_dataset.csv` contains only URLs of the images.\nActual images have to be downloaded before being used for experiments.\n\n`download_images.py` is a Python program which downloads images for all\nrecords in `dev_dataset.csv`. Usage:\n\n```\n# Replace CSV_FILE with path to dev_dataset.csv\nCSV_FILE=dev_dataset.csv\n# Replace OUTPUT_DIR with path to directory where all images should be stored\nOUTPUT_DIR=images\n# Download images\npython download_images.py --input_file=${CSV_FILE} --output_dir=${OUTPUT_DIR}\n```\n\nAll downloaded images will be cropped according to the bounding boxes in\n`dev_dataset.csv` and resized to 299x299 pixels.\nEach image will be saved in PNG format with filename `IMAGE_ID.png`\nwhere `IMAGE_ID` is the id of the image from `dev_dataset.csv`.\n"
},
{
"path": "dataset/dev_dataset.csv",
"content": "ImageId,URL,x1,y1,x2,y2,TrueLabel,TargetClass,OriginalLandingURL,License,Author,AuthorProfileURL\r\n0c7ac4a8c9dfa802,https://c1.staticflickr.com/9/8540/28821627444_0524012bdd_o.jpg,0,0,0.871838111,1,306,779,https://www.flickr.com/photos/gails_pictures/28821627444,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\nf43fbfe8a9ea876c,https://c1.staticflickr.com/9/8066/28892033183_6f675dcc03_o.jpg,0.25,0,1,0.599757674,884,378,https://www.flickr.com/photos/barty/28892033183,https://creativecommons.org/licenses/by/2.0/,Barry Badcock,https://www.flickr.com/people/barty/\r\n4fc263d35a3ad3ee,https://c1.staticflickr.com/8/7378/27465801596_a9dd11e5e2_o.jpg,0.333333333,0,1,1,244,123,https://www.flickr.com/photos/foxcroftacademy/27465801596,https://creativecommons.org/licenses/by/2.0/,Foxcroft Academy,https://www.flickr.com/people/foxcroftacademy/\r\ncc13c2bc5cdd1f44,https://c1.staticflickr.com/9/8864/28546467522_56229f2bef_o.jpg,0,0,0.5,0.75,560,741,https://www.flickr.com/photos/o_0/28546467522/,https://creativecommons.org/licenses/by/2.0/,Guilhem Vellut,https://www.flickr.com/people/o_0/\r\n73a52afd2f818ed5,https://c1.staticflickr.com/6/5607/31066602702_382b13646e_o.jpg,0.489195231,0,1,0.75,439,696,https://www.flickr.com/photos/chemiebw/31066602702/,https://creativecommons.org/licenses/by/2.0/,Chemie-Verb\\303\\244nde Baden-W\\303\\274rttemberg,https://www.flickr.com/people/chemiebw/\r\n58f0fd17c4a0e25a,https://c1.staticflickr.com/9/8262/29250758112_3147698dd2_o.jpg,0,0,0.5625,0.75,991,100,https://www.flickr.com/photos/zongo/29250758112,https://creativecommons.org/licenses/by/2.0/,DAVID HOLT,https://www.flickr.com/people/zongo/\r\n90e11aa7c36c64f2,https://c1.staticflickr.com/8/7528/26850127330_56022d63f7_o.jpg,0,0.25,0.5625,1,950,1,https://www.flickr.com/photos/cogdog/26850127330,https://creativecommons.org/licenses/by/2.0/,Alan Levine,https://www.flickr.com/people/cogdog/\r\n696f0f6bea562bf8,https://c1.staticflickr.com/6/5605/30947139580_468ba7e513_o.jpg,0,0,0.5,0.75,854,267,https://www.flickr.com/photos/jamesbradley/30947139580/,https://creativecommons.org/licenses/by/2.0/,Fr James Bradley,https://www.flickr.com/people/jamesbradley/\r\ndf58f94361c6d105,https://c1.staticflickr.com/8/7248/27047266920_8363816754_o.jpg,0.200627477,0,0.601882431,0.75,610,13,https://www.flickr.com/photos/roberthaandrikman/27047266920,https://creativecommons.org/licenses/by/2.0/,Robert Haandrikman,https://www.flickr.com/people/roberthaandrikman/\r\n1394faa319bd353c,https://c1.staticflickr.com/1/542/31667350163_b6906e0d48_o.jpg,0.333333333,0,1,1,610,507,https://www.flickr.com/photos/landrovermena/31667350163,https://creativecommons.org/licenses/by/2.0/,Land Rover MENA,https://www.flickr.com/people/landrovermena/\r\n6cae4a23623d142b,https://c1.staticflickr.com/8/7404/27305101801_5b422e51fe_o.jpg,0.28125,0,0.84375,0.75,583,440,https://flickr.com/39309480@N05/27305101801,https://creativecommons.org/licenses/by/2.0/,Embajada de EEUU en la Argentina,https://www.flickr.com/people/embajadaeeuubuenosaires/\r\nd02adcb9140880a1,https://c1.staticflickr.com/1/404/31746323273_467fb30873_o.jpg,0,0,0.714503817,1,916,783,https://www.flickr.com/photos/ruthanddave/31746323273/,https://creativecommons.org/licenses/by/2.0/,Ruth Hartnup,https://www.flickr.com/people/ruthanddave/\r\n71080285ad6cb4e9,https://c1.staticflickr.com/9/8444/29355103331_45d4b77595_o.jpg,0,0.25,0.75,1,456,478,https://www.flickr.com/photos/themarmot/29355103331,https://creativecommons.org/licenses/by/2.0/,The Marmot,https://www.flickr.com/people/themarmot/\r\n137ab6ca314e9e35,https://c1.staticflickr.com/9/8681/28102039640_11703a03f5_o.jpg,0,0.311985173,0.75,0.935955519,620,556,https://www.flickr.com/photos/130468093@N03/28102039640,https://creativecommons.org/licenses/by/2.0/,~ladyyoruichi~,https://www.flickr.com/people/130468093@N03/\r\nc819099fed6f0d73,https://c1.staticflickr.com/1/126/31638424475_3b36f926b1_o.jpg,0.329912023,0,0.98973607,1,962,777,https://www.flickr.com/photos/djackmanson/31638424475,https://creativecommons.org/licenses/by/2.0/,David Jackmanson,https://www.flickr.com/people/djackmanson/\r\n4b384c8247b56c0a,https://c1.staticflickr.com/9/8526/28626152335_955398ef80_o.jpg,0,0.25,0.5,1,542,925,https://www.flickr.com/photos/virtualpainter/28626152335,https://creativecommons.org/licenses/by/2.0/,Taka Umemura,https://www.flickr.com/people/virtualpainter/\r\n62ebd5f7ce015380,https://c1.staticflickr.com/9/8832/28846755931_9e77600319_o.jpg,0,0,0.66226138,1,631,692,https://www.flickr.com/photos/spera-designerschuhe/28846755931,https://creativecommons.org/licenses/by/2.0/,\"SPERA.de Designerschuhe, Taschen und Accessoires\",https://www.flickr.com/people/spera-designerschuhe/\r\nf24817d66024dfb6,https://c1.staticflickr.com/8/7286/28064769996_6bf78b5810_o.jpg,0,0,0.5,0.75,742,740,https://www.flickr.com/photos/jdlasica/28064769996,https://creativecommons.org/licenses/by/2.0/,JD Lasica,https://www.flickr.com/people/jdlasica/\r\n1b57d7bbd673c771,https://c1.staticflickr.com/9/8024/29035810873_9be68318c4_o.jpg,0.5,0.25,1,1,423,927,https://www.flickr.com/photos/fotosagenciabrasil/29035810873/,https://creativecommons.org/licenses/by/2.0/,Ag\\303\\252ncia Brasil Fotografias,https://www.flickr.com/people/fotosagenciabrasil/\r\na2de1f4f2c1021fa,https://c1.staticflickr.com/9/8598/29005552090_ee38f841ac_o.jpg,0.25,0.25,1,1,456,185,https://www.flickr.com/photos/sillygwailo/29005552090/,https://creativecommons.org/licenses/by/2.0/,Richard Eriksson,https://www.flickr.com/people/sillygwailo/\r\n3e7b01ba495f15f9,https://c1.staticflickr.com/8/7366/26897189910_049b6377e0_o.jpg,0,0,0.75,0.601503759,708,410,https://www.flickr.com/photos/darioalvarez/26897189910,https://creativecommons.org/licenses/by/2.0/,Dario Alvarez,https://www.flickr.com/people/darioalvarez/\r\n1a264165276e3c85,https://c1.staticflickr.com/8/7327/27500591886_a1b257229e_o.jpg,0.5,0,1,0.75,855,759,https://www.flickr.com/photos/presidentialoffice/27500591886,https://creativecommons.org/licenses/by/2.0/,\\347\\270\\275\\347\\265\\261\\345\\272\\234,https://www.flickr.com/people/presidentialoffice/\r\nbc9a5e01c02e759e,https://c1.staticflickr.com/8/7575/27719847416_e231168ac1_o.jpg,0,0.28125,0.75,0.84375,923,18,https://www.flickr.com/photos/jimg944/27719847416/,https://creativecommons.org/licenses/by/2.0/,Jim G,https://www.flickr.com/people/jimg944/\r\n01bdc0cfe670f708,https://c1.staticflickr.com/1/457/31837933011_3e57f56dc9_o.jpg,0,0,0.5,0.75,472,541,https://www.flickr.com/photos/quinet/31837933011,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n6af829a442e2e0e7,https://c1.staticflickr.com/6/5827/30663702381_910aca914a_o.jpg,0.25,0,1,1,130,231,https://www.flickr.com/photos/rusty_clark/30663702381,https://creativecommons.org/licenses/by/2.0/,Rusty Clark ~ 100K Photos,https://www.flickr.com/people/rusty_clark/\r\n493694833dc23399,https://c1.staticflickr.com/1/691/33087466756_27fcc6e909_o.jpg,0.25,0.25,0.75,1,538,835,https://www.flickr.com/photos/denalinps/33087466756/,https://creativecommons.org/licenses/by/2.0/,Denali National Park and Preserve,https://www.flickr.com/people/denalinps/\r\n6ac94c9244f84aa3,https://c1.staticflickr.com/9/8887/27621832723_abe5df1bf6_o.jpg,0.28125,0,0.84375,0.75,673,336,https://www.flickr.com/photos/sedoglia/27621832723,https://creativecommons.org/licenses/by/2.0/,Sergio,https://www.flickr.com/people/sedoglia/\r\n7529b42e71769777,https://c1.staticflickr.com/6/5573/30447660301_21f03bc832_o.jpg,0.28125,0,0.84375,0.75,477,618,https://www.flickr.com/photos/archiefalkmaar/30447660301/,https://creativecommons.org/licenses/by/2.0/,Regionaal Archief Alkmaar,https://www.flickr.com/people/archiefalkmaar/\r\nd204464fcf15ad99,https://c1.staticflickr.com/9/8735/28293562656_b6754b6f76_o.jpg,0.25,0.249079529,1,0.747238586,300,269,https://www.flickr.com/photos/zigazou76/28293562656,https://creativecommons.org/licenses/by/2.0/,Fr\\303\\251d\\303\\251ric BISSON,https://www.flickr.com/people/zigazou76/\r\n7fd7ce3eb2bf9944,https://c1.staticflickr.com/8/7060/26682014410_e1ae9f3f68_o.jpg,0.25,0,1,1,100,113,https://www.flickr.com/photos/starr-environmental/26682014410/,https://creativecommons.org/licenses/by/2.0/,Forest and Kim Starr,https://www.flickr.com/people/starr-environmental/\r\n303781188f1b3b27,https://c1.staticflickr.com/8/7196/27632686515_7c9e105ab4_o.jpg,0.578252033,0,1,0.75,477,339,https://www.flickr.com/photos/tomhilton/27632686515,https://creativecommons.org/licenses/by/2.0/,Tom Hilton,https://www.flickr.com/people/tomhilton/\r\n67967c9db11151e0,https://c1.staticflickr.com/6/5699/30482545566_9bb69e3a2c_o.jpg,0.5,0.25,1,1,252,820,https://www.flickr.com/photos/blumenbiene/30482545566,https://creativecommons.org/licenses/by/2.0/,Maja Dumat,https://www.flickr.com/people/blumenbiene/\r\nfa2a7d600b62174e,https://c1.staticflickr.com/9/8135/29796651083_af3421402e_o.jpg,0.25,0,1,0.729946524,521,572,https://www.flickr.com/photos/familymwr/29796651083,https://creativecommons.org/licenses/by/2.0/,U.S. Army,https://www.flickr.com/people/familymwr/\r\ncd2d2889ef7ac51c,https://c1.staticflickr.com/8/7108/27445698401_622147f593_o.jpg,0,0,0.666666667,1,924,607,https://www.flickr.com/photos/janitors/27445698401,https://creativecommons.org/licenses/by/2.0/,K\\304\\201rlis Dambr\\304\\201ns,https://www.flickr.com/people/janitors/\r\n11954d93a7051cb5,https://c1.staticflickr.com/9/8598/30096731475_f2050e80f0_o.jpg,0.5,0,1,0.75,23,503,https://www.flickr.com/photos/andymorffew/30096731475,https://creativecommons.org/licenses/by/2.0/,Andy Morffew,https://www.flickr.com/people/andymorffew/\r\n03f90f7138f761e8,https://c1.staticflickr.com/8/7358/27496838730_a300e17377_o.jpg,0.25,0.25,1,1,664,490,https://www.flickr.com/photos/pasa/27496838730,https://creativecommons.org/licenses/by/2.0/,Paul Sableman,https://www.flickr.com/people/pasa/\r\n19935b1c0a4d305d,https://c1.staticflickr.com/8/7560/28230694205_599d4915f1_o.jpg,0.304216867,0,1,1,761,642,https://www.flickr.com/photos/lge/28230694205,https://creativecommons.org/licenses/by/2.0/,LG\\354\\240\\204\\354\\236\\220,https://www.flickr.com/people/lge/\r\n9cf861a657c27cec,https://c1.staticflickr.com/1/737/31554472305_631b37cd13_o.jpg,0.28125,0.25,0.84375,1,583,897,https://www.flickr.com/photos/rusty_clark/31554472305/,https://creativecommons.org/licenses/by/2.0/,Rusty Clark ~ 100K Photos,https://www.flickr.com/people/rusty_clark/\r\n05817a675e7506d1,https://c1.staticflickr.com/8/7530/27222000794_49d44c7533_o.jpg,0.246522694,0,0.739568082,0.75,673,754,https://www.flickr.com/photos/trocaire/27222000794,https://creativecommons.org/licenses/by/2.0/,Trocaire,https://www.flickr.com/people/trocaire/\r\ne1c702ed9842dd1c,https://c1.staticflickr.com/8/7538/26750212822_874e61f2f5_o.jpg,0,0.25,0.75,1,693,757,https://www.flickr.com/photos/mcleod/26750212822,https://creativecommons.org/licenses/by/2.0/,Scott McLeod,https://www.flickr.com/people/mcleod/\r\ne7cafac174e90143,https://c1.staticflickr.com/8/7064/26911748774_e91c741c13_o.jpg,0.25,0.25,0.75,1,526,116,https://www.flickr.com/photos/tabor-roeder/26911748774,https://creativecommons.org/licenses/by/2.0/,Phil Roeder,https://www.flickr.com/people/tabor-roeder/\r\n1a680f33500acb4b,https://c1.staticflickr.com/6/5623/30353092243_549ce05481_o.jpg,0,0,0.5625,0.75,932,73,https://www.flickr.com/photos/londonmatt/30353092243,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\n81ddbe14219e9dcf,https://c1.staticflickr.com/1/599/32603915941_1f3f02a21f_o.jpg,0.436619718,0,1,0.75,318,672,http://www.flickr.com/photos/treegrow/32603915941/,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\n04cb9de291399d15,https://c1.staticflickr.com/9/8671/28725010671_e060588a1b_o.jpg,0.25,0.25,0.75,1,465,451,https://www.flickr.com/photos/photo-digest/28725010671,https://creativecommons.org/licenses/by/2.0/,media.digest,https://www.flickr.com/people/photo-digest/\r\n23f9d18d531eb148,https://c1.staticflickr.com/8/7545/28289771271_d972c53c03_o.jpg,0.249079529,0,0.747238586,0.75,479,851,https://www.flickr.com/photos/londonmatt/28289771271/,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\n63f2e14ab6be82cf,https://c1.staticflickr.com/8/7360/27324993991_9e7547e472_o.jpg,0.330871492,0,1,1,668,694,https://www.flickr.com/photos/mdgovpics/27324993991,https://creativecommons.org/licenses/by/2.0/,Maryland GovPics,https://www.flickr.com/people/mdgovpics/\r\n0dffbfa7ba43cb90,https://c1.staticflickr.com/1/599/31842595896_6c327628ee_o.jpg,0.28125,0,0.84375,0.75,962,95,https://flickr.com/10581108@N00/31842595896,https://creativecommons.org/licenses/by/2.0/,jencu,https://www.flickr.com/people/jennycu/\r\na97274c7e80a1764,https://c1.staticflickr.com/9/8770/28603315541_10de82290f_o.jpg,0,0.25,0.5625,1,866,926,https://flickr.com/10069023@N00/28603315541,https://creativecommons.org/licenses/by/2.0/,Charles Kremenak,https://www.flickr.com/people/charleskremenak/\r\n85e2eea48d912d31,https://c1.staticflickr.com/6/5728/30827582556_27a03b94b4_o.jpg,0.501840943,0.25,1,1,806,136,https://www.flickr.com/photos/eaglebrook/30827582556/,https://creativecommons.org/licenses/by/2.0/,EaglebrookSchool,https://www.flickr.com/people/eaglebrook/\r\nff5182599937c1ec,https://c1.staticflickr.com/9/8088/29255908970_542c7fcc99_o.jpg,0.25,0,1,0.75,325,210,https://www.flickr.com/photos/orangeaurochs/29255908970/,https://creativecommons.org/licenses/by/2.0/,Orangeaurochs,https://www.flickr.com/people/orangeaurochs/\r\n4bb12984b41d0834,https://c1.staticflickr.com/8/7155/26715421772_322982abb8_o.jpg,0.280784844,0,0.842354533,1,34,8,https://www.flickr.com/photos/lwpkommunikacio/26715421772,https://creativecommons.org/licenses/by/2.0/,Lwp Kommunik\\303\\241ci\\303\\263,https://www.flickr.com/people/lwpkommunikacio/\r\n150249123e81930d,https://c1.staticflickr.com/8/7637/27864748826_3611a80685_o.jpg,0.25,0,1,1,923,485,https://www.flickr.com/photos/36668466@N05/27864748826,https://creativecommons.org/licenses/by/2.0/,Fernando Medina,https://www.flickr.com/people/36668466@N05/\r\nf963957ccfa82f6b,https://c1.staticflickr.com/8/7321/27811646942_6af16e2c20_o.jpg,0.217675378,0,0.653026135,0.75,143,8,https://www.flickr.com/photos/bamyers4az/27811646942,https://creativecommons.org/licenses/by/2.0/,Brent Myers,https://www.flickr.com/people/bamyers4az/\r\n890fdaf4e1892e00,https://c1.staticflickr.com/9/8079/29641887231_ac78f6acb2_o.jpg,0,0,0.757861635,1,313,692,https://www.flickr.com/photos/gails_pictures/29641887231,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n9a1c67f467b9beea,https://c1.staticflickr.com/8/7313/26933714923_dae58c7fa8_o.jpg,0,0.25,0.70913884,1,303,90,https://www.flickr.com/photos/gails_pictures/26933714923,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n217597258ac12258,https://c1.staticflickr.com/8/7326/27290987603_52008d9de0_o.jpg,0.25,0.25,1,1,583,805,https://www.flickr.com/photos/iwashitashoichi/27290987603,https://creativecommons.org/licenses/by/2.0/,Shoichi Iwashita,https://www.flickr.com/people/iwashitashoichi/\r\n346c5575b60194c3,https://c1.staticflickr.com/6/5753/31338370671_675f18f5a9_o.jpg,0.25,0,0.75,0.75,949,625,https://flickr.com/77742560@N06/31338370671,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n296033c3a5910845,https://c1.staticflickr.com/8/7659/27497141411_2fdd591428_o.jpg,0,0,0.675074184,1,361,797,https://www.flickr.com/photos/100915417@N07/27497141411,https://creativecommons.org/licenses/by/2.0/,Charlie Marshall,https://www.flickr.com/people/100915417@N07/\r\n3a7d9bd310202e93,https://c1.staticflickr.com/6/5751/30906474272_60f5035a81_o.jpg,0,0,0.5,0.75,790,660,https://www.flickr.com/photos/sansebastian2016/30906474272,https://creativecommons.org/licenses/by/2.0/,Donostia/San Sebastian 2016,https://www.flickr.com/people/sansebastian2016/\r\n56e9d5f63cfc507e,https://c1.staticflickr.com/6/5767/31254643081_606fefaeef_o.jpg,0,0,1,1,441,624,https://flickr.com/31424626@N05/31254643081,https://creativecommons.org/licenses/by/2.0/,Pulencio,https://www.flickr.com/people/pulencio/\r\n602f2c3749d6021e,https://c1.staticflickr.com/6/5488/30010578363_ae1a2b5fc4_o.jpg,0.25,0,1,0.75,747,106,https://flickr.com/88307459@N08/30010578363,https://creativecommons.org/licenses/by/2.0/,Alessandro,https://www.flickr.com/people/88307459@N08/\r\n7c754696067fdfa9,https://c1.staticflickr.com/8/7322/26952063895_0d0990098b_o.jpg,0.53485064,0,1,0.75,765,469,https://www.flickr.com/photos/zachievenor/26952063895,https://creativecommons.org/licenses/by/2.0/,Zachi Evenor,https://www.flickr.com/people/zachievenor/\r\n8f5eda0b6a910b43,https://c1.staticflickr.com/1/525/31526646220_dbd7c0df37_o.jpg,0.5,0,1,0.75,950,918,https://www.flickr.com/photos/129816626@N04/31526646220/,https://creativecommons.org/licenses/by/2.0/,Elusive Muse,https://www.flickr.com/people/129816626@N04/\r\n89119361e9d9e02d,https://c1.staticflickr.com/9/8064/30107504355_f5a2453191_o.jpg,0,0,0.669128508,1,481,473,http://www.flickr.com/photos/79157069@N03/30107504355,https://creativecommons.org/licenses/by/2.0/,Matthew Paul Argall,https://www.flickr.com/people/79157069@N03/\r\nda63f2acf683c0a7,https://c1.staticflickr.com/3/2626/32878464786_81b7b5f560_o.jpg,0.25,0.25,0.75,1,134,617,https://www.flickr.com/photos/steve_childs/32878464786/,https://creativecommons.org/licenses/by/2.0/,Steve Childs,https://www.flickr.com/people/steve_childs/\r\n741fc1a3fc08731d,https://c1.staticflickr.com/1/333/30878601253_6a06af38ce_o.jpg,0.226157083,0.25,0.678471248,1,793,400,https://www.flickr.com/photos/ufv/30878601253,https://creativecommons.org/licenses/by/2.0/,University of the Fraser Valley,https://www.flickr.com/people/ufv/\r\nd77af00930283f4b,https://c1.staticflickr.com/6/5686/30135548405_290ce24c79_o.jpg,0,0.28125,0.75,0.84375,901,649,https://flickr.com/131810676@N03/30135548405,https://creativecommons.org/licenses/by/2.0/,baka_neko_baka,https://www.flickr.com/people/131810676@N03/\r\n97151fe7a0a6e2e3,https://c1.staticflickr.com/9/8108/28266576044_fe196cd052_o.jpg,0,0,1,1,734,228,https://www.flickr.com/photos/paulstudio/28266576044,https://creativecommons.org/licenses/by/2.0/,Paul Chan,https://www.flickr.com/people/paulstudio/\r\n393d40f3a6a79400,https://c1.staticflickr.com/8/7641/27798795492_12f1e08a40_o.jpg,0,0.25,0.496696035,1,328,846,https://www.flickr.com/photos/shankaronline/27798795492/,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n8f5e0855ca8de8ea,https://c1.staticflickr.com/1/594/31875644290_a507a79ca7_o.jpg,0.4375,0.25,1,1,468,716,https://www.flickr.com/photos/briansolis/31875644290,https://creativecommons.org/licenses/by/2.0/,Brian Solis,https://www.flickr.com/people/briansolis/\r\nd72f34526e12fd83,https://c1.staticflickr.com/9/8787/28375652543_039a76200f_o.jpg,0,0.280664063,0.75,0.841992188,442,325,https://www.flickr.com/photos/cogdog/28375652543,https://creativecommons.org/licenses/by/2.0/,Alan Levine,https://www.flickr.com/people/cogdog/\r\n957529a2abaa0581,https://c1.staticflickr.com/6/5493/31215689672_90fdc524c1_o.jpg,0.33773862,0,1,1,883,648,https://www.flickr.com/photos/yourbestdigs/31215689672/,https://creativecommons.org/licenses/by/2.0/,Your Best Digs,https://www.flickr.com/people/yourbestdigs/\r\n3bd1ba0432603b28,https://c1.staticflickr.com/9/8041/29388312836_4151756518_o.jpg,0,0,0.5625,0.75,921,89,https://www.flickr.com/photos/a_peach/29388312836,https://creativecommons.org/licenses/by/2.0/,A_Peach,https://www.flickr.com/people/a_peach/\r\n3228a465697fb2cf,https://c1.staticflickr.com/1/699/32034140530_62059de5a0_o.jpg,0,0,0.75,0.5625,840,922,https://www.flickr.com/photos/ufv/32034140530,https://creativecommons.org/licenses/by/2.0/,University of the Fraser Valley,https://www.flickr.com/people/ufv/\r\n871c99d5658c3189,https://c1.staticflickr.com/1/593/31013399073_e0c80204ed_o.jpg,0.5,0,1,0.75,956,587,https://flickr.com/64586261@N02/31013399073,https://creativecommons.org/licenses/by/2.0/,Ag\\303\\252ncia Bras\\303\\255lia,https://www.flickr.com/people/agenciabrasilia/\r\n1733cbc68a2ae05d,https://c1.staticflickr.com/1/465/32838931536_2e1b56309c_o.jpg,0,0,0.666666667,1,556,892,https://www.flickr.com/photos/147079914@N03/32838931536/in/pool-44124304756@N01,https://creativecommons.org/licenses/by/2.0/,sponki25,https://www.flickr.com/people/147079914@N03/\r\n69a3493d13953e1f,https://c1.staticflickr.com/1/486/31850894154_77cd0b743d_o.jpg,0.4375,0.25,1,1,520,407,https://www.flickr.com/photos/byzantiumbooks/31850894154,https://creativecommons.org/licenses/by/2.0/,Bill Smith,https://www.flickr.com/people/byzantiumbooks/\r\n8a82ddd7ab76efbb,https://c1.staticflickr.com/9/8073/29273707766_84a5a2321f_o.jpg,0,0,0.663235294,1,511,908,https://www.flickr.com/photos/shankaronline/29273707766,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\ne95ef0aeeab7b496,https://c1.staticflickr.com/3/2892/32904176405_735232107b_o.jpg,0.710762332,0,1,0.75,889,860,https://www.flickr.com/photos/arg_flickr/32904176405/,https://creativecommons.org/licenses/by/2.0/,Andrew,https://www.flickr.com/people/arg_flickr/\r\n8555611998ea426a,https://c1.staticflickr.com/6/5600/30019858070_5049f5d386_o.jpg,0,0,0.75,1,991,497,https://www.flickr.com/photos/eamoncurry/30019858070,https://creativecommons.org/licenses/by/2.0/,Eamon Curry,https://www.flickr.com/people/eamoncurry/\r\n1cc2caf8e5db27c3,https://c1.staticflickr.com/1/598/32069226561_2e29df4dce_o.jpg,0.25,0,1,0.5,431,607,https://flickr.com/21841998@N06/32069226561,https://creativecommons.org/licenses/by/2.0/,The People's Tribune,https://www.flickr.com/people/21841998@N06/\r\na00c8d22e63497c4,https://c1.staticflickr.com/8/7019/26876163625_b5ebe36eaf_o.jpg,0.27818818,0.25,0.834564541,1,431,892,https://www.flickr.com/photos/timmccune/26876163625,https://creativecommons.org/licenses/by/2.0/,Tim McCune,https://www.flickr.com/people/timmccune/\r\n4f553b721705d008,https://c1.staticflickr.com/8/7417/28111852685_716c19bcb5_o.jpg,0,0,0.75,1,397,3,https://www.flickr.com/photos/noaaphotolib/28111852685/in/photolist-hCZ5zh-7RqcE6-7Rtsyb-JQ9Apg/,https://creativecommons.org/licenses/by/2.0/,NOAA Photo Library,https://www.flickr.com/people/noaaphotolib/\r\nf1ce339a44d7a118,https://c1.staticflickr.com/9/8867/28838922156_09d615d6f5_o.jpg,0.333333333,0,1,1,98,702,https://www.flickr.com/photos/132033298@N04/28838922156,https://creativecommons.org/licenses/by/2.0/,C. P. Ewing,https://www.flickr.com/people/132033298@N04/\r\n2153d9869fb5a9ef,https://c1.staticflickr.com/9/8892/28233397700_b175333ccc_o.jpg,0.452546296,0.25,1,1,645,942,https://www.flickr.com/photos/katerha/28233397700,https://creativecommons.org/licenses/by/2.0/,Kate Ter Haar,https://www.flickr.com/people/katerha/\r\n7506caf0a49c3951,https://c1.staticflickr.com/8/7410/27688687271_a42aee521a_o.jpg,0,0,1,0.900181488,79,724,https://www.flickr.com/photos/136941790@N03/27688687271/,https://creativecommons.org/licenses/by/2.0/,laboratorio diagnostica ancona IZSUM,https://www.flickr.com/people/136941790@N03/\r\n1d4133f5bc7af193,https://c1.staticflickr.com/9/8584/28442098602_ed99e31eda_o.jpg,0.259962406,0.25,0.779887218,1,141,911,https://www.flickr.com/photos/25553993@N02/28442098602/,https://creativecommons.org/licenses/by/2.0/,Kev Chapman,https://www.flickr.com/people/25553993@N02/\r\nfd26c4733822dfcc,https://c1.staticflickr.com/9/8649/28250885435_30ea68123c_o.jpg,0.280664063,0.25,0.841992188,1,363,451,https://www.flickr.com/photos/usfwsmtnprairie/28250885435,https://creativecommons.org/licenses/by/2.0/,USFWS Mountain-Prairie,https://www.flickr.com/people/usfwsmtnprairie/\r\n60f5900a00d94b25,https://c1.staticflickr.com/6/5732/31328633496_995b71567c_o.jpg,0.28125,0,0.84375,0.75,706,558,https://www.flickr.com/photos/davidwilson1949/31328633496,https://creativecommons.org/licenses/by/2.0/,David Wilson,https://www.flickr.com/people/davidwilson1949/\r\n79f953be91aa8ce2,https://c1.staticflickr.com/8/7570/27808063061_740fc61f0c_o.jpg,0.28125,0.25,0.84375,1,660,588,https://flickr.com/70724896@N08/27808063061,https://creativecommons.org/licenses/by/2.0/,Betta Tryptophan,https://www.flickr.com/people/nagasakiosada/\r\nd9da4e5ef53d5e8b,https://c1.staticflickr.com/1/537/32179264340_6ea9260e9f_o.jpg,0,0.248826291,1,1,641,487,https://www.flickr.com/photos/ruthanddave/32179264340/,https://creativecommons.org/licenses/by/2.0/,Ruth Hartnup,https://www.flickr.com/people/ruthanddave/\r\n0411ca972e637297,https://c1.staticflickr.com/8/7761/26268791474_72626e9e6a_o.jpg,0.25,0.25,0.75,1,968,697,https://www.flickr.com/photos/nanpalmero/26268791474,https://creativecommons.org/licenses/by/2.0/,Nan Palmero,https://www.flickr.com/people/nanpalmero/\r\n92dbb18dd447fd75,https://c1.staticflickr.com/8/7127/27711989146_46b6ca79f6_o.jpg,0,0.25,0.563380282,1,490,194,https://www.flickr.com/photos/southbeachcars/27711989146,https://creativecommons.org/licenses/by/2.0/,Phillip Pessar,https://www.flickr.com/people/southbeachcars/\r\n65251969959ce0e6,https://c1.staticflickr.com/8/7623/28435254025_3c40331a75_o.jpg,0,0,1,1,938,837,https://www.flickr.com/photos/mcleod/28435254025,https://creativecommons.org/licenses/by/2.0/,Scott McLeod,https://www.flickr.com/people/mcleod/\r\nb76e552d4b3ec0c3,https://c1.staticflickr.com/6/5662/30119560842_26f15fba19_o.jpg,0.462155963,0,1,0.75,992,191,https://www.flickr.com/photos/127338544@N06/30119560842,https://creativecommons.org/licenses/by/2.0/,Sarah Ward,https://www.flickr.com/people/127338544@N06/\r\n828ade8df2d14c45,https://c1.staticflickr.com/6/5649/30456484724_643e11a460_o.jpg,0.5,0.25,1,1,888,988,https://www.flickr.com/photos/johnragai/30456484724,https://creativecommons.org/licenses/by/2.0/,John Ragai,https://www.flickr.com/people/johnragai/\r\n3dc0e8f854bdc9b3,https://c1.staticflickr.com/6/5682/31432432235_54a615c355_o.jpg,0.25,0.25,0.75,1,604,921,https://www.flickr.com/photos/lwpkommunikacio/31432432235,https://creativecommons.org/licenses/by/2.0/,Lwp Kommunik\\303\\241ci\\303\\263,https://www.flickr.com/people/lwpkommunikacio/\r\n6bb6cf032c1286c5,https://c1.staticflickr.com/8/7172/26559644113_e1c454ef1c_o.jpg,0.25,0.25,1,0.75,468,367,https://www.flickr.com/photos/quinet/26559644113,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n2e9b377877f64b3d,https://c1.staticflickr.com/9/8503/29411674165_57521acbf9_o.jpg,0,0.25,0.75,1,499,706,https://www.flickr.com/photos/clairity/29411674165/,https://creativecommons.org/licenses/by/2.0/,Sharon Mollerus,https://www.flickr.com/people/clairity/\r\n2842a69bf5b14526,https://c1.staticflickr.com/9/8612/28790928395_77327e1c83_o.jpg,0.267530488,0.25,0.802591463,1,880,757,https://www.flickr.com/photos/ruthanddave/28790928395/,https://creativecommons.org/licenses/by/2.0/,Ruth Hartnup,https://www.flickr.com/people/ruthanddave/\r\na74f197288b10a0d,https://c1.staticflickr.com/9/8532/29057660090_67701b5197_o.jpg,0.25,0,0.75,0.75,808,748,https://www.flickr.com/photos/usacehq/29057660090,https://creativecommons.org/licenses/by/2.0/,U.S. Army Corps of Engineers,https://www.flickr.com/people/usacehq/\r\n7a85f42383ec2f81,https://c1.staticflickr.com/9/8604/29022121774_c82af166b6_o.jpg,0,0.25,0.75,0.75,709,802,https://www.flickr.com/photos/greg_men/29022121774/,https://creativecommons.org/licenses/by/2.0/,Greg_Men,https://www.flickr.com/people/greg_men/\r\n55e85d7b58024966,https://c1.staticflickr.com/9/8611/28176653474_ef1665d13b_o.jpg,0,0,1,1,968,615,https://www.flickr.com/photos/ultrakml/28176653474,https://creativecommons.org/licenses/by/2.0/,Katherine Lim,https://www.flickr.com/people/ultrakml/\r\nd6d192c5786f29f5,https://c1.staticflickr.com/9/8094/29208873373_f929c21796_o.jpg,0.392682927,0.25,1,1,473,830,https://www.flickr.com/photos/vastateparksstaff/29208873373,https://creativecommons.org/licenses/by/2.0/,Virginia State Parks,https://www.flickr.com/people/vastateparksstaff/\r\nf533740a8d4618ce,https://c1.staticflickr.com/1/671/32497563865_bdd560906f_o.jpg,0,0,0.5,0.75,288,563,https://www.flickr.com/photos/30478819@N08/32497563865,https://creativecommons.org/licenses/by/2.0/,Marco Verch,https://www.flickr.com/people/30478819@N08/\r\n5292f5b36a0789fa,https://c1.staticflickr.com/9/8761/28372171573_06dbf9a8d4_o.jpg,0.249079529,0,0.747238586,0.75,854,795,https://www.flickr.com/photos/63363807@N03/28372171573,https://creativecommons.org/licenses/by/2.0/,Didit Putra,https://www.flickr.com/people/63363807@N03/\r\nba4b978112a20be1,https://c1.staticflickr.com/9/8245/29288650164_e95f3a9c80_o.jpg,0,0.25,0.5,1,972,16,https://www.flickr.com/photos/pahudson/29288650164/,https://creativecommons.org/licenses/by/2.0/,Paul Hudson,https://www.flickr.com/people/pahudson/\r\n3692f6967c632fee,https://c1.staticflickr.com/9/8729/28628001341_0ff5c44db3_o.jpg,0.499263623,0,0.998527246,0.75,806,569,https://www.flickr.com/photos/agenciabrasilia/28628001341,https://creativecommons.org/licenses/by/2.0/,Ag\\303\\252ncia Bras\\303\\255lia,https://www.flickr.com/people/agenciabrasilia/\r\nfae551840c1690fa,https://c1.staticflickr.com/9/8166/29643263176_aa56d7230d_o.jpg,0.333333333,0,1,1,720,424,https://www.flickr.com/photos/rusty_clark/29643263176,https://creativecommons.org/licenses/by/2.0/,Rusty Clark ~ 100K Photos,https://www.flickr.com/people/rusty_clark/\r\n8020698d79dc39f2,https://c1.staticflickr.com/3/2823/33351553831_7498912a81_o.jpg,0.25,0.25,0.75,1,855,954,https://www.flickr.com/photos/thejointstaff/33351553831/,https://creativecommons.org/licenses/by/2.0/,Chairman of the Joint Chiefs of Staff,https://www.flickr.com/people/thejointstaff/\r\nd392f0639c17f396,https://c1.staticflickr.com/8/7448/27262568354_6e93ef7b68_o.jpg,0.500736377,0.25,1,1,808,590,https://www.flickr.com/photos/inra_dist/27262568354/,https://creativecommons.org/licenses/by/2.0/,INRA DIST,https://www.flickr.com/people/inra_dist/\r\n8f97e3fface13e4d,https://c1.staticflickr.com/8/7108/27051340785_a3c9ab8b6d_o.jpg,0.333333333,0,1,1,472,548,https://www.flickr.com/photos/quinet/27051340785,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\na920691cf2d50f79,https://c1.staticflickr.com/9/8512/29944466315_4dabc1230d_o.jpg,0.25,0,0.75,0.75,947,952,https://www.flickr.com/photos/pellesten/29944466315/,https://creativecommons.org/licenses/by/2.0/,Pelle Sten,https://www.flickr.com/people/pellesten/\r\na482985ca1e77960,https://c1.staticflickr.com/9/8751/28431687365_edb15100b0_o.jpg,0,0.25,0.601941748,1,891,470,https://www.flickr.com/photos/janitors/28431687365,https://creativecommons.org/licenses/by/2.0/,K\\304\\201rlis Dambr\\304\\201ns,https://www.flickr.com/people/janitors/\r\n362e4ac62cf888f4,https://c1.staticflickr.com/1/531/32477078836_371e833f56_o.jpg,0,0,1,1,573,36,https://www.flickr.com/photos/freedomiiphotography/32477078836,https://creativecommons.org/licenses/by/2.0/,Freedom II Andres,https://www.flickr.com/people/freedomiiphotography/\r\n694e7617a4878041,https://c1.staticflickr.com/8/7087/27847654031_e4003bcdca_o.jpg,0,0,1,0.827586207,884,720,https://flickr.com/41383869@N07/27847654031,https://creativecommons.org/licenses/by/2.0/,Nathan Hughes Hamilton,https://www.flickr.com/people/nat507/\r\nfa5484e61e455eb2,https://c1.staticflickr.com/1/141/30799508623_1c015153fb_o.jpg,0.333333333,0,1,1,477,769,https://www.flickr.com/photos/scubabix/30799508623,https://creativecommons.org/licenses/by/2.0/,Rob Bixby,https://www.flickr.com/people/scubabix/\r\nab42aa5e9745c214,https://c1.staticflickr.com/1/439/30759196054_5d7ae31471_o.jpg,0.524100719,0,1,0.75,582,591,https://www.flickr.com/photos/andreboeni/30759196054,https://creativecommons.org/licenses/by/2.0/,Andrew Bone,https://www.flickr.com/people/andreboeni/\r\ne6e677a28e3bd6d7,https://c1.staticflickr.com/9/8774/28104802823_5748e74f68_o.jpg,0.334315169,0,1,1,604,160,https://www.flickr.com/photos/sierragoddess/28104802823,https://creativecommons.org/licenses/by/2.0/,Darla \\330\\257\\330\\247\\330\\261\\331\\204\\330\\247 Hueske,https://www.flickr.com/people/sierragoddess/\r\n2e541180b2323a5d,https://c1.staticflickr.com/6/5538/29704744484_7fee6ebbc3_o.jpg,0,0,0.75,0.75,24,905,https://www.flickr.com/photos/sarahmstewart/29704744484,https://creativecommons.org/licenses/by/2.0/,Sarah Stewart,https://www.flickr.com/people/sarahmstewart/\r\n01f824264783f58d,https://c1.staticflickr.com/6/5734/30178450721_1b2b4033a5_o.jpg,0,0,0.75,0.75,968,678,https://www.flickr.com/photos/naotakem/30178450721,https://creativecommons.org/licenses/by/2.0/,Naotake Murayama,https://www.flickr.com/people/naotakem/\r\n335e604bbe3eec78,https://c1.staticflickr.com/6/5615/30719295680_945ac30a0b_o.jpg,0.5,0,1,0.75,312,687,http://www.flickr.com/photos/52450054@N04/30719295680/,https://creativecommons.org/licenses/by/2.0/,Judy Gallagher,https://www.flickr.com/people/52450054@N04/\r\n4e1bdcb677f21fb4,https://c1.staticflickr.com/8/7646/27898036952_dc65e0387a_o.jpg,0.5,0,1,0.75,874,779,https://www.flickr.com/photos/64084569@N02/27898036952,https://creativecommons.org/licenses/by/2.0/,Singa Hitam,https://www.flickr.com/people/64084569@N02/\r\nc6b5f6c1d79dffc6,https://c1.staticflickr.com/9/8472/28615394904_41ea2d98ce_o.jpg,0,0.25,0.5,1,583,652,https://flickr.com/44440670@N04/28615394904,https://creativecommons.org/licenses/by/2.0/,Bilboko Konpartsak,https://www.flickr.com/people/bilbokokonpartsak/\r\n8f9c1e4fd9f4cb2e,https://c1.staticflickr.com/6/5511/30235489014_5fc7ff19a1_o.jpg,0.27818818,0,0.834564541,0.75,17,498,https://www.flickr.com/photos/taylar/30235489014,https://creativecommons.org/licenses/by/2.0/,Ingrid Taylar,https://www.flickr.com/people/taylar/\r\nf54616f2af588fd8,https://c1.staticflickr.com/9/8190/28571086580_2995522f6a_o.jpg,0.5,0.25,1,1,673,113,https://www.flickr.com/photos/41406729@N05/28571086580/,https://creativecommons.org/licenses/by/2.0/,Ben Fitzgerald-O'Connor,https://www.flickr.com/people/onega/\r\nf20911d36a287dce,https://c1.staticflickr.com/8/7010/26825773645_bb695521b3_o.jpg,0.249631811,0.25,0.748895434,1,781,492,https://www.flickr.com/photos/glynlowe/26825773645,https://creativecommons.org/licenses/by/2.0/,www.GlynLowe.com,https://www.flickr.com/people/glynlowe/\r\n81a7637c67f4f777,https://c1.staticflickr.com/8/7331/27680526741_b962934a0e_o.jpg,0.333333333,0,1,1,490,836,https://www.flickr.com/photos/144110575@N07/27680526741,https://creativecommons.org/licenses/by/2.0/,Jobs For Felons Hub,https://www.flickr.com/people/144110575@N07/\r\n65c1808390bd18c5,https://c1.staticflickr.com/9/8832/28094440470_78960e7877_o.jpg,0,0,0.5625,0.75,686,587,https://www.flickr.com/photos/faircompanies/28094440470,https://creativecommons.org/licenses/by/2.0/,Nicol\\303\\241s Boullosa,https://www.flickr.com/people/faircompanies/\r\ne570258f1bef4ade,https://c1.staticflickr.com/8/7479/27454174925_9950df0ff9_o.jpg,0,0,0.75,0.75,367,461,https://www.flickr.com/photos/rjbailey/27454174925,https://creativecommons.org/licenses/by/2.0/,Ron Bailey,https://www.flickr.com/people/rjbailey/\r\n58bbaa0c5ab520d5,https://c1.staticflickr.com/8/7361/26646332780_6916622aa8_o.jpg,0.25,0.25,1,1,488,716,https://www.flickr.com/photos/kattis_j/26646332780,https://creativecommons.org/licenses/by/2.0/,Katarina Jardenberg,https://www.flickr.com/people/kattis_j/\r\n556d8b457b94c285,https://c1.staticflickr.com/6/5584/30780053951_156743ba04_o.jpg,0,0,0.666666667,1,747,378,https://www.flickr.com/photos/usarak/30780053951/,https://creativecommons.org/licenses/by/2.0/,U.S. Army Alaska (USARAK),https://www.flickr.com/people/usarak/\r\n38546e7d7136c2ab,https://c1.staticflickr.com/6/5495/30613441041_60e2db1f58_o.jpg,0.28125,0,0.84375,0.75,600,114,https://flickr.com/17157315@N00/30613441041,https://creativecommons.org/licenses/by/2.0/,thekirbster,https://www.flickr.com/people/kirbyurner/\r\n9876fe92d003c9d4,https://c1.staticflickr.com/8/7426/27451221106_7d02f7a3b2_o.jpg,0.28125,0,0.84375,0.75,913,357,https://www.flickr.com/photos/10413717@N08/27451221106,https://creativecommons.org/licenses/by/2.0/,Smabs Sputzer,https://www.flickr.com/people/10413717@N08/\r\n8df3dc0c3fef1824,https://c1.staticflickr.com/9/8509/29548650262_4dba7e8039_o.jpg,0.333333333,0,1,1,951,843,https://flickr.com/143363015@N06/29548650262,https://creativecommons.org/licenses/by/2.0/,Tobias Begemann,https://www.flickr.com/people/tobiasbegemann/\r\n938ada2f2bfa2b68,https://c1.staticflickr.com/9/8149/29352102754_7ac0f1c311_o.jpg,0.25,0.25,0.75,1,844,918,https://www.flickr.com/photos/amslerpix/29352102754/,https://creativecommons.org/licenses/by/2.0/,amslerPIX,https://www.flickr.com/people/amslerpix/\r\n61131df39f19a4a0,https://c1.staticflickr.com/6/5497/31308303502_0e6f130313_o.jpg,0,0.25,0.5625,1,615,867,https://www.flickr.com/photos/davidstanleytravel/31308303502,https://creativecommons.org/licenses/by/2.0/,David Stanley,https://www.flickr.com/people/davidstanleytravel/\r\n7278c2b63080f08c,https://c1.staticflickr.com/1/338/32818896976_e770e081a2_o.jpg,0,0,0.666666667,1,349,287,http://www.flickr.com/photos/noortjvdb/32818896976/,https://creativecommons.org/licenses/by/2.0/,noortjevdB,https://www.flickr.com/people/noortjvdb/\r\nf734e87286835196,https://c1.staticflickr.com/6/5618/29467877973_f53d845cd7_o.jpg,0.249079529,0.25,0.747238586,1,324,660,https://www.flickr.com/photos/usfwsmtnprairie/29467877973/,https://creativecommons.org/licenses/by/2.0/,USFWS Mountain-Prairie,https://www.flickr.com/people/usfwsmtnprairie/\r\n06a0e36c443746b5,https://c1.staticflickr.com/1/689/31497463186_001b31fa44_o.jpg,0.5,0,1,0.75,354,232,https://www.flickr.com/photos/dkeats/31497463186,https://creativecommons.org/licenses/by/2.0/,Derek Keats,https://www.flickr.com/people/dkeats/\r\nd7d8fc1442a82a8e,https://c1.staticflickr.com/8/7041/26864593362_503898424a_o.jpg,0,0.25,0.599757674,1,22,846,https://www.flickr.com/photos/andymorffew/26864593362,https://creativecommons.org/licenses/by/2.0/,Andy Morffew,https://www.flickr.com/people/andymorffew/\r\nf9937696f421f4e4,https://c1.staticflickr.com/1/688/32401990166_57c8cdae3d_o.jpg,0.333333333,0,1,1,85,491,https://www.flickr.com/photos/kkoshy/32401990166,https://creativecommons.org/licenses/by/2.0/,Koshy Koshy,https://www.flickr.com/people/kkoshy/\r\n9429f0d1ad2283df,https://c1.staticflickr.com/6/5034/30184154626_f341307e8a_o.jpg,0,0.200323102,1,1,438,464,https://www.flickr.com/photos/janitors/30184154626,https://creativecommons.org/licenses/by/2.0/,K\\304\\201rlis Dambr\\304\\201ns,https://www.flickr.com/people/janitors/\r\n7524a7ef9de8edc3,https://c1.staticflickr.com/9/8825/28821400601_33e22535e7_o.jpg,0.13590604,0,1,1,947,307,https://www.flickr.com/photos/oceanyamaha/28821400601,https://creativecommons.org/licenses/by/2.0/,ocean yamaha,https://www.flickr.com/people/oceanyamaha/\r\nc98a8640dcfaa5e0,https://c1.staticflickr.com/1/656/32592379656_ef2d01af55_o.jpg,0,0,0.563380282,0.75,747,847,https://www.flickr.com/photos/mark6mauno/32592379656/,https://creativecommons.org/licenses/by/2.0/,mark6mauno,https://www.flickr.com/people/mark6mauno/\r\nd34c51d7523e9c0f,https://c1.staticflickr.com/9/8861/28145818093_4c661cab2c_o.jpg,0,0,0.663235294,1,647,655,https://www.flickr.com/photos/shankaronline/28145818093,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n24726d9054e9d78f,https://c1.staticflickr.com/9/8205/28857614150_fc34c1c771_o.jpg,0.421177267,0,0.842354533,0.75,545,455,https://www.flickr.com/photos/wittcogmbh/28857614150/,https://creativecommons.org/licenses/by/2.0/,wittco.gmbh,https://www.flickr.com/people/wittcogmbh/\r\n66aa56675136bbab,https://c1.staticflickr.com/8/7237/27131945991_c25202b461_o.jpg,0,0.331617647,1,0.994852941,470,607,https://www.flickr.com/photos/sterlingcollege/27131945991/,https://creativecommons.org/licenses/by/2.0/,Sterling College,https://www.flickr.com/people/sterlingcollege/\r\na8fe76c687b00e33,https://c1.staticflickr.com/1/538/31501305733_eba5c35a04_o.jpg,0.438671875,0.25,1,1,598,119,https://flickr.com/25187162@N03/31501305733,https://creativecommons.org/licenses/by/2.0/,BRUCE GIBSON,https://www.flickr.com/people/gibsongunleather/\r\n411c5d79017077f4,https://c1.staticflickr.com/8/7357/27945593082_c5faf86d90_o.jpg,0,0,0.744548287,1,82,677,https://www.flickr.com/photos/jkirkhart35/27945593082,https://creativecommons.org/licenses/by/2.0/,Jerry Kirkhart,https://www.flickr.com/people/jkirkhart35/\r\nbb6fcb4a52b4fc0f,https://c1.staticflickr.com/8/7796/26996485435_ce0ae48ae9_o.jpg,0.333333333,0,1,1,735,510,https://www.flickr.com/photos/transitomedellin/26996485435,https://creativecommons.org/licenses/by/2.0/,Secretar\\303\\255a de Movilidad de Medell\\303\\255n,https://www.flickr.com/people/transitomedellin/\r\n84640859ee739095,https://c1.staticflickr.com/9/8484/29052403164_26fb76527d_o.jpg,0.28125,0.25,0.84375,1,720,580,https://www.flickr.com/photos/rusty_clark/29052403164,https://creativecommons.org/licenses/by/2.0/,Rusty Clark ~ 100K Photos,https://www.flickr.com/people/rusty_clark/\r\n6d2d82bc1ba0a956,https://c1.staticflickr.com/1/642/31650552204_c0eed2f414_o.jpg,0,0,0.75,0.713028169,52,664,https://www.flickr.com/photos/jsjgeology/31650552204,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n3e1e47f5af2a3f16,https://c1.staticflickr.com/8/7515/27308980890_45a4af4039_o.jpg,0.251294379,0.25,0.753883136,1,294,357,https://www.flickr.com/photos/dkeats/27308980890,https://creativecommons.org/licenses/by/2.0/,Derek Keats,https://www.flickr.com/people/dkeats/\r\n1ac44ddb6a2d497a,https://c1.staticflickr.com/6/5782/31278384662_150be3d19e_o.jpg,0.333333333,0,1,1,898,27,https://www.flickr.com/photos/lge/31278384662,https://creativecommons.org/licenses/by/2.0/,LG\\354\\240\\204\\354\\236\\220,https://www.flickr.com/people/lge/\r\n7a291389e67a8a07,https://c1.staticflickr.com/9/8164/29763489606_b506759ed9_o.jpg,0,0.333333333,1,1,417,637,https://www.flickr.com/photos/rickmccharles/29763489606/,https://creativecommons.org/licenses/by/2.0/,Rick McCharles,https://www.flickr.com/people/rickmccharles/\r\nafa0f88aaab03427,https://c1.staticflickr.com/9/8284/29065395243_e0aa5811b3_o.jpg,0,0.25,0.5625,1,545,3,https://flickr.com/16813029@N08/29065395243,https://creativecommons.org/licenses/by/2.0/,wittco.gmbh,https://www.flickr.com/people/wittcogmbh/\r\n1e26772fc8c493bb,https://c1.staticflickr.com/6/5725/30011753244_385c391182_o.jpg,0,0,0.5,0.75,416,14,http://www.flickr.com/photos/147860906@N07/30011753244/,https://creativecommons.org/licenses/by/2.0/,DYKT Mohigan,https://www.flickr.com/people/147860906@N07/\r\n4d1237ccbd9a5b4b,https://c1.staticflickr.com/9/8015/29404892510_e505843f23_o.jpg,0.353159851,0.25,0.706319703,1,815,423,https://www.flickr.com/photos/yamahawatercraft/29404892510/,https://creativecommons.org/licenses/by/2.0/,Yamaha Watercraft Group,https://www.flickr.com/people/yamahawatercraft/\r\nc3f452167c39302f,https://c1.staticflickr.com/8/7571/27007736911_797f24353c_o.jpg,0.350800582,0,1,1,296,922,https://www.flickr.com/photos/usdagov/27007736911,https://creativecommons.org/licenses/by/2.0/,U.S. Department of Agriculture,https://www.flickr.com/people/usdagov/\r\n6841e385fe79177f,https://c1.staticflickr.com/1/689/31896146460_9cd904330e_o.jpg,0.255395683,0.25,1,1,830,833,https://flickr.com/45131642@N00/31896146460,https://creativecommons.org/licenses/by/2.0/,Phil Parker,https://www.flickr.com/people/45131642@N00/\r\nb48febf2b989205b,https://c1.staticflickr.com/8/7537/26855046792_36c804efd2_o.jpg,0.25,0.25,0.75,1,760,635,https://www.flickr.com/photos/barnimages/26855046792,https://creativecommons.org/licenses/by/2.0/,barnimages.com,https://www.flickr.com/people/barnimages/\r\nbaeb1300832574aa,https://c1.staticflickr.com/9/8775/29342994144_a7e3326c12_o.jpg,0.454002809,0,0.908005618,0.75,972,660,https://www.flickr.com/photos/hisgett/29342994144/,https://creativecommons.org/licenses/by/2.0/,Tony Hisgett,https://www.flickr.com/people/hisgett/\r\n7b6cc934fda859fa,https://c1.staticflickr.com/9/8752/28029665143_654198da68_o.jpg,0.25,0.28125,1,0.84375,307,653,https://www.flickr.com/photos/yto/28029665143/,https://creativecommons.org/licenses/by/2.0/,Tatsuo Yamashita,https://www.flickr.com/people/yto/\r\n39f82c728c956125,https://c1.staticflickr.com/1/400/32155857566_fb6080d6f9_o.jpg,0.296147673,0,0.888443018,0.75,638,762,http://www.flickr.com/photos/mark-gunn/32155857566/,https://creativecommons.org/licenses/by/2.0/,Mark Gunn,https://www.flickr.com/people/mark-gunn/\r\nc8231978dc765004,https://c1.staticflickr.com/8/7787/28887518643_2e0a3954f5_o.jpg,0.248348018,0.25,0.745044053,1,472,741,https://flickr.com/51434786@N06/28887518643,https://creativecommons.org/licenses/by/2.0/,crudmucosa,https://www.flickr.com/people/crudmucosa/\r\n823f8eae0df2d5c1,https://c1.staticflickr.com/8/7482/27862170094_b1b7c6a698_o.jpg,0,0.25,0.531155015,1,95,570,https://www.flickr.com/photos/usfwsmtnprairie/27862170094,https://creativecommons.org/licenses/by/2.0/,USFWS Mountain-Prairie,https://www.flickr.com/people/usfwsmtnprairie/\r\n038fd7a750733bb5,https://c1.staticflickr.com/1/654/31793584681_74a2ac4e3a_o.jpg,0,0,0.490510949,0.75,18,103,https://www.flickr.com/photos/kkoshy/31793584681,https://creativecommons.org/licenses/by/2.0/,Koshy Koshy,https://www.flickr.com/people/kkoshy/\r\n6d6078c60e59e9f4,https://c1.staticflickr.com/9/8136/30193749621_f4b1f30016_o.jpg,0,0,0.5625,0.75,985,797,https://www.flickr.com/photos/usfwsmidwest/30193749621/,https://creativecommons.org/licenses/by/2.0/,USFWSmidwest,https://www.flickr.com/people/usfwsmidwest/\r\n1f71668ab3dbe81b,https://c1.staticflickr.com/1/452/31908949435_f7e5ec059c_o.jpg,0,0,1,0.75,709,967,https://www.flickr.com/photos/londonmatt/31908949435,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\n70b3b833463d20b9,https://c1.staticflickr.com/8/7052/27277805330_6ee69eaf87_o.jpg,0,0.28125,0.75,0.84375,864,857,https://www.flickr.com/photos/crondeau/27277805330,https://creativecommons.org/licenses/by/2.0/,Christine Rondeau,https://www.flickr.com/people/crondeau/\r\ndd4d033a9a45552d,https://c1.staticflickr.com/9/8580/29739921130_3c98d8504c_o.jpg,0.282719436,0,0.848158307,0.75,392,126,https://www.flickr.com/photos/taylar/29739921130,https://creativecommons.org/licenses/by/2.0/,Ingrid Taylar,https://www.flickr.com/people/taylar/\r\n8800cff23fade16a,https://c1.staticflickr.com/8/7448/26655153174_a2122e0682_o.jpg,0.25,0.215753425,1,0.647260274,532,314,https://www.flickr.com/photos/51764518@N02/26655153174,https://creativecommons.org/licenses/by/2.0/,Joe Haupt,https://www.flickr.com/people/51764518@N02/\r\naf2ac1705dc19a5f,https://c1.staticflickr.com/9/8870/27676047423_1399759e15_o.jpg,0.333333333,0,1,1,384,99,https://www.flickr.com/photos/30802095@N04/27676047423,https://creativecommons.org/licenses/by/2.0/,Liam Lysaght,https://www.flickr.com/people/30802095@N04/\r\n87937b805302f958,https://c1.staticflickr.com/1/737/32203735982_b50d7c2ca2_o.jpg,0,0.25,0.421177267,1,418,873,https://www.flickr.com/photos/indigoskye/32203735982/in/photolist-R4JyKh-Q1hbGP-R1Gh67-HJixCL-QAEACy-QZmsqR-PTxYKd-PW9vX2-R7jm25-R7hjwW-QAqSgq-R7hhzu-R7hgyb-RazyPk-PSDtLK-PRt1D4-R5An1t-R2dxoq-PQYSyx-PQVJLV-Qv1VKs-QQAfjx-QPtAQH-QV1feS-QZUFDW-QiHqud-Qipyb1-QP34Gf-QF3Y5K-QCsidg-QzgK6Z-QGhQVL-QFLXCt-Pw4Qzr-QKAPYa-Ptjweh-Pw4AiR-QGhLiE-Qbw8CG-QKA5zr-QKzZVM-QKzUwk-QwtoyG-Pw3Y58-PtiNHS-PtiF79-Qbvq49-QGgLtS-Qbvg2m-PtioRJ,https://creativecommons.org/licenses/by/2.0/,John Frost,https://www.flickr.com/people/indigoskye/\r\nf33087679c5e7799,https://c1.staticflickr.com/6/5571/31004412114_2bbb2ce701_o.jpg,0.25,0,1,0.641068447,443,588,https://www.flickr.com/photos/duncanh1/31004412114,https://creativecommons.org/licenses/by/2.0/,Dun.can,https://www.flickr.com/people/duncanh1/\r\n6374e7e35ad90f50,https://c1.staticflickr.com/6/5513/31121388731_7eecd1eea3_o.jpg,0,0,0.666666667,1,39,771,https://www.flickr.com/photos/silkebaron/31121388731/,https://creativecommons.org/licenses/by/2.0/,prilfish,https://www.flickr.com/people/silkebaron/\r\n92c73d668e2cd3ce,https://c1.staticflickr.com/6/5444/29773329694_58d4166f7b_o.jpg,0,0,0.75,1,859,582,https://www.flickr.com/photos/togawanderings/29773329694,https://creativecommons.org/licenses/by/2.0/,Thomas Galvez,https://www.flickr.com/people/togawanderings/\r\ne8d5809ee6888f8f,https://c1.staticflickr.com/8/7742/26192609743_0b09bc425e_o.jpg,0,0.25,0.502588757,1,717,641,https://www.flickr.com/photos/riversdale_estate/26192609743,https://creativecommons.org/licenses/by/2.0/,Riversdale Estate,https://www.flickr.com/people/riversdale_estate/\r\n45da32a1cf073a4f,https://c1.staticflickr.com/8/7362/27048969652_bf3a3854f1_o.jpg,0,0.25,0.544615385,1,100,897,https://www.flickr.com/photos/brighton/27048969652,https://creativecommons.org/licenses/by/2.0/,Jim Linwood,https://www.flickr.com/people/brighton/\r\nfe94909a2fad4ffe,https://c1.staticflickr.com/8/7434/27542851474_00e37a272d_o.jpg,0,0.187296417,1,1,547,965,https://www.flickr.com/photos/haywirecustomguitars/27542851474,https://creativecommons.org/licenses/by/2.0/,Rick Mariner,https://www.flickr.com/people/haywirecustomguitars/\r\n00c3cd597f1ee96f,https://c1.staticflickr.com/6/5615/31076781681_e0a951dd43_o.jpg,0,0.464122137,0.75,1,138,96,https://flickr.com/143049688@N08/31076781681,https://creativecommons.org/licenses/by/2.0/,MR MAO PICS,https://www.flickr.com/people/143049688@N08/\r\na5949b0719d3ea77,https://c1.staticflickr.com/9/8667/28605783432_909eee5361_o.jpg,0.25,0.25,0.75,1,981,473,https://www.flickr.com/photos/amslerpix/28605783432/,https://creativecommons.org/licenses/by/2.0/,amslerPIX,https://www.flickr.com/people/amslerpix/\r\nd7984e6e1a51b3c4,https://c1.staticflickr.com/9/8883/28628911182_0982dd70ae_o.jpg,0.4375,0,1,0.75,518,937,https://flickr.com/32916425@N04/28628911182,https://creativecommons.org/licenses/by/2.0/,Jeremy Thompson,https://www.flickr.com/people/rollercoasterphilosophy/\r\n979757e16b67ac7e,https://c1.staticflickr.com/6/5590/31340775762_a825141a4d_o.jpg,0,0.25,0.636439268,1,323,798,https://www.flickr.com/photos/gails_pictures/31340775762/,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n60ab0fb53634bc3c,https://c1.staticflickr.com/9/8410/28986276455_81071dc27d_o.jpg,0,0.25,0.504444444,1,766,552,https://flickr.com/22711505@N05/28986276455,https://creativecommons.org/licenses/by/2.0/,Ron Cogswell,https://www.flickr.com/people/22711505@N05/\r\nad333e5d6953cbbf,https://c1.staticflickr.com/6/5788/30472796174_6eec033c0e_o.jpg,0.25,0,0.75,0.75,633,297,https://flickr.com/21178495@N06/30472796174,https://creativecommons.org/licenses/by/2.0/,Tiago Cassol Schvarstzhaupt,https://www.flickr.com/people/eletrosonico/\r\n2b7e6f0d6f73fa1f,https://c1.staticflickr.com/3/2649/32280811454_5755bb1af9_o.jpg,0,0,0.75,0.732589286,638,235,https://flickr.com/56041749@N02/32280811454,https://creativecommons.org/licenses/by/2.0/,Denise Krebs,https://www.flickr.com/people/mrsdkrebs/\r\n4063377442ea93d3,https://c1.staticflickr.com/1/653/32223796996_304c6b269d_o.jpg,0.25,0,0.75,0.75,596,814,https://www.flickr.com/photos/grimmegroup/32223796996/,https://creativecommons.org/licenses/by/2.0/,Grimme Group,https://www.flickr.com/people/grimmegroup/\r\n60ed157b21d49c4c,https://c1.staticflickr.com/3/2843/32866501816_486e532c1a_o.jpg,0,0.25,0.606331169,1,755,960,https://www.flickr.com/photos/51764518@N02/32866501816,https://creativecommons.org/licenses/by/2.0/,Joe Haupt,https://www.flickr.com/people/51764518@N02/\r\n8f07c92f0af97807,https://c1.staticflickr.com/8/7366/26918793702_778d2226ac_o.jpg,0.5,0.25,1,1,806,108,https://www.flickr.com/photos/claude_wians/26918793702,https://creativecommons.org/licenses/by/2.0/,Claude Wians,https://www.flickr.com/people/claude_wians/\r\nd67b326ef48bb7b7,https://c1.staticflickr.com/6/5350/29991729355_5ab8b027df_o.jpg,0.333333333,0,1,1,874,546,https://www.flickr.com/photos/bholl7510/29991729355/,https://creativecommons.org/licenses/by/2.0/,Brian Holland,https://www.flickr.com/people/bholl7510/\r\n9f3afb58c061c290,https://c1.staticflickr.com/9/8629/27671691554_a8bbdda66f_o.jpg,0.264204545,0.25,0.792613636,1,981,754,https://www.flickr.com/photos/noaaphotolib/27671691554,https://creativecommons.org/licenses/by/2.0/,NOAA Photo Library,https://www.flickr.com/people/noaaphotolib/\r\n1f7c25207bc3e303,https://c1.staticflickr.com/9/8605/29961048043_fe10e305d4_o.jpg,0.25,0,0.75,0.75,476,979,https://flickr.com/56537760@N03/29961048043,https://creativecommons.org/licenses/by/2.0/,Th\\303\\251o Paul,https://www.flickr.com/people/theopaul/\r\n70d812d368e3e828,https://c1.staticflickr.com/9/8261/29319398451_2005584856_o.jpg,0,0,0.75,0.75,456,202,https://www.flickr.com/photos/markmorgantrinidad/29319398451,https://creativecommons.org/licenses/by/2.0/,Mark Morgan,https://www.flickr.com/people/markmorgantrinidad/\r\ne19a59ad09d18497,https://c1.staticflickr.com/8/7300/27213625016_61cc784152_o.jpg,0,0.25,0.559282371,1,443,915,https://www.flickr.com/photos/rutiful/27213625016,https://creativecommons.org/licenses/by/2.0/,Ruth Hara,https://www.flickr.com/people/rutiful/\r\ndde33b51c8c152d2,https://c1.staticflickr.com/4/3784/32757850161_333a586b05_o.jpg,0,0,0.75,1,735,969,https://www.flickr.com/photos/27665395@N05/32757850161,https://creativecommons.org/licenses/by/2.0/,Jason Lawrence,https://www.flickr.com/people/27665395@N05/\r\na33af1b9f8b8a81b,https://c1.staticflickr.com/6/5592/29999027335_510c67fbc7_o.jpg,0.28125,0.25,0.84375,1,880,667,https://www.flickr.com/photos/31029865@N06/29999027335,https://creativecommons.org/licenses/by/2.0/,Dick Thomas Johnson,https://www.flickr.com/people/31029865@N06/\r\n1c924877ccc60129,https://c1.staticflickr.com/3/2002/32192998414_890b5b84db_o.jpg,0.4375,0.25,1,1,410,567,https://www.flickr.com/photos/stevendepolo/32192998414,https://creativecommons.org/licenses/by/2.0/,Steven Depolo,https://www.flickr.com/people/stevendepolo/\r\n74dbbe535861d709,https://c1.staticflickr.com/8/7687/27703779525_8a8baaaae3_o.jpg,0,0,0.666666667,1,702,194,https://www.flickr.com/photos/7armyjmtc/27703779525,https://creativecommons.org/licenses/by/2.0/,7th Army Training Command,https://www.flickr.com/people/7armyjmtc/\r\n858ec91996beb034,https://c1.staticflickr.com/9/8746/28705196386_80e810df65_o.jpg,0,0,0.75,0.75,686,962,https://www.flickr.com/photos/moto_club4ag/28705196386,https://creativecommons.org/licenses/by/2.0/,Moto "Club4AG" Miwa,https://www.flickr.com/people/moto_club4ag/\r\ne5cf220f8144c192,https://c1.staticflickr.com/8/7554/27328391223_6883936624_o.jpg,0.280784844,0,0.842354533,1,706,441,https://www.flickr.com/photos/abhinavnfr/27328391223,https://creativecommons.org/licenses/by/2.0/,Abhinav Phangcho Choudhury,https://www.flickr.com/people/abhinavnfr/\r\n34a39905ce228160,https://c1.staticflickr.com/8/7461/26512655533_d5fe870fbb_o.jpg,0,0.5,0.75,1,769,752,https://www.flickr.com/photos/30478819@N08/26512655533,https://creativecommons.org/licenses/by/2.0/,Marco Verch,https://www.flickr.com/people/30478819@N08/\r\n9455bd672cb9e4bd,https://c1.staticflickr.com/8/7442/27892201106_9c907e9c26_o.jpg,0.25,0,1,1,522,582,https://flickr.com/39160147@N03/27892201106,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n9c509bfb75167fb0,https://c1.staticflickr.com/1/558/32052180035_f042256367_o.jpg,0,0,1,0.831960461,641,748,https://flickr.com/44093058@N06/32052180035,https://creativecommons.org/licenses/by/2.0/,Blondinrikard Fr\\303\\266berg,https://www.flickr.com/people/blondinrikard/\r\n37e7dc99bf2704c7,https://c1.staticflickr.com/8/7278/26737658351_041174fc0e_o.jpg,0.25,0.25,1,0.75,664,610,https://www.flickr.com/photos/97199236@N04/26737658351,https://creativecommons.org/licenses/by/2.0/,Andrey Gaverdovsky,https://www.flickr.com/people/97199236@N04/\r\nf2bfa7a95e787fe1,https://c1.staticflickr.com/4/3918/33459563545_763a160b5a_o.jpg,0.248348018,0,0.745044053,0.75,721,218,https://www.flickr.com/photos/yourbestdigs/33459563545/,https://creativecommons.org/licenses/by/2.0/,Your Best Digs,https://www.flickr.com/people/yourbestdigs/\r\n0f9d8c86a9f38020,https://c1.staticflickr.com/9/8556/29567784475_4f78605d50_o.jpg,0.333333333,0,1,1,760,118,https://www.flickr.com/photos/markusspiske/29567784475/,https://creativecommons.org/licenses/by/2.0/,Markus Spiske,https://www.flickr.com/people/markusspiske/\r\nce05236afebb173d,https://c1.staticflickr.com/1/515/30802275984_2565161d0f_o.jpg,0,0.25,0.710456942,1,536,510,https://www.flickr.com/photos/glorycycles/30802275984,https://creativecommons.org/licenses/by/2.0/,Glory Cycles,https://www.flickr.com/people/glorycycles/\r\n12fb008451be82d4,https://c1.staticflickr.com/1/339/31842507145_e9db35db2b_o.jpg,0.28125,0,0.84375,0.75,583,555,https://www.flickr.com/photos/southbeachcars/31842507145,https://creativecommons.org/licenses/by/2.0/,Phillip Pessar,https://www.flickr.com/people/southbeachcars/\r\n1a7425e8f990e04a,https://c1.staticflickr.com/6/5635/30586945602_29b323c620_o.jpg,0.25,0.25,1,1,608,676,https://www.flickr.com/photos/powellburns/30586945602,https://creativecommons.org/licenses/by/2.0/,Powell Burns,https://www.flickr.com/people/powellburns/\r\n1ba816dc6937b62d,https://c1.staticflickr.com/9/8538/29489317193_4ea10616ea_o.jpg,0.496696035,0.25,0.99339207,1,860,768,https://www.flickr.com/photos/yourbestdigs/29489317193,https://creativecommons.org/licenses/by/2.0/,Your Best Digs,https://www.flickr.com/people/yourbestdigs/\r\na92e49d8bbdc2aae,https://c1.staticflickr.com/8/7292/28045564580_52c7cf78b1_o.jpg,0,0,0.684603886,1,533,564,https://www.flickr.com/photos/regnatarajan/28045564580,https://creativecommons.org/licenses/by/2.0/,Reg Natarajan,https://www.flickr.com/people/regnatarajan/\r\n0e0f1fd2ed183781,https://c1.staticflickr.com/6/5785/29863538926_4b27fc8b6a_o.jpg,0,0,0.80420712,1,114,988,https://www.flickr.com/photos/gails_pictures/29863538926,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\nc83cfca4bc9dd3c7,https://c1.staticflickr.com/8/7338/26885484975_11ac869fe2_o.jpg,0.421177267,0.25,0.842354533,1,696,730,https://flickr.com/65811939@N00/26885484975,https://creativecommons.org/licenses/by/2.0/,Radarsmum67,https://www.flickr.com/people/radarsmum67/\r\n172f992e1a0ebbc3,https://c1.staticflickr.com/6/5598/31398988035_25b2ca9970_o.jpg,0.306930693,0,0.920792079,1,566,492,https://www.flickr.com/photos/kansasscanner/31398988035/,https://creativecommons.org/licenses/by/2.0/,Tyler Silvest,https://www.flickr.com/people/kansasscanner/\r\n565c67fd230f18cc,https://c1.staticflickr.com/9/8783/28325823061_8c659541be_o.jpg,0,0,0.5,0.75,555,854,https://www.flickr.com/photos/115391424@N05/28325823061,https://creativecommons.org/licenses/by/2.0/,lasta29,https://www.flickr.com/people/115391424@N05/\r\nb8596bba57a73794,https://c1.staticflickr.com/8/7380/26874457623_c0dc46b3fa_o.jpg,0.475215517,0.25,0.950431034,1,352,41,https://www.flickr.com/photos/dkeats/26874457623,https://creativecommons.org/licenses/by/2.0/,Derek Keats,https://www.flickr.com/people/dkeats/\r\nd94940216544b2c5,https://c1.staticflickr.com/4/3684/32889855711_216843d54b_o.jpg,0,0,0.745709828,1,312,622,https://www.flickr.com/photos/gails_pictures/32889855711,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n9f55ac35ce9703e5,https://c1.staticflickr.com/8/7326/26785585893_a25ac89a89_o.jpg,0.226157083,0.25,0.678471248,1,9,194,https://www.flickr.com/photos/94066540@N08/26785585893,https://creativecommons.org/licenses/by/2.0/,Stuart Mckiggan,https://www.flickr.com/people/94066540@N08/\r\n55323ce1ff485412,https://c1.staticflickr.com/1/772/32544267962_4093028c88_o.jpg,0,0,0.5,0.75,386,571,https://www.flickr.com/photos/30478819@N08/32544267962,https://creativecommons.org/licenses/by/2.0/,Marco Verch,https://www.flickr.com/people/30478819@N08/\r\n526f322d78bd5b44,https://c1.staticflickr.com/6/5674/30963905825_0ff9253d66_o.jpg,0.271186441,0,1,1,571,288,https://www.flickr.com/photos/taymtaym/30963905825/,https://creativecommons.org/licenses/by/2.0/,taymtaym,https://www.flickr.com/people/taymtaym/\r\n229fbe790098d2e5,https://c1.staticflickr.com/9/8366/29338309226_66c4557833_o.jpg,0,0,0.563380282,0.75,481,576,https://www.flickr.com/photos/jeepersmedia/29338309226,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\ne0c0f412d78571dc,https://c1.staticflickr.com/8/7079/27237503592_028a10898b_o.jpg,0.25,0,1,0.748646209,734,409,https://www.flickr.com/photos/alvy/27237503592,https://creativecommons.org/licenses/by/2.0/,\\303\\201lvaro Ib\\303\\241\\303\\261ez,https://www.flickr.com/people/alvy/\r\nd12aad0d18252b40,https://c1.staticflickr.com/9/8155/29394215765_0a9fb71d55_o.jpg,0.5,0,1,0.75,325,39,https://www.flickr.com/photos/yutakaseki/29394215765,https://creativecommons.org/licenses/by/2.0/,Yutaka Seki,https://www.flickr.com/people/yutakaseki/\r\n22712d0324d43ef2,https://c1.staticflickr.com/8/7783/29065239841_856237083c_o.jpg,0,0.25,0.75,1,898,847,https://www.flickr.com/photos/lge/29065239841,https://creativecommons.org/licenses/by/2.0/,LG\\354\\240\\204\\354\\236\\220,https://www.flickr.com/people/lge/\r\n3201d10ec4dca056,https://c1.staticflickr.com/9/8335/28756723693_0d013b2d23_o.jpg,0.25,0,0.75,0.75,739,605,https://www.flickr.com/photos/wuipdesign/28756723693,https://creativecommons.org/licenses/by/2.0/,Wuipdesign,https://www.flickr.com/people/wuipdesign/\r\n62c7958e7cd80cca,https://c1.staticflickr.com/9/8422/29506847095_d1d503ca6e_o.jpg,0,0.323327616,0.75,1,402,757,https://www.flickr.com/photos/atoach/29506847095/,https://creativecommons.org/licenses/by/2.0/,Tim Green,https://www.flickr.com/people/atoach/\r\n12cf4de43c70273e,https://c1.staticflickr.com/9/8107/29584374652_9e00ea0c4c_o.jpg,0.608865711,0.25,1,1,815,833,https://flickr.com/42219086@N05/29584374652,https://creativecommons.org/licenses/by/2.0/,Yamaha Watercraft Group,https://www.flickr.com/people/yamahawatercraft/\r\n56dbd7d814ef5c58,https://c1.staticflickr.com/8/7633/27183091666_8c5fc3c63a_o.jpg,0,0,0.669128508,1,254,277,https://www.flickr.com/photos/70253321@N00/27183091666,https://creativecommons.org/licenses/by/2.0/,fugzu,https://www.flickr.com/people/70253321@N00/\r\ne91abe0f1a7a30f4,https://c1.staticflickr.com/9/8428/29430666931_e770d94da0_o.jpg,0.480130058,0,0.960260116,0.75,752,495,https://www.flickr.com/photos/omnious100/29430666931,https://creativecommons.org/licenses/by/2.0/,Arturo Hurtado,https://www.flickr.com/people/omnious100/\r\n641d03527848417f,https://c1.staticflickr.com/6/5341/30230914936_690327db34_o.jpg,0,0.25,0.551470588,1,341,450,https://www.flickr.com/photos/happymillerman/30230914936,https://creativecommons.org/licenses/by/2.0/,happymillerman,https://www.flickr.com/people/happymillerman/\r\nde282cea97167923,https://c1.staticflickr.com/9/8491/28883167640_a8d08d7c53_o.jpg,0,0.330871492,1,1,305,656,https://www.flickr.com/photos/52450054@N04/28883167640,https://creativecommons.org/licenses/by/2.0/,Judy Gallagher,https://www.flickr.com/people/52450054@N04/\r\n088560551b91ceb0,https://c1.staticflickr.com/8/7287/26823150326_c79c4cf404_o.jpg,0.25,0.4375,1,1,983,797,https://www.flickr.com/photos/ewolivera/26823150326,https://creativecommons.org/licenses/by/2.0/,Edgardo W. Olivera,https://www.flickr.com/people/ewolivera/\r\na70b1cda0776cba6,https://c1.staticflickr.com/1/527/32735802985_cf85631809_o.jpg,0,0.25,0.5,1,143,119,https://www.flickr.com/photos/ddebold/32735802985/,https://creativecommons.org/licenses/by/2.0/,Don DeBold,https://www.flickr.com/people/ddebold/\r\nc036b20afde859b7,https://c1.staticflickr.com/9/8254/28443278150_007ab658ba_o.jpg,0,0,0.5625,0.75,281,508,https://www.flickr.com/photos/californiadfg/28443278150/,https://creativecommons.org/licenses/by/2.0/,California Department of Fish and Wildlife,https://www.flickr.com/people/californiadfg/\r\n5d6c4759c26bae8c,https://c1.staticflickr.com/9/8174/28972300461_1d85484b62_o.jpg,0.25,0.25,0.75,1,547,746,https://www.flickr.com/photos/30478819@N08/28972300461,https://creativecommons.org/licenses/by/2.0/,Marco Verch,https://www.flickr.com/people/30478819@N08/\r\n3eded0a8d5f16059,https://c1.staticflickr.com/8/7448/27356260545_152e77a25a_o.jpg,0.239370079,0,1,1,100,457,https://www.flickr.com/photos/brighton/27356260545,https://creativecommons.org/licenses/by/2.0/,Jim Linwood,https://www.flickr.com/people/brighton/\r\n9db1070b1b9f03d2,https://c1.staticflickr.com/4/3818/33059801071_2681885776_o.jpg,0,0,0.522624434,0.75,324,870,https://www.flickr.com/photos/132033298@N04/33059801071,https://creativecommons.org/licenses/by/2.0/,C. P. Ewing,https://www.flickr.com/people/132033298@N04/\r\ne3c3f8889ac99aec,https://c1.staticflickr.com/9/8786/28331257953_f5d077c2c1_o.jpg,0,0.248713235,0.75,0.746139706,536,897,https://www.flickr.com/photos/boostsamurai/28331257953,https://creativecommons.org/licenses/by/2.0/,JAK SIE MASZ,https://www.flickr.com/people/boostsamurai/\r\nd6d42f8a1f6b4460,https://c1.staticflickr.com/9/8230/29839990905_33b2231132_o.jpg,0.564649243,0,1,0.75,666,374,https://www.flickr.com/photos/pedrosimoes7/29839990905,https://creativecommons.org/licenses/by/2.0/,Pedro Ribeiro Sim\\303\\265es,https://www.flickr.com/people/pedrosimoes7/\r\n517dc655edeaf344,https://c1.staticflickr.com/8/7309/26590993833_7b17146eda_o.jpg,0,0,0.5,0.75,430,920,https://www.flickr.com/photos/slgc/26590993833,https://creativecommons.org/licenses/by/2.0/,slgckgc,https://www.flickr.com/people/slgc/\r\n10db986bad48a9d5,https://c1.staticflickr.com/8/7092/26592880754_0af21a54f5_o.jpg,0.25,0,1,0.75,894,134,https://www.flickr.com/photos/vertstone/26592880754,https://creativecommons.org/licenses/by/2.0/,Sunil Dass,https://www.flickr.com/people/vertstone/\r\n17cfccd62451a233,https://c1.staticflickr.com/8/7005/27391093362_0c39bc1805_o.jpg,0.25,0.25,1,1,966,571,https://www.flickr.com/photos/rjbailey/27391093362/,https://creativecommons.org/licenses/by/2.0/,Ron Bailey,https://www.flickr.com/people/rjbailey/\r\n212476b1b58a0cfe,https://c1.staticflickr.com/6/5349/29372567534_c17516e486_o.jpg,0.255402385,0.25,0.766207154,1,512,697,http://www.flickr.com/photos/andreboeni/29372567534/,https://creativecommons.org/licenses/by/2.0/,Andrew Bone,https://www.flickr.com/people/andreboeni/\r\n919a6f1d23c6d05e,https://c1.staticflickr.com/6/5348/31279804625_839baf7892_o.jpg,0,0.25,0.5,1,145,983,https://www.flickr.com/photos/ddebold/31279804625,https://creativecommons.org/licenses/by/2.0/,Don DeBold,https://www.flickr.com/people/ddebold/\r\nda4f4d6944d93ae0,https://c1.staticflickr.com/6/5260/29956436271_478885e6f4_o.jpg,0,0,0.75,1,735,779,https://www.flickr.com/photos/27665395@N05/29956436271,https://creativecommons.org/licenses/by/2.0/,Jason Lawrence,https://www.flickr.com/people/27665395@N05/\r\n1461ed9402248ddc,https://c1.staticflickr.com/8/7699/27042345425_591120613f_o.jpg,0.25,0.25,0.75,1,514,725,https://www.flickr.com/photos/jimfischer/27042345425/,https://creativecommons.org/licenses/by/2.0/,Jim Fischer,https://www.flickr.com/people/jimfischer/\r\nf0dcd519b5499d48,https://c1.staticflickr.com/6/5218/29421210913_9b476c8d85_o.jpg,0.230524079,0.25,0.691572238,1,664,231,https://www.flickr.com/photos/eltb/29421210913,https://creativecommons.org/licenses/by/2.0/,Catedrales e Iglesias/Cathedrals and Churches,https://www.flickr.com/people/eltb/\r\n9a5913e0fb6d0866,https://c1.staticflickr.com/9/8795/28922147966_1362994487_o.jpg,0,0,0.5625,0.75,767,204,https://www.flickr.com/photos/londonmatt/28922147966/,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\ncfab8f880da31ad8,https://c1.staticflickr.com/1/35/31390789132_d7ab56b8eb_o.jpg,0.28125,0,0.84375,0.75,541,308,https://www.flickr.com/photos/jsjgeology/31390789132,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n17df043d369008f5,https://c1.staticflickr.com/9/8249/28958829961_fe65fb3bb9_o.jpg,0.281690141,0.25,0.845070423,1,991,802,https://www.flickr.com/photos/scotnelson/28958829961,https://creativecommons.org/licenses/by/2.0/,Scot Nelson,https://www.flickr.com/people/scotnelson/\r\n95d6a4087f394d8b,https://c1.staticflickr.com/9/8129/28719483854_d03b5d7da2_o.jpg,0.333333333,0,1,1,750,541,https://flickr.com/96541566@N06/28719483854,https://creativecommons.org/licenses/by/2.0/,Jan Helebrant,https://www.flickr.com/people/96541566@N06/\r\n0be391239ccba0f2,https://c1.staticflickr.com/4/3905/32853501241_fc35720ce1_o.jpg,0,0,0.7484375,1,719,996,https://www.flickr.com/photos/feeney1982/32853501241,https://creativecommons.org/licenses/by/2.0/,MK Feeney,https://www.flickr.com/people/feeney1982/\r\n8d68931ba23fdeb7,https://c1.staticflickr.com/8/7102/27075766490_313dd1d866_o.jpg,0,0.267530488,0.75,0.802591463,842,800,https://www.flickr.com/photos/spreadshirt/27075766490,https://creativecommons.org/licenses/by/2.0/,Spreadshirt,https://www.flickr.com/people/spreadshirt/\r\n6bff4ecb94f1f84e,https://c1.staticflickr.com/9/8347/28860027516_8f8d9bee70_o.jpg,0.257672156,0,0.773016467,0.75,859,866,https://www.flickr.com/photos/governosp/28860027516,https://creativecommons.org/licenses/by/2.0/,Governo do Estado de S\\303\\243o Paulo,https://www.flickr.com/people/governosp/\r\n96fcb76927274db6,https://c1.staticflickr.com/9/8705/28851545335_aa85baa5f3_o.jpg,0.5,0.25,1,1,472,316,https://www.flickr.com/photos/quinet/28851545335,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n4de063bd5b838f11,https://c1.staticflickr.com/6/5669/30863954931_bc7e58710c_o.jpg,0.28125,0,0.84375,0.75,2,289,https://www.flickr.com/photos/tamakisono/30863954931/,https://creativecommons.org/licenses/by/2.0/,Tamaki Sono,https://www.flickr.com/people/tamakisono/\r\n55b85f4336d9603a,https://c1.staticflickr.com/8/7245/27089308192_03759b12c3_o.jpg,0,0,1,1,880,30,https://www.flickr.com/photos/10459174@N03/27089308192/,https://creativecommons.org/licenses/by/2.0/,Fred Faulkner,https://www.flickr.com/people/10459174@N03/\r\nf6437ed4f1b8fe23,https://c1.staticflickr.com/6/5695/31218448105_c5d0e22fd3_o.jpg,0.160367171,0,0.481101512,1,302,843,http://www.flickr.com/photos/131914450@N06/31218448105,https://creativecommons.org/licenses/by/2.0/,burgerking1975,https://www.flickr.com/people/131914450@N06/\r\na90d9627e2145090,https://c1.staticflickr.com/9/8059/29463573752_e27f82a1e6_o.jpg,0.510583942,0.25,1,1,426,722,https://www.flickr.com/photos/60712129@N06/29463573752/,https://creativecommons.org/licenses/by/2.0/,Dianne Lacourciere,https://www.flickr.com/people/60712129@N06/\r\nf4bde5facfdca0c6,https://c1.staticflickr.com/1/405/32476678205_491d504f85_o.jpg,0,0,1,1,788,194,https://flickr.com/14385354@N00/32476678205,https://creativecommons.org/licenses/by/2.0/,Ungry Young Man,https://www.flickr.com/people/markusunger/\r\n803450d936b35ebb,https://c1.staticflickr.com/8/7476/29073407732_3a52be689b_o.jpg,0.249631811,0,0.748895434,0.75,934,258,https://www.flickr.com/photos/comedynose/29073407732/,https://creativecommons.org/licenses/by/2.0/,Pete,https://www.flickr.com/people/comedynose/\r\nfff35cdcce3cde43,https://c1.staticflickr.com/9/8463/29106205880_18aa360993_o.jpg,0,0,0.5,0.75,913,409,https://www.flickr.com/photos/davebloggs007/29106205880,https://creativecommons.org/licenses/by/2.0/,Thank you for visiting my page,https://www.flickr.com/people/davebloggs007/\r\nb58e8d136aff5a50,https://c1.staticflickr.com/9/8188/29745896581_ff26e547ca_o.jpg,0.25,0.25,0.75,1,625,565,https://www.flickr.com/photos/fst_oeb_nrw/29745896581,https://creativecommons.org/licenses/by/2.0/,Fachstelle f\\303\\274r \\303\\226ffentliche Bibliotheken NRW,https://www.flickr.com/people/fst_oeb_nrw/\r\n76d7e2d1bf3a31fe,https://c1.staticflickr.com/9/8072/29787724155_f733ca9146_o.jpg,0.421177267,0.25,0.842354533,1,312,732,https://www.flickr.com/photos/31031835@N08/29787724155,https://creativecommons.org/licenses/by/2.0/,John Tann,https://www.flickr.com/people/31031835@N08/\r\na777db736ecdfb65,https://c1.staticflickr.com/8/7324/27415463920_2958a226d7_o.jpg,0,0.128161889,1,1,326,56,https://www.flickr.com/photos/wildreturn/27415463920,https://creativecommons.org/licenses/by/2.0/,Andy Reago & Chrissy McClarren,https://www.flickr.com/people/wildreturn/\r\nf1a6fdadaa132225,https://c1.staticflickr.com/6/5455/30574341260_ec628a854c_o.jpg,0.25,0,1,0.75,990,18,https://www.flickr.com/photos/hills_alive/30574341260,https://creativecommons.org/licenses/by/2.0/,Caroline,https://www.flickr.com/people/hills_alive/\r\n6bea02f26beef50d,https://c1.staticflickr.com/6/5700/29664825844_e3a32c8217_o.jpg,0.249079529,0,0.747238586,0.75,709,267,https://www.flickr.com/photos/londonmatt/29664825844,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\n7b6a9bf5b175e3b2,https://c1.staticflickr.com/9/8270/30005296161_6f99755f62_o.jpg,0.372611465,0,0.74522293,0.75,558,599,https://www.flickr.com/photos/imo-un/30005296161,https://creativecommons.org/licenses/by/2.0/,International Maritime Organization,https://www.flickr.com/people/imo-un/\r\n7c249a944594a62e,https://c1.staticflickr.com/6/5810/30483710105_e24bbc2699_o.jpg,0,0,0.5625,0.75,990,620,https://www.flickr.com/photos/verzo/30483710105,https://creativecommons.org/licenses/by/2.0/,Roberto Verzo,https://www.flickr.com/people/verzo/\r\n5c2d823a4d30e3d5,https://c1.staticflickr.com/6/5613/30328326823_cd02f11183_o.jpg,0.248348018,0.25,0.745044053,1,718,894,https://www.flickr.com/photos/thartz00/30328326823,https://creativecommons.org/licenses/by/2.0/,likeaduck,https://www.flickr.com/people/thartz00/\r\nfc94d8ed9eca8c51,https://c1.staticflickr.com/2/1478/26763589175_4f80c5497b_o.jpg,0,0,0.666666667,1,390,584,https://www.flickr.com/photos/stevendepolo/26763589175,https://creativecommons.org/licenses/by/2.0/,Steven Depolo,https://www.flickr.com/people/stevendepolo/\r\n73728c249317ccb9,https://c1.staticflickr.com/1/623/31659729526_41139a4893_o.jpg,0.28125,0,0.84375,0.75,848,145,http://www.flickr.com/photos/mohit_s/31659729526/,https://creativecommons.org/licenses/by/2.0/,Mohit S,https://www.flickr.com/people/mohit_s/\r\n0b1d45bd9ab1064e,https://c1.staticflickr.com/9/8272/29868746026_7a9165f2d9_o.jpg,0,0,0.666666667,1,561,563,https://www.flickr.com/photos/jill_carlson/29868746026,https://creativecommons.org/licenses/by/2.0/,Jill Carlson (jillcarlson.org),https://www.flickr.com/people/jill_carlson/\r\n4b52fd42977215b9,https://c1.staticflickr.com/8/7555/27346942261_e7b059637c_o.jpg,0,0,0.5625,0.75,901,627,https://www.flickr.com/photos/rusty_clark/27346942261,https://creativecommons.org/licenses/by/2.0/,Rusty Clark ~ 100K Photos,https://www.flickr.com/people/rusty_clark/\r\nc27e3bde18fba21b,https://c1.staticflickr.com/6/5127/30107077575_dce955ff33_o.jpg,0,0.25,0.492324561,1,323,128,https://www.flickr.com/photos/postmanpetecoluk/30107077575,https://creativecommons.org/licenses/by/2.0/,postman.pete,https://www.flickr.com/people/postmanpetecoluk/\r\n5dbfff56258cf9e6,https://c1.staticflickr.com/1/613/31473791884_176af43beb_o.jpg,0,0.262273414,0.75,0.786820242,318,385,https://flickr.com/52450054@N04/31473791884,https://creativecommons.org/licenses/by/2.0/,Judy Gallagher,https://www.flickr.com/people/52450054@N04/\r\nc47ca11294a9aefe,https://c1.staticflickr.com/9/8724/28243588105_0e8673ea4b_o.jpg,0.25,0.25,1,0.75,547,119,https://www.flickr.com/photos/80037792@N03/28243588105/,https://creativecommons.org/licenses/by/2.0/,Adri\\303\\241n Castillo Rivera,https://www.flickr.com/people/80037792@N03/\r\n97917829607b6189,https://c1.staticflickr.com/1/435/31542814264_1481334d51_o.jpg,0.25,0,0.75,0.75,859,660,https://www.flickr.com/photos/72821066@N04/31542814264,https://creativecommons.org/licenses/by/2.0/,Kathleen,https://www.flickr.com/people/72821066@N04/\r\nd386396825739960,https://c1.staticflickr.com/6/5671/31045268386_89505e1441_o.jpg,0.514898256,0,1,0.75,854,227,https://www.flickr.com/photos/13476480@N07/31045268386,https://creativecommons.org/licenses/by/2.0/,manhhai,https://www.flickr.com/people/13476480@N07/\r\n44fa09bc5f8e69a5,https://c1.staticflickr.com/8/7556/27540902503_548da9039e_o.jpg,0.4108,0,1,0.75,304,990,https://www.flickr.com/photos/gails_pictures/27540902503,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n70db5a1689f89129,https://c1.staticflickr.com/8/7368/28008608932_93cb96fee0_o.jpg,0.238832853,0,0.716498559,0.75,434,472,https://www.flickr.com/photos/familymwr/28008608932/,https://creativecommons.org/licenses/by/2.0/,U.S. Army,https://www.flickr.com/people/familymwr/\r\n89d60f53d478c4f7,https://c1.staticflickr.com/8/7303/26926132240_65cf222cf4_o.jpg,0,0.25,0.5,1,668,435,https://www.flickr.com/photos/codnewsroom/26926132240,https://creativecommons.org/licenses/by/2.0/,COD Newsroom,https://www.flickr.com/people/codnewsroom/\r\nd531a31add433f46,https://c1.staticflickr.com/9/8621/28763062545_b79f4f72ff_o.jpg,0.5,0.25,1,1,764,769,https://www.flickr.com/photos/rusty_clark/28763062545,https://creativecommons.org/licenses/by/2.0/,Rusty Clark ~ 100K Photos,https://www.flickr.com/people/rusty_clark/\r\n4d0c0f35a0c00bdc,https://c1.staticflickr.com/8/7262/26732253551_7b396172cf_o.jpg,0,0,0.75,0.636439268,680,432,https://www.flickr.com/photos/kayadams/26732253551,https://creativecommons.org/licenses/by/2.0/,Kay Adams,https://www.flickr.com/people/kayadams/\r\n33197fa6ae88508a,https://c1.staticflickr.com/9/8032/29583923580_143f4756d0_o.jpg,0.398058252,0.25,1,1,848,30,https://www.flickr.com/photos/zachievenor/29583923580,https://creativecommons.org/licenses/by/2.0/,Zachi Evenor,https://www.flickr.com/people/zachievenor/\r\n78059d1afd7b9fea,https://c1.staticflickr.com/9/8268/28771428103_98f31c968c_o.jpg,0,0,1,1,58,248,https://www.flickr.com/photos/brutchersp/28771428103,https://creativecommons.org/licenses/by/2.0/,Sam Brutcher,https://www.flickr.com/people/brutchersp/\r\n542c07e8bf8e283c,https://c1.staticflickr.com/1/530/32275825531_4a40f15002_o.jpg,0,0,0.5625,0.75,145,706,https://flickr.com/77110886@N07/32275825531,https://creativecommons.org/licenses/by/2.0/,AmishRob,https://www.flickr.com/people/77110886@N07/\r\ne8c99e0220ab46d2,https://c1.staticflickr.com/8/7408/27465191295_e5f6865548_o.jpg,0,0.244318182,1,0.732954545,591,535,https://www.flickr.com/photos/50019407@N03/27465191295,https://creativecommons.org/licenses/by/2.0/,striegel,https://www.flickr.com/people/50019407@N03/\r\nc755f3b0f89028fc,https://c1.staticflickr.com/6/5656/30912005906_eeb50d4284_o.jpg,0.4375,0,1,0.75,566,827,https://www.flickr.com/photos/railwayofaustralia/30912005906/,https://creativecommons.org/licenses/by/2.0/,Railways of Australia by Daryle Phillips,https://www.flickr.com/people/railwayofaustralia/\r\n76db169867bf2e9d,https://c1.staticflickr.com/6/5759/30161761356_e4053ff3fb_o.jpg,0.25,0,1,0.75,964,908,https://flickr.com/93362968@N00/30161761356,https://creativecommons.org/licenses/by/2.0/,Ruben,https://www.flickr.com/people/surfzone/\r\n784232355d49601b,https://c1.staticflickr.com/1/624/31694410765_dea6857dfb_o.jpg,0.321273517,0,0.96382055,1,327,161,https://www.flickr.com/photos/42244964@N03/31694410765,https://creativecommons.org/licenses/by/2.0/,Frank Vassen,https://www.flickr.com/people/42244964@N03/\r\ne99ceff3e420ef95,https://c1.staticflickr.com/8/7381/27954856402_191862c411_o.jpg,0.27818818,0.25,0.834564541,1,831,768,https://www.flickr.com/photos/photographingtravis/27954856402/,https://creativecommons.org/licenses/by/2.0/,Travis Wise,https://www.flickr.com/people/photographingtravis/\r\n23955512cd0e1191,https://c1.staticflickr.com/8/7578/30160381445_b3e6087902_o.jpg,0,0.25,0.497426471,1,476,742,https://www.flickr.com/photos/kylemcdonald/30160381445,https://creativecommons.org/licenses/by/2.0/,Kyle McDonald,https://www.flickr.com/people/kylemcdonald/\r\nd6eac6858474111c,https://c1.staticflickr.com/8/7320/27399110256_d3fe95b790_o.jpg,0.248826291,0,1,1,323,250,https://www.flickr.com/photos/j_benson/27399110256/,https://creativecommons.org/licenses/by/2.0/,John Benson,https://www.flickr.com/people/j_benson/\r\n8d90c523b27410a7,https://c1.staticflickr.com/8/7364/27529300064_90409b557b_o.jpg,0,0.491827637,0.75,1,321,563,https://www.flickr.com/photos/postmanpetecoluk/27529300064,https://creativecommons.org/licenses/by/2.0/,postman.pete,https://www.flickr.com/people/postmanpetecoluk/\r\n5cf72b48fdf5e419,https://c1.staticflickr.com/6/5718/30045803555_c7a2fba567_o.jpg,0.495552732,0,0.991105464,1,630,744,https://www.flickr.com/photos/135366503@N05/30045803555,https://creativecommons.org/licenses/by/2.0/,CLAUDIA DEA,https://www.flickr.com/people/135366503@N05/\r\n228fec6cab5f560d,https://c1.staticflickr.com/1/668/31760002531_3b7716366a_o.jpg,0.333333333,0,1,1,974,931,https://flickr.com/39136124@N00/31760002531,https://creativecommons.org/licenses/by/2.0/,Paul Asman and Jill Lenoble,https://www.flickr.com/people/pauljill/\r\ne7b61bec5e4ca40e,https://c1.staticflickr.com/6/5663/30514982830_d4031e75fe_o.jpg,0,0.333049404,0.75,1,324,674,https://flickr.com/44465748@N00/30514982830,https://creativecommons.org/licenses/by/2.0/,benet2006,https://www.flickr.com/people/benetd/\r\nd3551db0b83c8fc0,https://c1.staticflickr.com/9/8310/29174628816_300ec9d85c_o.jpg,0.5,0.25,1,1,618,324,https://www.flickr.com/photos/dccentralkitchen/29174628816,https://creativecommons.org/licenses/by/2.0/,DC Central Kitchen,https://www.flickr.com/people/dccentralkitchen/\r\nc7d0c73d636300f5,https://c1.staticflickr.com/4/3390/3194974577_b7573f6c32_o.jpg,0.489195231,0.25,1,1,830,801,https://www.flickr.com/photos/davidwilson1949/3194974577,https://creativecommons.org/licenses/by/2.0/,David Wilson,https://www.flickr.com/people/davidwilson1949/\r\n61ed2594a0524631,https://c1.staticflickr.com/8/7456/27590651131_7c9941b3ce_o.jpg,0,0.25,0.5,1,472,996,https://www.flickr.com/photos/prayitnophotography/27590651131,https://creativecommons.org/licenses/by/2.0/,Prayitno / Thank you for (12 millions +) view,https://www.flickr.com/people/prayitnophotography/\r\n89f48bcf97aa6600,https://c1.staticflickr.com/9/8478/29134440636_664709aa0b_o.jpg,0.25,0.25,0.75,1,418,70,https://www.flickr.com/photos/pavdw/29134440636,https://creativecommons.org/licenses/by/2.0/,Paul VanDerWerf,https://www.flickr.com/people/pavdw/\r\ned786d792c4309a4,https://c1.staticflickr.com/8/7630/27108493571_721c8b15ac_o.jpg,0.25,0,1,0.75,499,962,https://www.flickr.com/photos/jjeff/27108493571,https://creativecommons.org/licenses/by/2.0/,Jeff Robbins,https://www.flickr.com/people/jjeff/\r\n9fe19d5c6dab83a3,https://c1.staticflickr.com/8/7507/29079497195_c97c2f18c1_o.jpg,0.5,0,1,0.75,455,805,https://www.flickr.com/photos/stevendepolo/29079497195,https://creativecommons.org/licenses/by/2.0/,Steven Depolo,https://www.flickr.com/people/stevendepolo/\r\nd5d5f843f100328c,https://c1.staticflickr.com/8/7366/26927066371_055fb093dc_o.jpg,0.25,0.25,1,1,886,538,https://www.flickr.com/photos/garlandcannon/26927066371,https://creativecommons.org/licenses/by/2.0/,garlandcannon,https://www.flickr.com/people/garlandcannon/\r\n789417c091011f53,https://c1.staticflickr.com/6/5632/30859118166_ac1328ed72_o.jpg,0.5,0.25,1,1,345,177,https://flickr.com/145232592@N06/30859118166,https://creativecommons.org/licenses/by/2.0/,Martin7d2,https://www.flickr.com/people/martin7d2/\r\ne648427e38e80ff8,https://c1.staticflickr.com/8/7440/27289001152_11922d530c_o.jpg,0.466894977,0.25,1,1,888,617,https://www.flickr.com/photos/marinelson/27289001152,https://creativecommons.org/licenses/by/2.0/,Marinelson Almeida - Traveling through Brazil,https://www.flickr.com/people/marinelson/\r\n1478a0fc99a988db,https://c1.staticflickr.com/8/7775/29139964033_0509a1d34c_o.jpg,0.25,0.28125,1,0.84375,776,928,https://www.flickr.com/photos/juggernautco/29139964033,https://creativecommons.org/licenses/by/2.0/,Daniel X. O'Neil,https://www.flickr.com/people/juggernautco/\r\ne758de2cbb3d10a0,https://c1.staticflickr.com/6/5341/30948537862_3f3461e1a1_o.jpg,0,0.25,0.63875,1,305,905,https://www.flickr.com/photos/gails_pictures/30948537862,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n8af30e5c3fa48a08,https://c1.staticflickr.com/8/7646/29079347226_a8454b4835_o.jpg,0.249079529,0.25,0.747238586,1,696,71,https://www.flickr.com/photos/brandedbulltank/29079347226/,https://creativecommons.org/licenses/by/2.0/,Tank,https://www.flickr.com/people/brandedbulltank/\r\n38287d2e5a4a9a57,https://c1.staticflickr.com/9/8518/28286972964_4a723a160a_o.jpg,0.25,0.25,1,1,960,798,https://www.flickr.com/photos/g4gti/28286972964/,https://creativecommons.org/licenses/by/2.0/,Ryohei Noda,https://www.flickr.com/people/g4gti/\r\nc6e0d2ab3bbf4a3e,https://c1.staticflickr.com/8/7053/27295097944_149d29443d_o.jpg,0.28125,0,0.84375,0.75,706,838,https://www.flickr.com/photos/davidwilson1949/27295097944,https://creativecommons.org/licenses/by/2.0/,David Wilson,https://www.flickr.com/people/davidwilson1949/\r\n4cccb215bd60b3c3,https://c1.staticflickr.com/2/1476/26151802513_eba98c8b71_o.jpg,0.333333333,0,1,1,33,341,https://www.flickr.com/photos/briangratwicke/26151802513,https://creativecommons.org/licenses/by/2.0/,Brian Gratwicke,https://www.flickr.com/people/briangratwicke/\r\n9a49001b63903f72,https://c1.staticflickr.com/9/8320/29938183125_fbbd772391_o.jpg,0.333333333,0,1,1,769,804,https://www.flickr.com/photos/abaecherliphotography/29938183125/,https://creativecommons.org/licenses/by/2.0/,Samuel Ab\\303\\244cherli,https://www.flickr.com/people/abaecherliphotography/\r\n1cbae41091a750dc,https://c1.staticflickr.com/8/7437/27918206525_1045d33569_o.jpg,0,0.25,0.561328125,1,473,814,http://www.flickr.com/16915761@N08/27918206525,https://creativecommons.org/licenses/by/2.0/,Gordon Watt,https://www.flickr.com/people/gordon_watt/\r\nde40907c1e2c5515,https://c1.staticflickr.com/8/7717/26258443973_8a326bac94_o.jpg,0,0,1,0.75,629,486,https://www.flickr.com/photos/yutakaseki/26258443973,https://creativecommons.org/licenses/by/2.0/,Yutaka Seki,https://www.flickr.com/people/yutakaseki/\r\n446f75fdea0e1da0,https://c1.staticflickr.com/9/8197/28459775774_861c11bd49_o.jpg,0.28125,0,0.84375,0.75,922,201,https://www.flickr.com/photos/pussreboots/28459775774/,https://creativecommons.org/licenses/by/2.0/,caligula1995,https://www.flickr.com/people/pussreboots/\r\n059da9a174262427,https://c1.staticflickr.com/8/7529/26612408253_d5156925ff_o.jpg,0,0.25,0.521428571,1,254,630,https://www.flickr.com/photos/70253321@N00/26612408253/,https://creativecommons.org/licenses/by/2.0/,fugzu,https://www.flickr.com/people/70253321@N00/\r\n35c344ca0584ba14,https://c1.staticflickr.com/6/5759/31004746922_f545e4c9e3_o.jpg,0.504405286,0,1,0.75,688,831,https://www.flickr.com/photos/eltb/31004746922,https://creativecommons.org/licenses/by/2.0/,Catedrales e Iglesias/Cathedrals and Churches,https://www.flickr.com/people/eltb/\r\nc335ce88293384f2,https://c1.staticflickr.com/8/7345/27701978545_2cf121bede_o.jpg,0,0,0.75,1,735,629,https://www.flickr.com/photos/27665395@N05/27701978545,https://creativecommons.org/licenses/by/2.0/,Jason Lawrence,https://www.flickr.com/people/27665395@N05/\r\ndaa6404780b1af27,https://c1.staticflickr.com/6/5785/29921098450_5016da1899_o.jpg,0.25,0.25,0.75,1,638,999,https://www.flickr.com/photos/bazzadarambler/29921098450,https://creativecommons.org/licenses/by/2.0/,bazzadarambler,https://www.flickr.com/people/bazzadarambler/\r\n446b24f903b488fc,https://c1.staticflickr.com/6/5521/31389531466_ba559c04c8_o.jpg,0,0,0.435350757,0.75,144,839,https://www.flickr.com/photos/winnu/31389531466/,https://creativecommons.org/licenses/by/2.0/,Nigel,https://www.flickr.com/people/winnu/\r\nd7ffed8f21024a14,https://c1.staticflickr.com/9/8728/28249502800_8a99581af4_o.jpg,0.25,0.25,0.75,1,752,922,https://www.flickr.com/photos/42220226@N07/28249502800,https://creativecommons.org/licenses/by/2.0/,Sicnag,https://www.flickr.com/people/42220226@N07/\r\n5efeca6c4c233f01,https://c1.staticflickr.com/6/5500/31284538395_f1ef2882f9_o.jpg,0.48619186,0,0.972383721,0.75,761,464,https://www.flickr.com/photos/lge/31284538395,https://creativecommons.org/licenses/by/2.0/,LG\\354\\240\\204\\354\\236\\220,https://www.flickr.com/people/lge/\r\n7d599306068f4594,https://c1.staticflickr.com/8/7715/27614734306_5cd65c4100_o.jpg,0.4375,0.25,1,1,717,514,https://www.flickr.com/photos/142136365@N08/27614734306,https://creativecommons.org/licenses/by/2.0/,Sery Content Development MT,https://www.flickr.com/people/142136365@N08/\r\n67a52d9155241752,https://c1.staticflickr.com/9/8571/27757091954_a8b6697969_o.jpg,0,0.25,0.5625,1,830,420,https://www.flickr.com/photos/raver_mikey/27757091954,https://creativecommons.org/licenses/by/2.0/,Mikey,https://www.flickr.com/people/raver_mikey/\r\ndc12a25cc85ee9cf,https://c1.staticflickr.com/6/5220/29764054890_0309cd8fb8_o.jpg,0,0,0.374521073,0.75,647,896,http://www.flickr.com/photos/vitroids/29764054890/,https://creativecommons.org/licenses/by/2.0/,Masakazu Matsumoto,https://www.flickr.com/people/vitroids/\r\n0d6463e9a7ef6185,https://c1.staticflickr.com/9/8147/28957674191_b862ba9c0c_o.jpg,0,0,0.66226138,1,771,786,https://www.flickr.com/photos/spera-designerschuhe/28957674191,https://creativecommons.org/licenses/by/2.0/,\"SPERA.de Designerschuhe, Taschen und Accessoires\",https://www.flickr.com/people/spera-designerschuhe/\r\naf543b5f19ed1b1d,https://c1.staticflickr.com/2/1659/26658174432_76a1fc6cd7_o.jpg,0.28125,0,0.84375,0.75,40,391,https://www.flickr.com/photos/briangratwicke/26658174432,https://creativecommons.org/licenses/by/2.0/,Brian Gratwicke,https://www.flickr.com/people/briangratwicke/\r\na7151de51c2f1d0b,https://c1.staticflickr.com/9/8622/27804271193_73fe11b408_o.jpg,0.25,0.340254237,1,1,345,214,https://www.flickr.com/photos/jared422/27804271193,https://creativecommons.org/licenses/by/2.0/,Jared,https://www.flickr.com/people/jared422/\r\nc5c9d17672ff88ce,https://c1.staticflickr.com/8/7654/27071795905_8908bd10e9_o.jpg,0.25,0,1,0.561328125,458,266,https://www.flickr.com/photos/140473922@N07/27071795905/,https://creativecommons.org/licenses/by/2.0/,malnasfalu,https://www.flickr.com/people/140473922@N07/\r\n5bdb8268eadf57d8,https://c1.staticflickr.com/6/5765/30333126152_38071946f0_o.jpg,0,0,0.666666667,1,852,968,https://www.flickr.com/photos/37491467@N07/30333126152,https://creativecommons.org/licenses/by/2.0/,Say_No_To_Turtles,https://www.flickr.com/people/37491467@N07/\r\n80177288cda51d95,https://c1.staticflickr.com/1/712/32401234622_f7ec1efbe5_o.jpg,0,0,0.75,1,641,605,https://flickr.com/57868312@N00/32401234622,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\n5b60477a29c7ef81,https://c1.staticflickr.com/1/648/32597389405_80e847a182_o.jpg,0.333333333,0,1,1,416,700,http://www.flickr.com/photos/bexwalton/32597389405/,https://creativecommons.org/licenses/by/2.0/,Bex Walton,https://www.flickr.com/people/bexwalton/\r\nda7e65bc80814f37,https://c1.staticflickr.com/6/5337/30469224383_6aa5c4a26b_o.jpg,0.5,0,1,0.75,788,603,https://www.flickr.com/photos/aloha75/30469224383/,https://creativecommons.org/licenses/by/2.0/,Sam Howzit,https://www.flickr.com/people/aloha75/\r\n4aeb33c9e565a201,https://c1.staticflickr.com/8/7631/27704602521_784d784312_o.jpg,0,0,0.735294118,1,98,840,https://www.flickr.com/photos/7326810@N08/27704602521,https://creativecommons.org/licenses/by/2.0/,Jean,https://www.flickr.com/people/7326810@N08/\r\n19d43f8204d3183d,https://c1.staticflickr.com/8/7100/27097357596_2e5f7f8ddb_o.jpg,0,0,0.75,0.646812081,985,378,https://www.flickr.com/photos/atoach/27097357596,https://creativecommons.org/licenses/by/2.0/,Tim Green,https://www.flickr.com/people/atoach/\r\n2b741c5c635a7295,https://c1.staticflickr.com/9/8870/28112675323_f9aa3a3600_o.jpg,0,0,0.681073025,1,108,634,https://www.flickr.com/photos/flowcomm/28112675323,https://creativecommons.org/licenses/by/2.0/,flowcomm,https://www.flickr.com/people/flowcomm/\r\n398720b8f19d5b84,https://c1.staticflickr.com/1/502/32384871825_3a1708f568_o.jpg,0.5,0.25,1,1,697,700,https://flickr.com/146625745@N08/32384871825,https://creativecommons.org/licenses/by/2.0/,home thods,https://www.flickr.com/people/146625745@N08/\r\nab8329c28d8538c3,https://c1.staticflickr.com/6/5507/31042117711_9b835b691b_o.jpg,0.481981982,0.25,1,1,469,13,https://www.flickr.com/photos/fernando680411/31042117711,https://creativecommons.org/licenses/by/2.0/,Nelson Fernando Sotelo Castro,https://www.flickr.com/people/fernando680411/\r\n348618ddbde3a39c,https://c1.staticflickr.com/9/8604/27892878974_931dc78b34_o.jpg,0,0,0.75,0.535877863,386,460,https://www.flickr.com/photos/bvi4092/27892878974,https://creativecommons.org/licenses/by/2.0/,bvi4092,https://www.flickr.com/people/bvi4092/\r\nd0029b3675b792ae,https://c1.staticflickr.com/9/8739/28789525922_b9ea40444c_o.jpg,0.322236181,0,0.966708543,0.75,790,902,https://www.flickr.com/photos/84906483@N08/28789525922/,https://creativecommons.org/licenses/by/2.0/,spinster cardigan,https://www.flickr.com/people/84906483@N08/\r\n69854b68476d5068,https://c1.staticflickr.com/9/8239/29052269695_ba56b1b270_o.jpg,0.498159057,0,0.996318115,0.75,893,733,https://www.flickr.com/photos/zigazou76/29052269695,https://creativecommons.org/licenses/by/2.0/,Fr\\303\\251d\\303\\251ric BISSON,https://www.flickr.com/people/zigazou76/\r\n4bdb561770b52424,https://c1.staticflickr.com/8/7429/27735562062_14d90e477c_o.jpg,0.25,0.25,0.75,1,147,184,https://www.flickr.com/photos/blachswan/27735562062,https://creativecommons.org/licenses/by/2.0/,Ed Dunens,https://www.flickr.com/people/blachswan/\r\na92ab9e5fb9c5d64,https://c1.staticflickr.com/9/8121/29350360356_21b52b834f_o.jpg,0.500736377,0.25,1,1,629,457,https://www.flickr.com/photos/glynlowe/29350360356,https://creativecommons.org/licenses/by/2.0/,www.GlynLowe.com,https://www.flickr.com/people/glynlowe/\r\n8c6edf1ddfb61102,https://c1.staticflickr.com/8/7310/27095263593_6c217dcf2d_o.jpg,0.503303965,0,1,0.75,889,123,https://www.flickr.com/photos/sybarite48/27095263593,https://creativecommons.org/licenses/by/2.0/,Daniel Jolivet,https://www.flickr.com/people/sybarite48/\r\nc135b139d293f7f2,https://c1.staticflickr.com/6/5511/31483820115_746e4778f1_o.jpg,0.324925816,0,1,1,347,491,https://www.flickr.com/photos/13476480@N07/31483820115,https://creativecommons.org/licenses/by/2.0/,manhhai,https://www.flickr.com/people/13476480@N07/\r\n7d2e9b03ec1b630b,https://c1.staticflickr.com/8/7391/27341486272_663d67f419_o.jpg,0.25,0,0.75,0.75,401,535,https://www.flickr.com/photos/joi/27341486272,https://creativecommons.org/licenses/by/2.0/,Joi Ito,https://www.flickr.com/people/joi/\r\n29fe8d653950a631,https://c1.staticflickr.com/8/7694/26743914583_5a94c79b35_o.jpg,0,0,0.5,0.75,80,542,https://www.flickr.com/photos/quinet/26743914583/,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\nd4c817db6be62b71,https://c1.staticflickr.com/1/611/32446502452_f6b691f7cc_o.jpg,0.375178317,0,1,1,95,164,https://www.flickr.com/photos/tdlucas5000/32446502452,https://creativecommons.org/licenses/by/2.0/,tdlucas5000,https://www.flickr.com/people/tdlucas5000/\r\n1e77361718213a43,https://c1.staticflickr.com/1/606/32230247896_6f7d356e7e_o.jpg,0,0,1,1,440,435,https://flickr.com/73833659@N00/32230247896,https://creativecommons.org/licenses/by/2.0/,MK Feeney,https://www.flickr.com/people/feeney1982/\r\n6711e27733cfd050,https://c1.staticflickr.com/6/5510/31404867966_99715bd367_o.jpg,0.435350757,0.25,0.870701513,1,718,733,https://www.flickr.com/photos/dok1/31404867966,https://creativecommons.org/licenses/by/2.0/,Don O'Brien,https://www.flickr.com/people/dok1/\r\ndf874264566c58a9,https://c1.staticflickr.com/8/7319/27496297684_111a2f5a16_o.jpg,0.25,0.25,1,1,742,152,https://flickr.com/60944636@N00/27496297684,https://creativecommons.org/licenses/by/2.0/,el cajon yacht club,https://www.flickr.com/people/el_cajon_yacht_club/\r\n67c0339b1dfda5f6,https://c1.staticflickr.com/6/5196/29898630576_fceec38d6a_o.jpg,0,0.25,0.499263623,1,558,496,https://www.flickr.com/photos/thejointstaff/29898630576,https://creativecommons.org/licenses/by/2.0/,Chairman of the Joint Chiefs of Staff,https://www.flickr.com/people/thejointstaff/\r\nabd121ef53a69ef8,https://c1.staticflickr.com/8/7724/27965730726_4baf0c6722_o.jpg,0,0,0.658629442,0.75,240,335,https://www.flickr.com/photos/megs_pics/27965730726/,https://creativecommons.org/licenses/by/2.0/,Megan Cole,https://www.flickr.com/people/megs_pics/\r\n5021e7da00bf43b2,https://c1.staticflickr.com/6/5500/31195547582_09c4533c1c_o.jpg,0.25,0.25,1,1,480,937,https://flickr.com/96774554@N03/31195547582,https://creativecommons.org/licenses/by/2.0/,Wheels ON,https://www.flickr.com/people/96774554@N03/\r\nebdfdcba5dc3b72a,https://c1.staticflickr.com/1/422/32445585326_8074f13658_o.jpg,0,0,0.5,0.75,868,600,https://www.flickr.com/photos/greggjerdingen/32445585326/,https://creativecommons.org/licenses/by/2.0/,Greg Gjerdingen,https://www.flickr.com/people/greggjerdingen/\r\n68274b8293271807,https://c1.staticflickr.com/8/7281/27344963241_0feb45e01d_o.jpg,0,0.25,0.499263623,1,834,579,https://www.flickr.com/photos/ministreiodadefesa/27344963241,https://creativecommons.org/licenses/by/2.0/,Minist\\303\\251rio da Defesa,https://www.flickr.com/people/ministreiodadefesa/\r\na84ead15406596ce,https://c1.staticflickr.com/9/8300/29557884001_05493785a6_o.jpg,0,0.25,0.5,1,480,937,https://www.flickr.com/photos/concavowheels/29557884001,https://creativecommons.org/licenses/by/2.0/,CONCAVO WHEELS,https://www.flickr.com/people/concavowheels/\r\n5d1df4748743d42e,https://c1.staticflickr.com/8/7368/27924672426_3fde6b8850_o.jpg,0.25,0.436619718,1,1,964,322,https://www.flickr.com/photos/dalecruse/27924672426,https://creativecommons.org/licenses/by/2.0/,Dale Cruse,https://www.flickr.com/people/dalecruse/\r\n16ff75228179469c,https://c1.staticflickr.com/6/5564/29957754894_f57a1c39f2_o.jpg,0.502573529,0.25,1,1,228,386,https://flickr.com/32234827@N03/29957754894,https://creativecommons.org/licenses/by/2.0/,Dick Thompson,https://www.flickr.com/people/dick-thompson-sandian/\r\n6e840c209f5733d5,https://c1.staticflickr.com/9/8596/28375656583_c817d204c3_o.jpg,0,0,0.75,0.561328125,442,750,https://www.flickr.com/photos/cogdog/28375656583,https://creativecommons.org/licenses/by/2.0/,Alan Levine,https://www.flickr.com/people/cogdog/\r\n16bbe588e0ecc6ab,https://c1.staticflickr.com/8/7078/26813986145_f515146bfa_o.jpg,0.482608696,0,0.965217391,0.75,142,168,https://www.flickr.com/photos/79452129@N02/26813986145,https://creativecommons.org/licenses/by/2.0/,Fyn Kynd Photography,https://www.flickr.com/people/79452129@N02/\r\n46b657df12fd7c26,https://c1.staticflickr.com/8/7129/27258007632_658fac681c_o.jpg,0,0.25,0.5625,1,284,724,https://www.flickr.com/photos/akras/27258007632/,https://creativecommons.org/licenses/by/2.0/,Andrey,https://www.flickr.com/people/akras/\r\nbc7dff0960bb3335,https://c1.staticflickr.com/8/7616/28648009596_6e65ac7ace_o.jpg,0,0.260752688,0.75,1,443,765,https://www.flickr.com/photos/atoach/28648009596/,https://creativecommons.org/licenses/by/2.0/,Tim Green,https://www.flickr.com/people/atoach/\r\n128fca690a11f68c,https://c1.staticflickr.com/8/7369/26875470084_43357ac827_o.jpg,0.250923191,0.25,0.752769572,1,98,785,https://www.flickr.com/photos/quinet/26875470084,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\ned9b2813b9245e52,https://c1.staticflickr.com/9/8770/29809879405_c8cf2a2b31_o.jpg,0,0,0.5,0.75,880,676,https://www.flickr.com/photos/konstantinti/29809879405,https://creativecommons.org/licenses/by/2.0/,Konstantin Tilberg,https://www.flickr.com/people/konstantinti/\r\na187afb98f295057,https://c1.staticflickr.com/1/319/31741319460_db1f31d4b9_o.jpg,0,0,0.5625,0.75,121,755,https://www.flickr.com/photos/jsjgeology/31741319460,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\nd2d2b255935f0aae,https://c1.staticflickr.com/6/5563/30545415915_150b5097e6_o.jpg,0,0,0.75,1,948,432,https://www.flickr.com/photos/torsten-behrens/30545415915/in/photostream/,https://creativecommons.org/licenses/by/2.0/,Torsten Behrens,https://www.flickr.com/people/torsten-behrens/\r\n4d008626b4b68665,https://c1.staticflickr.com/1/296/32309427086_5dcd538057_o.jpg,0.4375,0.25,1,1,108,780,https://flickr.com/21203533@N00/32309427086,https://creativecommons.org/licenses/by/2.0/,Jim G,https://www.flickr.com/people/jimg944/\r\n10bc8784a5eb8843,https://c1.staticflickr.com/6/5582/30724498173_0e082b8ff0_o.jpg,0.488702624,0,0.977405248,0.75,744,210,https://www.flickr.com/photos/13476480@N07/30724498173,https://creativecommons.org/licenses/by/2.0/,manhhai,https://www.flickr.com/people/13476480@N07/\r\n77ac8bf95825dfcc,https://c1.staticflickr.com/8/7259/26898012321_a3265271dd_o.jpg,0.248713235,0,0.746139706,0.75,635,728,https://www.flickr.com/photos/shankaronline/26898012321,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n5c8bcfa0dbb246ab,https://c1.staticflickr.com/8/7728/26215432693_2ee23efc4d_o.jpg,0.234375,0.25,0.703125,1,818,581,https://flickr.com/123760891@N03/26215432693,https://creativecommons.org/licenses/by/2.0/,Automobile Italia,https://www.flickr.com/people/automobileitalia/\r\nafb0233b2e0545c4,https://c1.staticflickr.com/9/8439/29240681086_52945bd07d_o.jpg,0.333333333,0,1,1,581,959,https://www.flickr.com/photos/dalbera/29240681086,https://creativecommons.org/licenses/by/2.0/,Jean-Pierre Dalb\\303\\251ra,https://www.flickr.com/people/dalbera/\r\n1bb5f1337dfb1edd,https://c1.staticflickr.com/1/589/32214674481_5a087370d6_o.jpg,0.25,0.405546624,1,1,641,293,https://flickr.com/13194817@N00/32214674481,https://creativecommons.org/licenses/by/2.0/,F Delventhal,https://www.flickr.com/people/krossbow/\r\na3c4dd88edacb575,https://c1.staticflickr.com/1/561/31438357361_6a749db0f2_o.jpg,0,0.25,0.75,1,400,351,http://www.flickr.com/photos/condominiocriativo/31438357361/,https://creativecommons.org/licenses/by/2.0/,Condom\\303\\255nio Criativo,https://www.flickr.com/people/condominiocriativo/\r\n64ed3f0e8a2dc320,https://c1.staticflickr.com/1/323/31117919653_3c5225af88_o.jpg,0.555632823,0,1,0.75,653,410,https://flickr.com/42310076@N04/31117919653,https://creativecommons.org/licenses/by/2.0/,Chairman of the Joint Chiefs of Staff,https://www.flickr.com/people/thejointstaff/\r\nd25df1a8259d3fdf,https://c1.staticflickr.com/6/5676/30846111770_bc0995d221_o.jpg,0.248826291,0,1,1,161,833,https://www.flickr.com/photos/adactio/30846111770,https://creativecommons.org/licenses/by/2.0/,Jeremy Keith,https://www.flickr.com/people/adactio/\r\n184a160b32ff4c72,https://c1.staticflickr.com/8/7111/26820702803_77558c6aba_o.jpg,0.328216374,0,0.984649123,1,518,717,https://www.flickr.com/photos/paytonc/26820702803,https://creativecommons.org/licenses/by/2.0/,Payton Chung,https://www.flickr.com/people/paytonc/\r\n4f722cab5573a5fb,https://c1.staticflickr.com/8/7567/28064754896_4c6440e3e0_o.jpg,0.28125,0,0.84375,0.75,885,392,https://www.flickr.com/photos/jdlasica/28064754896/,https://creativecommons.org/licenses/by/2.0/,JD Lasica,https://www.flickr.com/people/jdlasica/\r\n1b546695167d3f6e,https://c1.staticflickr.com/6/5768/30206052140_ae3086703e_o.jpg,0.436619718,0,1,0.75,960,837,https://www.flickr.com/photos/kudumomo/30206052140/,https://creativecommons.org/licenses/by/2.0/,momo,https://www.flickr.com/people/kudumomo/\r\n6a940d1bf69f1d10,https://c1.staticflickr.com/9/8122/29786339486_cf9f752823_o.jpg,0,0.187296417,1,1,558,433,https://www.flickr.com/photos/southbeachcars/29786339486/,https://creativecommons.org/licenses/by/2.0/,Phillip Pessar,https://www.flickr.com/people/southbeachcars/\r\nf4902a569fa12171,https://c1.staticflickr.com/8/7448/27323270341_7fc97564c7_o.jpg,0.5,0,1,0.75,950,992,https://flickr.com/141397992@N02/27323270341,https://creativecommons.org/licenses/by/2.0/,Vegan Photo,https://www.flickr.com/people/141397992@N02/\r\n54a6b2cd7f89220e,https://c1.staticflickr.com/2/1462/26144464314_ae66e3b1da_o.jpg,0.28125,0,0.84375,0.75,918,171,https://www.flickr.com/photos/londonmatt/26144464314,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\n7b71925e989ee1d8,https://c1.staticflickr.com/1/348/31544121611_8bf40c6800_o.jpg,0.210873984,0.25,0.632621951,1,934,35,https://www.flickr.com/photos/hile/31544121611,https://creativecommons.org/licenses/by/2.0/,Helen Penjam,https://www.flickr.com/people/hile/\r\neae13c52554a7783,https://c1.staticflickr.com/1/283/31571942652_56aeae1bf8_o.jpg,0.245614035,0.25,0.736842105,1,781,848,https://flickr.com/149503105@N07/31571942652,https://creativecommons.org/licenses/by/2.0/,Ajith Kumar,https://www.flickr.com/people/ajithdauk/\r\nb1e69e9c3d740f4c,https://c1.staticflickr.com/9/8733/28200514302_2f59d09947_o.jpg,0,0.25,0.5625,1,4,833,https://www.flickr.com/photos/noaaphotolib/28200514302,https://creativecommons.org/licenses/by/2.0/,NOAA Photo Library,https://www.flickr.com/people/noaaphotolib/\r\n493e327b171ab0c3,https://c1.staticflickr.com/6/5475/30311997820_5114dcc89a_o.jpg,0,0,0.672592593,1,831,856,https://www.flickr.com/photos/13476480@N07/30311997820,https://creativecommons.org/licenses/by/2.0/,manhhai,https://www.flickr.com/people/13476480@N07/\r\n2efe4916e679e875,https://c1.staticflickr.com/1/166/30669575014_c1861446e4_o.jpg,0.210873984,0,0.632621951,0.75,866,926,https://www.flickr.com/photos/franganillo/30669575014,https://creativecommons.org/licenses/by/2.0/,Jorge Franganillo,https://www.flickr.com/people/franganillo/\r\n588bd41c9aa5a81b,https://c1.staticflickr.com/8/7138/26571751464_519eeae08c_o.jpg,0.25,0,1,0.75,534,154,https://www.flickr.com/photos/annethelibrarian/26571751464,https://creativecommons.org/licenses/by/2.0/,anneheathen,https://www.flickr.com/people/annethelibrarian/\r\n49b1adfeb8ffcabd,https://c1.staticflickr.com/8/7556/27284219884_84e1fbcaeb_o.jpg,0.4375,0,1,0.75,932,671,https://www.flickr.com/photos/55229469@N07/27284219884,https://creativecommons.org/licenses/by/2.0/,Forsaken Fotos,https://www.flickr.com/people/55229469@N07/\r\n225e06c2dfa68a62,https://c1.staticflickr.com/9/8055/28952537693_18d3e06407_o.jpg,0.25,0,0.75,0.75,882,945,https://www.flickr.com/photos/barnimages/28952537693/,https://creativecommons.org/licenses/by/2.0/,barnimages.com,https://www.flickr.com/people/barnimages/\r\nd7d5b34ea0a0ffcf,https://c1.staticflickr.com/9/8066/29898270981_73e0076c26_o.jpg,0.500736377,0.25,1,1,878,970,https://www.flickr.com/photos/presidentialoffice/29898270981,https://creativecommons.org/licenses/by/2.0/,\\347\\270\\275\\347\\265\\261\\345\\272\\234,https://www.flickr.com/people/presidentialoffice/\r\n0212aa8aa912ee45,https://c1.staticflickr.com/6/5466/30266928545_e63bcb8041_o.jpg,0.327407407,0,1,1,327,416,https://www.flickr.com/photos/dany13/30266928545,https://creativecommons.org/licenses/by/2.0/,dany13,https://www.flickr.com/people/dany13/\r\nd7de327bbf412965,https://c1.staticflickr.com/8/7568/28879744260_7f6ee27057_o.jpg,0,0.25,0.556376361,1,490,660,https://www.flickr.com/photos/stevensnodgrass/28879744260,https://creativecommons.org/licenses/by/2.0/,Steve Snodgrass,https://www.flickr.com/people/stevensnodgrass/\r\n2200794c59b1808e,https://c1.staticflickr.com/1/508/32766800345_d32328469f_o.jpg,0.363560732,0,1,0.75,311,214,https://www.flickr.com/photos/gails_pictures/32766800345,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n83896b9fb7e9c2f8,https://c1.staticflickr.com/9/8537/29037936543_0f751d4cdc_o.jpg,0.25,0,0.75,0.75,537,547,https://www.flickr.com/photos/elsie/29037936543/,https://creativecommons.org/licenses/by/2.0/,Les Chatfield,https://www.flickr.com/people/elsie/\r\nf9f36f7774d73820,https://c1.staticflickr.com/3/2013/32659953892_0199b33b51_o.jpg,0,0,0.563380282,0.75,305,723,https://www.flickr.com/photos/treegrow/32659953892,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\n796e23f8dbf9a7a2,https://c1.staticflickr.com/8/7421/27680391126_32219fb627_o.jpg,0,0,0.666666667,1,607,835,https://www.flickr.com/photos/quinet/27680391126,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\ndad40e355e4e8e96,https://c1.staticflickr.com/9/8027/29028347661_d9a3dc8d21_o.jpg,0.25,0,0.75,0.75,655,848,https://www.flickr.com/photos/karen_roe/29028347661,https://creativecommons.org/licenses/by/2.0/,Karen Roe,https://www.flickr.com/people/karen_roe/\r\n064e1cc405062c85,https://c1.staticflickr.com/1/645/31869882565_e6c7e4558a_o.jpg,0,0.25,0.5625,1,108,135,https://www.flickr.com/photos/jimg944/31869882565,https://creativecommons.org/licenses/by/2.0/,Jim G,https://www.flickr.com/people/jimg944/\r\n1484479b7630af0a,https://c1.staticflickr.com/8/7520/27703755495_e5dd0273c4_o.jpg,0.25,0,0.75,0.75,702,904,https://www.flickr.com/photos/7armyjmtc/27703755495,https://creativecommons.org/licenses/by/2.0/,7th Army Training Command,https://www.flickr.com/people/7armyjmtc/\r\n18d5f99fe237a306,https://c1.staticflickr.com/9/8496/29071092183_9562a5b8dc_o.jpg,0.246827411,0,0.740482234,1,815,77,https://www.flickr.com/photos/yamahawatercraft/29071092183/,https://creativecommons.org/licenses/by/2.0/,Yamaha Watercraft Group,https://www.flickr.com/people/yamahawatercraft/\r\nd0c6c29954f0380b,https://c1.staticflickr.com/1/517/32675860980_0c40361cf8_o.jpg,0.25,0,1,0.5,431,96,https://www.flickr.com/photos/tjspix/32675860980,https://creativecommons.org/licenses/by/2.0/,Teresa Johns,https://www.flickr.com/people/tjspix/\r\nc777da68bec01dcd,https://c1.staticflickr.com/9/8562/28193582220_b2580f280c_o.jpg,0.249631811,0.25,0.748895434,1,487,584,https://www.flickr.com/photos/mtaphotos/28193582220,https://creativecommons.org/licenses/by/2.0/,Metropolitan Transportation Authority of the State of New York,https://www.flickr.com/people/mtaphotos/\r\n5cd8864b6e35df82,https://c1.staticflickr.com/8/7105/27793628895_a1d2e5e492_o.jpg,0,0,0.75,1,866,898,https://www.flickr.com/photos/jeepersmedia/27793628895,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n444109f3ccfacc7b,https://c1.staticflickr.com/9/8064/29945301806_292581209b_o.jpg,0.25,0,0.75,0.75,336,221,https://www.flickr.com/photos/amslerpix/29945301806,https://creativecommons.org/licenses/by/2.0/,amslerPIX,https://www.flickr.com/people/amslerpix/\r\nb722c809657aecee,https://c1.staticflickr.com/1/356/30881368874_ed2a887e15_o.jpg,0,0.5,0.75,1,777,640,https://www.flickr.com/photos/eebeephoto/30881368874/,https://creativecommons.org/licenses/by/2.0/,Ed Berger,https://www.flickr.com/people/eebeephoto/\r\na193f7ae7d550a82,https://c1.staticflickr.com/8/7285/27193523532_37b77df506_o.jpg,0.4375,0.25,1,1,738,631,https://www.flickr.com/photos/jeepersmedia/27193523532,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n97025d82ec9860dc,https://c1.staticflickr.com/9/8123/28630695394_9e536782e6_o.jpg,0.264605463,0,0.793816388,0.75,676,574,https://www.flickr.com/photos/jeepersmedia/28630695394,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n23f39187c0d00e30,https://c1.staticflickr.com/6/5581/30083343153_3e32030455_o.jpg,0,0,0.663235294,1,972,658,https://flickr.com/14642507@N08/30083343153,https://creativecommons.org/licenses/by/2.0/,sasastro,https://www.flickr.com/people/sasastro/\r\n515c6586db87bbd8,https://c1.staticflickr.com/6/5590/30456955046_3073b879aa_o.jpg,0,0,0.665684831,1,874,218,https://www.flickr.com/photos/bikashdas/30456955046,https://creativecommons.org/licenses/by/2.0/,Bikash Das,https://www.flickr.com/people/bikashdas/\r\n42e9b41ce96dbf0a,https://c1.staticflickr.com/9/8058/29670008630_bd1bfc6c1a_o.jpg,0,0,0.75,0.5625,851,647,https://www.flickr.com/photos/stellahyc/29670008630,https://creativecommons.org/licenses/by/2.0/,stellahyc910,https://www.flickr.com/people/stellahyc/\r\n126175abfdcb1a21,https://c1.staticflickr.com/1/386/32014236205_a98a5d344a_o.jpg,0.25,0.25,0.75,1,471,486,https://flickr.com/14771153@N04/32014236205,https://creativecommons.org/licenses/by/2.0/,slgckgc,https://www.flickr.com/people/slgc/\r\n4b0c9bf2584b35d3,https://c1.staticflickr.com/9/8394/29489318570_f677a927e0_o.jpg,0,0.25,0.75,1,960,79,https://www.flickr.com/photos/naotakem/29489318570/,https://creativecommons.org/licenses/by/2.0/,Naotake Murayama,https://www.flickr.com/people/naotakem/\r\n19b5bea9a7608a92,https://c1.staticflickr.com/3/2598/32087920803_43b48bab4b_o.jpg,0.28125,0.25,0.84375,1,778,435,https://www.flickr.com/photos/capcase/32087920803,https://creativecommons.org/licenses/by/2.0/,James Case,https://www.flickr.com/people/capcase/\r\n1a94dc47fdd84a95,https://c1.staticflickr.com/6/5761/30269979564_df6e9df309_o.jpg,0,0.277925532,0.75,1,870,830,https://www.flickr.com/photos/ringkong/30269979564,https://creativecommons.org/licenses/by/2.0/,typexnick,https://www.flickr.com/people/ringkong/\r\nc35cd0d5673ec1db,https://c1.staticflickr.com/8/7531/29226165661_4d4c2264c7_o.jpg,0,0.25,0.5625,1,963,367,https://www.flickr.com/photos/stevendepolo/29226165661,https://creativecommons.org/licenses/by/2.0/,Steven Depolo,https://www.flickr.com/people/stevendepolo/\r\nb239b5b3e1642f3b,https://c1.staticflickr.com/6/5494/25330616759_529bdf9b94_o.jpg,0.512008734,0,1,0.75,914,746,https://www.flickr.com/photos/sidelong/25330616759,https://creativecommons.org/licenses/by/2.0/,DaveBleasdale,https://www.flickr.com/people/sidelong/\r\ne22a1c8dab22f256,https://c1.staticflickr.com/8/7453/27575870760_ab7aeb3fe9_o.jpg,0,0.25,0.63875,1,311,11,https://www.flickr.com/photos/gails_pictures/27575870760,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\nd959fd91192d5717,https://c1.staticflickr.com/6/5459/22848076308_16f2d07763_o.jpg,0,0,0.5,0.75,629,596,https://www.flickr.com/photos/prayitnophotography/22848076308,https://creativecommons.org/licenses/by/2.0/,Prayitno / Thank you for (12 millions +) view,https://www.flickr.com/people/prayitnophotography/\r\n692016122b9e07a6,https://c1.staticflickr.com/6/5489/29804286543_9cba22e186_o.jpg,0.25,0.25,0.75,1,960,123,https://www.flickr.com/photos/39419174@N00/29804286543/,https://creativecommons.org/licenses/by/2.0/,danieleloreto,https://www.flickr.com/people/39419174@N00/\r\n8749fd2540a0cb01,https://c1.staticflickr.com/8/7491/26781987183_ff723c7a40_o.jpg,0,0.578822733,0.75,1,833,448,https://www.flickr.com/photos/ankurp/26781987183,https://creativecommons.org/licenses/by/2.0/,Ankur P,https://www.flickr.com/people/ankurp/\r\n8e9e976fc98e3cfa,https://c1.staticflickr.com/8/7702/27224046005_eaa4681fee_o.jpg,0.28125,0.25,0.84375,1,666,740,https://www.flickr.com/photos/65542891@N05/27224046005,https://creativecommons.org/licenses/by/2.0/,bsa rwd 10hp,https://www.flickr.com/people/65542891@N05/\r\nd7c8008bf74f2324,https://c1.staticflickr.com/9/8691/28869458692_b60ba5200d_o.jpg,0.439189189,0,1,1,815,672,https://www.flickr.com/photos/watts_photos/28869458692,https://creativecommons.org/licenses/by/2.0/,C Watts,https://www.flickr.com/people/watts_photos/\r\n330b9cb3cd80c88b,https://c1.staticflickr.com/8/7658/27109056921_f299a6739a_o.jpg,0.280784844,0,0.842354533,1,534,431,https://www.flickr.com/photos/annethelibrarian/27109056921,https://creativecommons.org/licenses/by/2.0/,anneheathen,https://www.flickr.com/people/annethelibrarian/\r\n1ef172b89f4a39cf,https://c1.staticflickr.com/6/5508/31450600235_a463d20bb6_o.jpg,0.333333333,0,1,1,51,536,https://www.flickr.com/photos/52450054@N04/31450600235/,https://creativecommons.org/licenses/by/2.0/,Judy Gallagher,https://www.flickr.com/people/52450054@N04/\r\nb02b553998809e17,https://c1.staticflickr.com/8/7614/26966736165_2a6493858a_o.jpg,0.5,0.25,1,1,704,847,https://www.flickr.com/photos/30845197@N00/26966736165/,https://creativecommons.org/licenses/by/2.0/,TLV and more,https://www.flickr.com/people/30845197@N00/\r\n74b2e35cda75fe27,https://c1.staticflickr.com/1/316/32602465092_4522030924_o.jpg,0,0.25,0.576740506,1,629,466,https://www.flickr.com/photos/archivesnz/32602465092/,https://creativecommons.org/licenses/by/2.0/,Archives New Zealand,https://www.flickr.com/people/archivesnz/\r\nf5a2cf5c66d313fa,https://c1.staticflickr.com/8/7090/27455698002_a951356516_o.jpg,0.25,0,1,1,977,656,https://www.flickr.com/photos/crondeau/27455698002,https://creativecommons.org/licenses/by/2.0/,Christine Rondeau,https://www.flickr.com/people/crondeau/\r\n5ceb9f2e8405b0de,https://c1.staticflickr.com/6/5666/30835439000_59b2871139_o.jpg,0.400242326,0.25,1,1,412,64,https://www.flickr.com/photos/mdgovpics/30835439000/,https://creativecommons.org/licenses/by/2.0/,Maryland GovPics,https://www.flickr.com/people/mdgovpics/\r\n114c98c078c5aab0,https://c1.staticflickr.com/9/8705/28272303681_3f64c45d57_o.jpg,0.25,0,1,0.75,50,108,https://www.flickr.com/photos/austinevan/28272303681,https://creativecommons.org/licenses/by/2.0/,Evan Bench,https://www.flickr.com/people/austinevan/\r\nd741b583a123562b,https://c1.staticflickr.com/9/8728/29612129273_477bf62411_o.jpg,0.254086181,0.25,0.762258544,1,321,792,https://flickr.com/24025807@N06/29612129273,https://creativecommons.org/licenses/by/2.0/,CLUBE TREKKING SANTA MARIA RS BRASIL,https://www.flickr.com/people/hikingsantamaria/\r\n324278ff7f22b4eb,https://c1.staticflickr.com/4/3747/32487584933_83dbbe0cb8_o.jpg,0.25,0.25,1,1,498,303,https://flickr.com/25633870@N00/32487584933,https://creativecommons.org/licenses/by/2.0/,Dave Haygarth,https://www.flickr.com/people/minnellium/\r\n1f4227ce24232867,https://c1.staticflickr.com/8/7229/27970534102_43ed025ff3_o.jpg,0.4375,0,1,0.75,34,942,https://www.flickr.com/photos/hockeyholic/27970534102/,https://creativecommons.org/licenses/by/2.0/,Andy Blackledge,https://www.flickr.com/people/hockeyholic/\r\n5973419efd52f783,https://c1.staticflickr.com/6/5508/31120625660_9594495e7d_o.jpg,0.348396501,0,1,1,625,795,https://www.flickr.com/photos/usnavalwarcollegeri/31120625660,https://creativecommons.org/licenses/by/2.0/,U.S. Naval War College,https://www.flickr.com/people/usnavalwarcollegeri/\r\nfdedd06cdf58084e,https://c1.staticflickr.com/8/7718/27331217545_3b853360be_o.jpg,0,0.200323102,1,1,982,428,https://www.flickr.com/photos/slgc/27331217545,https://creativecommons.org/licenses/by/2.0/,slgckgc,https://www.flickr.com/people/slgc/\r\n911aebf30626ce02,https://c1.staticflickr.com/3/2430/32287607674_b11095a2e5_o.jpg,0.521645022,0.25,1,1,875,731,https://www.flickr.com/photos/didbygraham/32287607674/,https://creativecommons.org/licenses/by/2.0/,Graham Richardson,https://www.flickr.com/people/didbygraham/\r\n55599050147071cc,https://c1.staticflickr.com/8/7412/27512988611_20873a4f90_o.jpg,0.246522694,0.25,0.739568082,1,473,962,https://www.flickr.com/photos/vastateparksstaff/27512988611/,https://creativecommons.org/licenses/by/2.0/,Virginia State Parks,https://www.flickr.com/people/vastateparksstaff/\r\neee6f4bcc8468f77,https://c1.staticflickr.com/1/652/31882765966_b27b4fcf19_o.jpg,0,0,1,0.666666667,964,714,https://www.flickr.com/photos/oikeuttaelaimille/31882765966/,https://creativecommons.org/licenses/by/2.0/,Oikeutta el\\303\\244imille,https://www.flickr.com/people/oikeuttaelaimille/\r\n7f3c20f487834846,https://c1.staticflickr.com/8/7448/26824382232_3b4a0eddcd_o.jpg,0,0,0.678571429,1,408,883,https://www.flickr.com/photos/zongo/26824382232,https://creativecommons.org/licenses/by/2.0/,DAVID HOLT,https://www.flickr.com/people/zongo/\r\n8c8bc30a62a1ee23,https://c1.staticflickr.com/1/431/30989283343_7cae478677_o.jpg,0,0.25,0.713028169,1,304,2,https://www.flickr.com/photos/33398884@N03/30989283343/,https://creativecommons.org/licenses/by/2.0/,Ben Sale,https://www.flickr.com/people/33398884@N03/\r\n7e0fb7f80ffa514b,https://c1.staticflickr.com/1/717/31662466562_b2b1553c3d_o.jpg,0.25,0,1,1,655,405,https://www.flickr.com/photos/didbygraham/31662466562,https://creativecommons.org/licenses/by/2.0/,Graham Richardson,https://www.flickr.com/people/didbygraham/\r\nc02ff39fe5832f4b,https://c1.staticflickr.com/6/5597/30495370923_f689e3460c_o.jpg,0.25,0.25,0.75,1,905,675,https://www.flickr.com/photos/o_0/30495370923/,https://creativecommons.org/licenses/by/2.0/,Guilhem Vellut,https://www.flickr.com/people/o_0/\r\n891fe6baf8896b77,https://c1.staticflickr.com/8/7795/28741196845_ca00f0104f_o.jpg,0.25,0,1,0.75,325,535,https://www.flickr.com/photos/kellymcpherson/28741196845/,https://creativecommons.org/licenses/by/2.0/,Kelly Parker McPherson,https://www.flickr.com/people/kellymcpherson/\r\ne96d52f9fbe3ca0b,https://c1.staticflickr.com/3/2540/32722690932_52ea2c48f3_o.jpg,0,0,0.660792952,1,631,435,https://flickr.com/29971610@N02/32722690932,https://creativecommons.org/licenses/by/2.0/,Esther Max,https://www.flickr.com/people/esthermax/\r\n29279d12ed3e29ff,https://c1.staticflickr.com/1/502/32144969220_9f0703dc8e_o.jpg,0.5,0.25,1,1,458,328,https://www.flickr.com/photos/foreignoffice/32144969220/,https://creativecommons.org/licenses/by/2.0/,Foreign and Commonwealth Office,https://www.flickr.com/people/foreignoffice/\r\nac295a63ac9d2cd7,https://c1.staticflickr.com/8/7292/27671965670_4d9e6a72c8_o.jpg,0.280784844,0,0.842354533,1,693,864,https://www.flickr.com/photos/shoppeolina/27671965670,https://creativecommons.org/licenses/by/2.0/,owlin aolin,https://www.flickr.com/people/shoppeolina/\r\n5d23b04cc45bc37b,https://c1.staticflickr.com/9/8178/29212985530_24749869f4_o.jpg,0.4375,0.25,1,1,518,737,https://www.flickr.com/photos/zongo/29212985530/,https://creativecommons.org/licenses/by/2.0/,DAVID HOLT,https://www.flickr.com/people/zongo/\r\n11c53fe6feb856c6,https://c1.staticflickr.com/8/7162/26707259244_e47c43b2b7_o.jpg,0,0.299878837,0.75,0.899636511,477,766,https://www.flickr.com/photos/edenpictures/26707259244,https://creativecommons.org/licenses/by/2.0/,\"Eden, Janine and Jim\",https://www.flickr.com/people/edenpictures/\r\ncbcb7444cc827812,https://c1.staticflickr.com/8/7549/26276389703_ecd645cc3e_o.jpg,0.333333333,0,1,1,763,786,https://www.flickr.com/photos/agenciabrasilia/26276389703,https://creativecommons.org/licenses/by/2.0/,Ag\\303\\252ncia Bras\\303\\255lia,https://www.flickr.com/people/agenciabrasilia/\r\nfe1314dd6f75551d,https://c1.staticflickr.com/8/7144/26370391824_3159ca42b2_o.jpg,0,0.25,0.75,1,921,565,https://www.flickr.com/photos/brendanadkins/26370391824,https://creativecommons.org/licenses/by/2.0/,Brendan Adkins,https://www.flickr.com/people/brendanadkins/\r\na8fca010774a1187,https://c1.staticflickr.com/1/425/31809450395_45b02082bb_o.jpg,0.4375,0,1,0.75,829,810,https://www.flickr.com/photos/fdecomite/31809450395/,https://creativecommons.org/licenses/by/2.0/,fdecomite,https://www.flickr.com/people/fdecomite/\r\n5ec826683757ffae,https://c1.staticflickr.com/6/5731/30742412415_c35c2ff310_o.jpg,0,0,0.599757674,0.75,23,369,https://www.flickr.com/photos/andymorffew/30742412415,https://creativecommons.org/licenses/by/2.0/,Andy Morffew,https://www.flickr.com/people/andymorffew/\r\nb9310b9080c06e4f,https://c1.staticflickr.com/8/7398/27330948851_cd9be2c204_o.jpg,0.25,0.318416523,1,1,867,269,https://www.flickr.com/photos/60900612@N08/27330948851,https://creativecommons.org/licenses/by/2.0/,Thomas's Pics,https://www.flickr.com/people/60900612@N08/\r\n4f9f0b75f5b04524,https://c1.staticflickr.com/8/7587/26933811080_cf8040335a_o.jpg,0,0.25,0.5,1,58,897,https://www.flickr.com/photos/gregthebusker/26933811080,https://creativecommons.org/licenses/by/2.0/,Greg Schechter,https://www.flickr.com/people/gregthebusker/\r\n4abadb721e74b99b,https://c1.staticflickr.com/8/7612/26811635786_e5a333566f_o.jpg,0,0,0.666666667,1,945,781,https://www.flickr.com/photos/mwf2005/26811635786/,https://creativecommons.org/licenses/by/2.0/,muffinn,https://www.flickr.com/people/mwf2005/\r\na3c651cf20ee2d34,https://c1.staticflickr.com/8/7710/26294912563_ba1be3ec95_o.jpg,0,0,0.75,0.75,641,561,https://www.flickr.com/photos/49889874@N05/26294912563,https://creativecommons.org/licenses/by/2.0/,marc falardeau,https://www.flickr.com/people/49889874@N05/\r\n5c4eec952ca114a1,https://c1.staticflickr.com/1/528/31431409824_729b5c64d9_o.jpg,0.25,0.25,0.75,1,348,525,https://flickr.com/51986662@N05/31431409824,https://creativecommons.org/licenses/by/2.0/,USFWS Mountain-Prairie,https://www.flickr.com/people/usfwsmtnprairie/\r\n7f045354f4577c16,https://c1.staticflickr.com/9/8268/30074145430_c1afafabc2_o.jpg,0.25,0,1,0.75,558,970,https://www.flickr.com/photos/rodrigoparedes/30074145430,https://creativecommons.org/licenses/by/2.0/,Rodrigo Paredes,https://www.flickr.com/people/rodrigoparedes/\r\n53211f8a1b131de6,https://c1.staticflickr.com/9/8732/28580034102_984ce9d944_o.jpg,0.220095694,0,1,1,946,578,https://www.flickr.com/photos/elzey/28580034102,https://creativecommons.org/licenses/by/2.0/,Richard Elzey,https://www.flickr.com/people/elzey/\r\n9c0e25e2f72438e8,https://c1.staticflickr.com/6/5607/31102907732_def8083c3f_o.jpg,0,0,0.747269891,1,512,506,https://www.flickr.com/photos/andreboeni/31102907732,https://creativecommons.org/licenses/by/2.0/,Andrew Bone,https://www.flickr.com/people/andreboeni/\r\n55512a6c2a45e2e8,https://c1.staticflickr.com/8/7340/28096885425_4c07a3cf83_o.jpg,0,0.320512821,0.75,0.961538462,928,387,https://www.flickr.com/photos/uwlideas/28096885425/,https://creativecommons.org/licenses/by/2.0/,Med Coolman,https://www.flickr.com/people/uwlideas/\r\n869adfbad4e38261,https://c1.staticflickr.com/6/5697/30241927150_b62ec0df8a_o.jpg,0.25,0,1,0.75,592,866,https://www.flickr.com/photos/sonyanews/30241927150/,https://creativecommons.org/licenses/by/2.0/,sonyanews,https://www.flickr.com/people/sonyanews/\r\nadcf48b42266309d,https://c1.staticflickr.com/9/8057/29233160570_87ebbbc643_o.jpg,0.248713235,0.25,0.746139706,1,505,589,https://www.flickr.com/photos/mdpettitt/29233160570,https://creativecommons.org/licenses/by/2.0/,Martin Pettitt,https://www.flickr.com/people/mdpettitt/\r\n3b67bec3793efcb4,https://c1.staticflickr.com/6/5682/30920421081_33e27f1675_o.jpg,0.548319328,0,1,0.75,822,746,https://www.flickr.com/photos/gaby1/30920421081,https://creativecommons.org/licenses/by/2.0/,Gabriel Gonz\\303\\241lez,https://www.flickr.com/people/gaby1/\r\n67a08d685abf7fb6,https://c1.staticflickr.com/8/7793/27924211143_79cef8b65a_o.jpg,0.25,0,0.75,0.75,739,710,https://www.flickr.com/photos/blondinrikard/27924211143,https://creativecommons.org/licenses/by/2.0/,Blondinrikard Fr\\303\\266berg,https://www.flickr.com/people/blondinrikard/\r\n22a313cc9fc2b4f1,https://c1.staticflickr.com/9/8612/28340861743_d50c5784e8_o.jpg,0.248826291,0,1,1,310,940,https://www.flickr.com/photos/pamas/28340861743,https://creativecommons.org/licenses/by/2.0/,Esin \\303\\234st\\303\\274n,https://www.flickr.com/people/pamas/\r\n0d85e0103989c850,https://c1.staticflickr.com/9/8124/28994100466_472d225974_o.jpg,0.272727273,0,1,1,324,722,https://www.flickr.com/photos/usfwsmtnprairie/28994100466,https://creativecommons.org/licenses/by/2.0/,USFWS Mountain-Prairie,https://www.flickr.com/people/usfwsmtnprairie/\r\nc3cf76709c87255b,https://c1.staticflickr.com/9/8340/29178027716_ed24cc96f2_o.jpg,0,0,0.75,0.75,676,224,https://www.flickr.com/photos/soulhammer/29178027716/,https://creativecommons.org/licenses/by/2.0/,Nathaniel Bruno,https://www.flickr.com/people/soulhammer/\r\n74355a8c878b6814,https://c1.staticflickr.com/1/408/31898209835_867bd3b167_o.jpg,0,0,0.75,1,875,622,https://flickr.com/69203347@N03/31898209835,https://creativecommons.org/licenses/by/2.0/,AmigoDeBusesEMT3,https://www.flickr.com/people/69203347@N03/\r\n6efed50f84ec5e01,https://c1.staticflickr.com/6/5168/29928350415_8935e09ba7_o.jpg,0.5,0,1,0.75,947,602,https://www.flickr.com/photos/lsmith2010/29928350415,https://creativecommons.org/licenses/by/2.0/,Larry Smith,https://www.flickr.com/people/lsmith2010/\r\nd6887e8d19a06139,https://c1.staticflickr.com/6/5753/30992363516_d62f652ef9_o.jpg,0.25,0,1,1,563,45,https://www.flickr.com/photos/rbowen/30992363516/,https://creativecommons.org/licenses/by/2.0/,Rich Bowen,https://www.flickr.com/people/rbowen/\r\n21fef4d9b3dce49a,https://c1.staticflickr.com/9/8126/28295230322_8c45fd586e_o.jpg,0,0.25,0.411096257,1,975,460,https://www.flickr.com/photos/jsjgeology/28295230322,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n0d65cb9d26de4d37,https://c1.staticflickr.com/8/7605/28137606406_e04b692957_o.jpg,0.25,0,0.75,0.75,571,336,https://www.flickr.com/photos/chrisdag/28137606406,https://creativecommons.org/licenses/by/2.0/,ChrisDag,https://www.flickr.com/people/chrisdag/\r\n3df888fc3b1df464,https://c1.staticflickr.com/8/7574/26519073574_355cd52e23_o.jpg,0.25,0,1,0.75,955,383,https://www.flickr.com/photos/vincentteeuwen/26519073574,https://creativecommons.org/licenses/by/2.0/,Vincent Teeuwen,https://www.flickr.com/people/vincentteeuwen/\r\n5ab3f1edd2f0cb06,https://c1.staticflickr.com/8/7716/27804279183_27b9180bda_o.jpg,0,0.079722704,1,1,387,116,https://www.flickr.com/photos/jared422/27804279183,https://creativecommons.org/licenses/by/2.0/,Jared,https://www.flickr.com/people/jared422/\r\n51b390bb124a6d9d,https://c1.staticflickr.com/9/8037/29219102264_ded8d317fa_o.jpg,0,0,0.75,0.657142857,968,422,https://www.flickr.com/photos/141897087@N06/29219102264,https://creativecommons.org/licenses/by/2.0/,JV Santore,https://www.flickr.com/people/141897087@N06/\r\n5bf502c4a4b7d04d,https://c1.staticflickr.com/6/5590/31190125660_b80ecdcb10_o.jpg,0.5,0,1,0.75,10,350,https://flickr.com/65847397@N05/31190125660,https://creativecommons.org/licenses/by/2.0/,Paul Balfe,https://www.flickr.com/people/paul_e_balfe/\r\n6ad1230d612d7eb7,https://c1.staticflickr.com/3/2413/31980917214_4992777b36_o.jpg,0.25,0.498159057,1,0.996318115,438,680,https://www.flickr.com/photos/davehamster/31980917214/,https://creativecommons.org/licenses/by/2.0/,David Merrett,https://www.flickr.com/people/davehamster/\r\n76a5715019fe11c7,https://c1.staticflickr.com/9/8223/28686786171_cee3ce47f5_o.jpg,0,0,0.690690691,1,821,768,https://www.flickr.com/photos/grassrootsgroundswell/28686786171,https://creativecommons.org/licenses/by/2.0/,grassrootsgroundswell,https://www.flickr.com/people/grassrootsgroundswell/\r\n207fd3b5def218d1,https://c1.staticflickr.com/8/7223/27019150086_7ba8c473c7_o.jpg,0.25,0.4375,1,1,991,279,https://www.flickr.com/photos/ndrwfgg/27019150086,https://creativecommons.org/licenses/by/2.0/,Andy / Andrew Fogg,https://www.flickr.com/people/ndrwfgg/\r\nad43e858c614db4d,https://c1.staticflickr.com/9/8263/28394550932_d93637d26b_o.jpg,0,0.25,0.5625,1,968,71,https://flickr.com/7831824@N04/28394550932,https://creativecommons.org/licenses/by/2.0/,Bex Walton,https://www.flickr.com/people/bexwalton/\r\nb3ba6f0352a29b2d,https://c1.staticflickr.com/6/5459/30102051053_1c9c298a3c_o.jpg,0,0,0.498159057,0.75,752,563,https://www.flickr.com/photos/davehamster/30102051053/,https://creativecommons.org/licenses/by/2.0/,David Merrett,https://www.flickr.com/people/davehamster/\r\n95185508805329aa,https://c1.staticflickr.com/9/8354/29365580064_0e6908f14d_o.jpg,0.299878837,0.25,0.899636511,1,653,721,https://www.flickr.com/photos/ocarchives/29365580064,https://creativecommons.org/licenses/by/2.0/,Orange County Archives,https://www.flickr.com/people/ocarchives/\r\n73ee2cc19ef92e50,https://c1.staticflickr.com/8/7189/27594432472_dfaf365016_o.jpg,0,0.25,0.501846381,1,982,686,https://www.flickr.com/photos/jorge-11/27594432472,https://creativecommons.org/licenses/by/2.0/,George M. Groutas,https://www.flickr.com/people/jorge-11/\r\nbe1c43056c5d0085,https://c1.staticflickr.com/6/5525/30762478116_b134a3ee51_o.jpg,0.28125,0,0.84375,0.75,145,436,https://www.flickr.com/photos/usfwsmtnprairie/30762478116,https://creativecommons.org/licenses/by/2.0/,USFWS Mountain-Prairie,https://www.flickr.com/people/usfwsmtnprairie/\r\n84a5f6ee2af29208,https://c1.staticflickr.com/8/7421/26497262564_8dc1a3b161_o.jpg,0.331617647,0,0.994852941,1,321,513,https://www.flickr.com/photos/ilouque/26497262564/,https://creativecommons.org/licenses/by/2.0/,ilouque,https://www.flickr.com/people/ilouque/\r\n50b9e11ae686ae32,https://c1.staticflickr.com/6/5591/31118494856_52df471331_o.jpg,0,0,0.991007194,1,321,34,https://www.flickr.com/photos/gails_pictures/31118494856,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\nfe2edb6cad72d2d2,https://c1.staticflickr.com/8/7468/26505971443_c1368c14e0_o.jpg,0.25,0,1,1,918,488,https://www.flickr.com/photos/jeepersmedia/26505971443,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n58420b727fcab02b,https://c1.staticflickr.com/6/5768/30823764900_d486a3f904_o.jpg,0,0,0.714503817,1,138,168,https://flickr.com/143049688@N08/30823764900,https://creativecommons.org/licenses/by/2.0/,MR MAO PICS,https://www.flickr.com/people/143049688@N08/\r\n47b087263b33728b,https://c1.staticflickr.com/9/8228/29000044324_0eb54060d7_o.jpg,0.298790323,0.25,0.896370968,1,318,152,https://flickr.com/33398884@N03/29000044324,https://creativecommons.org/licenses/by/2.0/,Ben Sale,https://www.flickr.com/people/33398884@N03/\r\n0e0299d6352a3fb8,https://c1.staticflickr.com/1/636/31002720224_411dc0a03f_o.jpg,0.25,0.25,0.75,1,539,4,https://flickr.com/22539273@N00/31002720224,https://creativecommons.org/licenses/by/2.0/,Guilhem Vellut,https://www.flickr.com/people/o_0/\r\n2ee4c1b2b9bcc6c8,https://c1.staticflickr.com/9/8354/29761961012_02ecff90a4_o.jpg,0.25,0,0.75,0.75,763,785,https://www.flickr.com/photos/deltanewshub/29761961012,https://creativecommons.org/licenses/by/2.0/,Delta News Hub,https://www.flickr.com/people/deltanewshub/\r\n24312de5c34473c0,https://c1.staticflickr.com/6/5584/29688881533_e0c64dc987_o.jpg,0.25,0.438671875,1,1,874,820,https://www.flickr.com/photos/marsupilami92/29688881533/,https://creativecommons.org/licenses/by/2.0/,patrick janicek,https://www.flickr.com/people/marsupilami92/\r\nbb748281da5e8681,https://c1.staticflickr.com/6/5281/29946906871_845417f8ac_o.jpg,0.210588633,0,0.6317659,0.75,348,694,https://www.flickr.com/photos/39090590@N02/29946906871,https://creativecommons.org/licenses/by/2.0/,Cyril Fluck,https://www.flickr.com/people/39090590@N02/\r\nb648e8dfcb4fbc99,https://c1.staticflickr.com/1/679/31918129032_23e9a7057b_o.jpg,0,0,0.666666667,1,982,572,https://flickr.com/148488209@N08/31918129032,https://creativecommons.org/licenses/by/2.0/,makaiyla willis,https://www.flickr.com/people/makaiylaw/\r\nd2cc2eabfc7bf54a,https://c1.staticflickr.com/6/5509/31244583052_a8416401f1_o.jpg,0.25,0,0.75,0.75,556,721,https://www.flickr.com/photos/rpavich/31244583052,https://creativecommons.org/licenses/by/2.0/,rpavich,https://www.flickr.com/people/rpavich/\r\nda4b3e38b069d735,https://c1.staticflickr.com/9/8206/28992945534_a0fcc6e623_o.jpg,0.25,0.25,1,1,777,691,https://www.flickr.com/photos/wwworks/28992945534,https://creativecommons.org/licenses/by/2.0/,woodleywonderworks,https://www.flickr.com/people/wwworks/\r\nb0bb1ea9bcc45800,https://c1.staticflickr.com/9/8433/28813586781_49bed3bfb8_o.jpg,0,0,0.5625,0.75,845,456,https://www.flickr.com/photos/renaissancechambara/28813586781/,https://creativecommons.org/licenses/by/2.0/,Ged Carroll,https://www.flickr.com/people/renaissancechambara/\r\nac668c13ccfd4b79,https://c1.staticflickr.com/1/538/31770839465_f1c3beb91e_o.jpg,0,0,0.666666667,1,598,751,https://www.flickr.com/photos/quinet/31770839465,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\nb6c6a5b7829fbc9b,https://c1.staticflickr.com/9/8290/28794830561_2c903f75c1_o.jpg,0.336764706,0,1,1,473,506,https://www.flickr.com/photos/shankaronline/28794830561,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n014f0024918a7b6f,https://c1.staticflickr.com/8/7564/27345252166_6948b71ff3_o.jpg,0.28125,0,0.84375,0.75,10,885,https://www.flickr.com/photos/amanderson/27345252166,https://creativecommons.org/licenses/by/2.0/,amanderson2,https://www.flickr.com/people/amanderson/\r\n93114736f072d6d3,https://c1.staticflickr.com/8/7396/26691839134_a072ec1162_o.jpg,0.25,0,0.75,0.75,617,763,https://www.flickr.com/photos/cogdog/26691839134,https://creativecommons.org/licenses/by/2.0/,Alan Levine,https://www.flickr.com/people/cogdog/\r\nc9ddfb1043177de8,https://c1.staticflickr.com/6/5517/31689316216_b954ab5538_o.jpg,0.5,0.25,1,1,975,117,http://www.flickr.com/photos/richardmoross/31689316216/,https://creativecommons.org/licenses/by/2.0/,Richard Moross,https://www.flickr.com/people/richardmoross/\r\n7e4c41563370bed1,https://c1.staticflickr.com/9/8497/29201470542_e661660313_o.jpg,0,0,0.666666667,1,719,945,https://www.flickr.com/photos/herdie/29201470542,https://creativecommons.org/licenses/by/2.0/,herdiephoto,https://www.flickr.com/people/herdie/\r\n328d99215b19c8c7,https://c1.staticflickr.com/1/662/32533031665_e6c123d8eb_o.jpg,0.5,0,1,0.75,806,926,https://www.flickr.com/photos/30478819@N08/32533031665,https://creativecommons.org/licenses/by/2.0/,Marco Verch,https://www.flickr.com/people/30478819@N08/\r\n2a1a60df9a58bb7c,https://c1.staticflickr.com/8/7780/26705680280_a7f5f24425_o.jpg,0,0.333333333,1,1,650,19,https://www.flickr.com/photos/travelnevada/26705680280,https://creativecommons.org/licenses/by/2.0/,Travel Nevada,https://www.flickr.com/people/travelnevada/\r\n263c7bcd59502625,https://c1.staticflickr.com/9/8092/29347734850_29561b6ec6_o.jpg,0.5,0,1,0.75,480,651,https://www.flickr.com/photos/concavowheels/29347734850,https://creativecommons.org/licenses/by/2.0/,CONCAVO WHEELS,https://www.flickr.com/people/concavowheels/\r\nd05076189b7fd185,https://c1.staticflickr.com/1/727/32381120905_712248f38a_o.jpg,0.25,0,1,0.646812081,708,765,https://www.flickr.com/photos/jdlasica/32381120905,https://creativecommons.org/licenses/by/2.0/,JD Lasica,https://www.flickr.com/people/jdlasica/\r\nfe193683db61ac2d,https://c1.staticflickr.com/6/5669/30159267186_7663a5ba9c_o.jpg,0.4375,0,1,0.75,744,1,https://www.flickr.com/photos/meemainseen/30159267186,https://creativecommons.org/licenses/by/2.0/,Muhammad Imran Saeed,https://www.flickr.com/people/meemainseen/\r\nd8b0101e2459882a,https://c1.staticflickr.com/8/7451/27873714556_8452970117_o.jpg,0,0.25,0.5,1,655,88,https://www.flickr.com/photos/impulse_2016/27873714556/,https://creativecommons.org/licenses/by/2.0/,Impulse Theater Festival 2016,https://www.flickr.com/people/impulse_2016/\r\n3841b33c9797cbf1,https://c1.staticflickr.com/6/5563/30857323591_f9687379e7_o.jpg,0,0,0.5625,0.75,909,953,https://www.flickr.com/photos/brownpau/30857323591,https://creativecommons.org/licenses/by/2.0/,brownpau,https://www.flickr.com/people/brownpau/\r\n232c22314e21f764,https://c1.staticflickr.com/6/5505/30587416664_22b14394ec_o.jpg,0.560165975,0,1,0.75,115,904,https://www.flickr.com/photos/gails_pictures/30587416664,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n30b62313af2b5d74,https://c1.staticflickr.com/8/7570/30228833946_9b000a7dbd_o.jpg,0,0,0.561328125,0.75,316,795,https://www.flickr.com/photos/pamas/30228833946,https://creativecommons.org/licenses/by/2.0/,Esin \\303\\234st\\303\\274n,https://www.flickr.com/people/pamas/\r\n2c54b60a2851b8ed,https://c1.staticflickr.com/9/8523/29243698212_6d465d0527_o.jpg,0,0,0.5625,0.75,657,577,https://www.flickr.com/photos/zombieite/29243698212,https://creativecommons.org/licenses/by/2.0/,zombieite,https://www.flickr.com/people/zombieite/\r\nc96c3eab644075d3,https://c1.staticflickr.com/8/7790/26778459800_6328b3ac5d_o.jpg,0,0,0.932055749,1,348,300,https://www.flickr.com/photos/lac-bac/26778459800,https://creativecommons.org/licenses/by/2.0/,BiblioArchives / LibraryArchives,https://www.flickr.com/people/lac-bac/\r\n060dfe239abfd547,https://c1.staticflickr.com/8/7475/28717385362_9e5d05dfe7_o.jpg,0.473457676,0.25,0.946915352,1,704,396,https://www.flickr.com/photos/alan-light/28717385362/,https://creativecommons.org/licenses/by/2.0/,Alan Light,https://www.flickr.com/people/alan-light/\r\n0bd5123017590e39,https://c1.staticflickr.com/6/5553/30409656042_0a59691936_o.jpg,0.4375,0.25,1,1,566,430,https://www.flickr.com/photos/davidwilson1949/30409656042,https://creativecommons.org/licenses/by/2.0/,David Wilson,https://www.flickr.com/people/davidwilson1949/\r\nbd3617fcc985fe31,https://c1.staticflickr.com/6/5624/30871247636_63a2a6a4fa_o.jpg,0,0.25,0.75,0.75,833,61,https://www.flickr.com/photos/xiquinho/30871247636/,https://creativecommons.org/licenses/by/2.0/,xiquinhosilva,https://www.flickr.com/people/xiquinho/\r\n54666421b4a6998f,https://c1.staticflickr.com/8/7701/26783338114_ced2a4e55f_o.jpg,0.5,0,1,0.75,920,759,https://www.flickr.com/photos/zeevveez/26783338114,https://creativecommons.org/licenses/by/2.0/,zeevveez,https://www.flickr.com/people/zeevveez/\r\n4e34df87968bd882,https://c1.staticflickr.com/1/621/31058803964_6d583c4eea_o.jpg,0.438671875,0.25,1,1,339,379,https://flickr.com/30139045@N00/31058803964,https://creativecommons.org/licenses/by/2.0/,Tim Walker,https://www.flickr.com/people/timjoyfamily/\r\n3f48b1e74589d310,https://c1.staticflickr.com/9/8038/28644165984_de12eb6b3e_o.jpg,0,0,0.666666667,1,653,508,https://www.flickr.com/photos/secdef/28644165984,https://creativecommons.org/licenses/by/2.0/,Jim Mattis,https://www.flickr.com/people/secdef/\r\n96e77c20bf94ecb1,https://c1.staticflickr.com/9/8342/28986701940_8aa08f43c6_o.jpg,0.4108,0.25,1,1,818,518,https://www.flickr.com/photos/autohistorian/28986701940,https://creativecommons.org/licenses/by/2.0/,Alden Jewell,https://www.flickr.com/people/autohistorian/\r\n4edeacef9fb8ec90,https://c1.staticflickr.com/6/5579/31278004155_6f6d083d49_o.jpg,0,0,0.5625,0.75,719,997,https://www.flickr.com/photos/cambodia4kidsorg/31278004155/,https://creativecommons.org/licenses/by/2.0/,Cambodia4kids.org Beth Kanter,https://www.flickr.com/people/cambodia4kidsorg/\r\nc9302f3b2b0452e4,https://c1.staticflickr.com/8/7413/27864002812_f3fc0a7878_o.jpg,0,0,0.875,1,884,892,https://www.flickr.com/photos/hisgett/27864002812?rb=1,https://creativecommons.org/licenses/by/2.0/,Tony Hisgett,https://www.flickr.com/people/hisgett/\r\n0af0a5dfee6b84ff,https://c1.staticflickr.com/8/7382/26663873503_fc70ec8c08_o.jpg,0.280226209,0,0.840678627,0.75,484,808,https://www.flickr.com/photos/65344061@N06/26663873503/,https://creativecommons.org/licenses/by/2.0/,Andrew Bone,https://www.flickr.com/people/andreboeni/\r\n02dbe78a73db2808,https://c1.staticflickr.com/9/8089/27862798673_b874427662_o.jpg,0.369771863,0.25,0.739543726,1,975,255,https://flickr.com/47445767@N05/27862798673,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n592c7424f13b2d29,https://c1.staticflickr.com/9/8639/28634730405_a12dd81623_o.jpg,0.419316375,0.25,1,1,604,517,https://www.flickr.com/photos/lac-bac/28634730405,https://creativecommons.org/licenses/by/2.0/,BiblioArchives / LibraryArchives,https://www.flickr.com/people/lac-bac/\r\nf5b0ed0fbb3b5f74,https://c1.staticflickr.com/1/131/30836327244_86e3f6cee9_o.jpg,0.5,0.25,1,1,567,645,https://www.flickr.com/photos/nationalassemblyforwales/30836327244,https://creativecommons.org/licenses/by/2.0/,National Assembly for Wales,https://www.flickr.com/people/nationalassemblyforwales/\r\n1b1ffd7000d0749f,https://c1.staticflickr.com/6/5763/30233213804_63bd4007fd_o.jpg,0,0,0.75,0.75,960,991,https://www.flickr.com/photos/ilove9and23/30233213804/,https://creativecommons.org/licenses/by/2.0/,Debbie Tingzon,https://www.flickr.com/people/ilove9and23/\r\n70a2529a6b2c0d46,https://c1.staticflickr.com/9/8631/29164479153_b29b11670b_o.jpg,0.5,0.25,1,1,580,550,https://www.flickr.com/photos/usarmyband/29164479153/,https://creativecommons.org/licenses/by/2.0/,The United States Army Band,https://www.flickr.com/people/usarmyband/\r\nbba5c56cc3eb07c9,https://c1.staticflickr.com/1/546/31014776813_b295235e55_o.jpg,0.267938931,0.25,0.803816794,1,921,594,https://www.flickr.com/photos/thomasrousing/31014776813,https://creativecommons.org/licenses/by/2.0/,Thomas Rousing,https://www.flickr.com/people/thomasrousing/\r\n000b7d55b6184b08,https://c1.staticflickr.com/8/7450/27510795501_46c2084b6c_o.jpg,0,0,0.5625,0.75,389,963,https://www.flickr.com/photos/40205555@N06/27510795501,https://creativecommons.org/licenses/by/2.0/,mrho101,https://www.flickr.com/people/40205555@N06/\r\n8a02b6196885c394,https://c1.staticflickr.com/8/7296/27350770975_3f596317b8_o.jpg,0,0,1,1,760,447,https://www.flickr.com/photos/graficmedia/27350770975,https://creativecommons.org/licenses/by/2.0/,[-_-] JORGE,https://www.flickr.com/people/graficmedia/\r\n582663fe666aac9c,https://c1.staticflickr.com/6/5592/29610060673_fe4f7b8d01_o.jpg,0,0,0.56375,0.75,923,928,https://www.flickr.com/photos/kelvynskee/29610060673,https://creativecommons.org/licenses/by/2.0/,Kelvyn Skee,https://www.flickr.com/people/kelvynskee/\r\nf6515f31b750a573,https://c1.staticflickr.com/1/719/32474210662_fa858156a3_o.jpg,0,0,0.666666667,1,477,723,https://www.flickr.com/photos/quinet/32474210662,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\nacd99e9b7128395a,https://c1.staticflickr.com/1/445/32794739211_d6f5324607_o.jpg,0,0,0.75,1,859,753,http://www.flickr.com/photos/callmewhatever/32794739211/,https://creativecommons.org/licenses/by/2.0/,Daniel Zimmermann,https://www.flickr.com/people/callmewhatever/\r\n83d35591dfff99ac,https://c1.staticflickr.com/8/7232/27285496573_d5bdd3aa87_o.jpg,0,0,0.75,1,581,857,https://www.flickr.com/photos/juggernautco/27285496573,https://creativecommons.org/licenses/by/2.0/,Daniel X. O'Neil,https://www.flickr.com/people/juggernautco/\r\n3dfbdd066af9ca1e,https://c1.staticflickr.com/9/8816/28135299942_b870fd0ab6_o.jpg,0.25,0,0.75,0.75,821,5,https://www.flickr.com/photos/ben_salter/28135299942/,https://creativecommons.org/licenses/by/2.0/,Ben Salter,https://www.flickr.com/people/ben_salter/\r\n59204d6bb0c234eb,https://c1.staticflickr.com/9/8752/28207273334_3d05b11ea6_o.jpg,0,0.421177267,0.75,0.842354533,418,703,https://www.flickr.com/photos/oatsy40/28207273334/,https://creativecommons.org/licenses/by/2.0/,oatsy40,https://www.flickr.com/people/oatsy40/\r\nd740663eeccdcecd,https://c1.staticflickr.com/9/8130/29191810554_9f08be9c5f_o.jpg,0,0,0.75,0.5625,438,762,https://www.flickr.com/photos/simon2207/29191810554/,https://creativecommons.org/licenses/by/2.0/,Simon220771,https://www.flickr.com/people/simon2207/\r\n9226b783cfb770bd,https://c1.staticflickr.com/9/8637/28516867321_5b709d78c7_o.jpg,0,0.25,0.5625,1,826,34,https://www.flickr.com/photos/jsjgeology/28516867321,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n01dd15caa1b2c7b4,https://c1.staticflickr.com/8/7767/26780389003_e95a6d7cee_o.jpg,0,0.25,0.561328125,1,304,551,https://www.flickr.com/photos/treegrow/26780389003,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\nb97f4b95ae696316,https://c1.staticflickr.com/8/7186/26690393874_7d7f8dafc9_o.jpg,0,0.333333333,1,1,518,399,http://www.flickr.com/photos/n28307/26690393874/,https://creativecommons.org/licenses/by/2.0/,Glenn Beltz,https://www.flickr.com/people/n28307/\r\nc42c88890b5fc86d,https://c1.staticflickr.com/6/5755/30476251310_f45f8b81b7_o.jpg,0.4375,0.25,1,1,671,112,https://www.flickr.com/photos/paulmiller/30476251310/,https://creativecommons.org/licenses/by/2.0/,Paul Miller,https://www.flickr.com/people/paulmiller/\r\nd158837db31ed892,https://c1.staticflickr.com/1/445/31796400213_c6e650f7d5_o.jpg,0.5,0,1,0.75,519,804,https://www.flickr.com/photos/box_repsol/31796400213,https://creativecommons.org/licenses/by/2.0/,Box Repsol,https://www.flickr.com/people/box_repsol/\r\n4a1a40a116088962,https://c1.staticflickr.com/9/8587/28162416971_8e105a5657_o.jpg,0,0,0.75,1,598,624,https://www.flickr.com/photos/gibsongunleather/28162416971,https://creativecommons.org/licenses/by/2.0/,BRUCE GIBSON,https://www.flickr.com/people/gibsongunleather/\r\nba4508962af69d90,https://c1.staticflickr.com/9/8202/29005733052_a9f711f4bf_o.jpg,0.248348018,0,0.745044053,0.75,673,827,https://flickr.com/37922399@N05/29005733052,https://creativecommons.org/licenses/by/2.0/,Virginia State Parks,https://www.flickr.com/people/vastateparksstaff/\r\n7069a117c92d27a4,https://c1.staticflickr.com/8/7605/26251509774_96c84ae16f_o.jpg,0.421747967,0.25,0.843495935,1,546,186,https://www.flickr.com/photos/psychlist1972/26251509774/,https://creativecommons.org/licenses/by/2.0/,Pete Brown,https://www.flickr.com/people/psychlist1972/\r\n79fa9855119ffeea,https://c1.staticflickr.com/9/8090/28857580530_b7ba1ea2f1_o.jpg,0,0,1,1,479,506,https://www.flickr.com/photos/wwworks/28857580530,https://creativecommons.org/licenses/by/2.0/,woodleywonderworks,https://www.flickr.com/people/wwworks/\r\n82e2334925339ef8,https://c1.staticflickr.com/6/5632/31268100236_4975f1a2db_o.jpg,0.25,0,1,0.75,818,769,https://www.flickr.com/photos/96774554@N03/31268100236,https://creativecommons.org/licenses/by/2.0/,Wheels ON,https://www.flickr.com/people/96774554@N03/\r\n4f5464b7d4682441,https://c1.staticflickr.com/3/2859/32842901716_635c080982_o.jpg,0,0.25,0.47277937,1,976,607,https://www.flickr.com/photos/blmoregon/32842901716/,https://creativecommons.org/licenses/by/2.0/,Bureau of Land Management Oregon and Washington,https://www.flickr.com/people/blmoregon/\r\n72f96c6dc5df2de8,https://c1.staticflickr.com/4/3869/32108968233_aa2d42d481_o.jpg,0.5,0.25,1,1,885,784,https://www.flickr.com/photos/jlascar/32108968233/,https://creativecommons.org/licenses/by/2.0/,Jorge L\\303\\241scar,https://www.flickr.com/people/jlascar/\r\n6609c25b3314b5b3,https://c1.staticflickr.com/9/8031/29690105171_f0e16e643e_o.jpg,0.333333333,0,1,1,512,878,https://www.flickr.com/photos/8782469@N05/29690105171,https://creativecommons.org/licenses/by/2.0/,Graham Robertson,https://www.flickr.com/people/8782469@N05/\r\n52d8515abbfaa8ac,https://c1.staticflickr.com/6/5629/30139827383_2b90e9254e_o.jpg,0.496696035,0,0.99339207,0.75,230,634,https://www.flickr.com/photos/vastateparksstaff/30139827383,https://creativecommons.org/licenses/by/2.0/,Virginia State Parks,https://www.flickr.com/people/vastateparksstaff/\r\na7f9b28e30d19d85,https://c1.staticflickr.com/6/5712/30477407072_911714e259_o.jpg,0,0,0.504444444,0.75,662,699,https://www.flickr.com/photos/andreboeni/30477407072,https://creativecommons.org/licenses/by/2.0/,Andrew Bone,https://www.flickr.com/people/andreboeni/\r\n8b7a5446dae48ac1,https://c1.staticflickr.com/1/316/31908704910_2c6375f0ef_o.jpg,0.266552511,0,0.799657534,0.75,309,363,https://www.flickr.com/photos/gails_pictures/31908704910,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\nb2a7e558f1a3ff16,https://c1.staticflickr.com/9/8730/28819631221_819c793b59_o.jpg,0.303165584,0.25,0.909496753,1,620,536,https://www.flickr.com/photos/raveejaya/28819631221,https://creativecommons.org/licenses/by/2.0/,Raveendra Jayashantha,https://www.flickr.com/people/raveejaya/\r\n1a95932427129cda,https://c1.staticflickr.com/8/7124/27017461135_e00ed6ac0d_o.jpg,0,0.25,0.5,1,657,465,https://www.flickr.com/photos/antoniseb/27017461135/,https://creativecommons.org/licenses/by/2.0/,Jay Cross,https://www.flickr.com/people/antoniseb/\r\nd4579363596fbf21,https://c1.staticflickr.com/1/520/31675116444_127982bddf_o.jpg,0.287322275,0,0.861966825,0.75,833,889,https://www.flickr.com/photos/fran001/31675116444/,https://creativecommons.org/licenses/by/2.0/,Francisco Anzola,https://www.flickr.com/people/fran001/\r\nf7dd570a0cfe1dcc,https://c1.staticflickr.com/9/8264/29543383031_e6d047580a_o.jpg,0,0,0.75,0.75,669,55,https://www.flickr.com/photos/alazaat/29543383031,https://creativecommons.org/licenses/by/2.0/,Hussein Alazaat,https://www.flickr.com/people/alazaat/\r\n2d23b1dab7e3e9c0,https://c1.staticflickr.com/1/456/31638318370_25bb9334d8_o.jpg,0,0.25,0.5,1,511,971,https://www.flickr.com/photos/45958601@N02/31638318370,https://creativecommons.org/licenses/by/2.0/,joey zanotti,https://www.flickr.com/people/45958601@N02/\r\nb7d9e1d8cf1d7912,https://c1.staticflickr.com/6/5697/30316954084_6b364b382d_o.jpg,0,0,0.666666667,1,519,485,https://www.flickr.com/photos/box_repsol/30316954084,https://creativecommons.org/licenses/by/2.0/,Box Repsol,https://www.flickr.com/people/box_repsol/\r\n6b27728f0f816dc6,https://c1.staticflickr.com/1/483/32823130292_df26e1cd49_o.jpg,0,0.25,0.499263623,1,638,874,https://www.flickr.com/photos/rpavich/32823130292/,https://creativecommons.org/licenses/by/2.0/,rpavich,https://www.flickr.com/people/rpavich/\r\nc96358fe7b8eaf28,https://c1.staticflickr.com/8/7262/27455525114_9b1d3ed1ac_o.jpg,0.578822733,0.25,1,1,706,492,https://www.flickr.com/photos/12216515@N00/27455525114,https://creativecommons.org/licenses/by/2.0/,Charles Fulton,https://www.flickr.com/people/12216515@N00/\r\n9b4cab2ffb0c0754,https://c1.staticflickr.com/9/8256/30009146415_d412513e5f_o.jpg,0,0,1,0.75,975,963,https://www.flickr.com/photos/jsjgeology/30009146415,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n1685d066f8070081,https://c1.staticflickr.com/6/5684/29941070052_c4c12092e6_o.jpg,0,0,0.421747967,0.75,979,361,https://www.flickr.com/photos/scotnelson/29941070052,https://creativecommons.org/licenses/by/2.0/,Scot Nelson,https://www.flickr.com/people/scotnelson/\r\ne071bf346787d971,https://c1.staticflickr.com/8/7244/27006991522_4a72303846_o.jpg,0.578252033,0.25,1,1,563,990,https://www.flickr.com/photos/respres/27006991522,https://creativecommons.org/licenses/by/2.0/,Jeff Turner,https://www.flickr.com/people/respres/\r\ne66d732354c6baf0,https://c1.staticflickr.com/9/8349/29382034812_b8cca8606d_o.jpg,0,0,0.5,0.75,301,72,https://www.flickr.com/photos/52450054@N04/29382034812,https://creativecommons.org/licenses/by/2.0/,Judy Gallagher,https://www.flickr.com/people/52450054@N04/\r\n33adbc27fdcadd65,https://c1.staticflickr.com/8/7130/26834523912_fb0b3fe3ed_o.jpg,0.28125,0.25,0.84375,1,924,716,https://www.flickr.com/photos/68147320@N02/26834523912,https://creativecommons.org/licenses/by/2.0/,T.Tseng,https://www.flickr.com/people/68147320@N02/\r\neb53098ebcf05e64,https://c1.staticflickr.com/6/5591/30938192751_f6a6fb1739_o.jpg,0,0,0.641099855,1,564,601,https://www.flickr.com/photos/68704638@N04/30938192751,https://creativecommons.org/licenses/by/2.0/,M Dreibelbis,https://www.flickr.com/people/68704638@N04/\r\n8c609b06a60af060,https://c1.staticflickr.com/8/7199/27864572096_73f40f5d09_o.jpg,0,0,0.5625,0.75,499,291,https://flickr.com/77742560@N06/27864572096,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\nc66310d8b709d6d5,https://c1.staticflickr.com/9/8371/29266955473_d24ddfa05c_o.jpg,0,0,1,1,763,106,https://www.flickr.com/photos/alper/29266955473,https://creativecommons.org/licenses/by/2.0/,Alper \\303\\207u\\304\\237un,https://www.flickr.com/people/alper/\r\n740981d161a38bad,https://c1.staticflickr.com/1/578/32214662255_3bc38af004_o.jpg,0.5,0,1,0.75,572,149,http://www.flickr.com/photos/batcula/32214662255/,https://creativecommons.org/licenses/by/2.0/,Jim,https://www.flickr.com/people/batcula/\r\n4c8a4414419ab0bb,https://c1.staticflickr.com/9/8377/30076813705_00b7bc68ff_o.jpg,0.25,0,0.75,0.75,662,848,https://www.flickr.com/photos/greggjerdingen/30076813705,https://creativecommons.org/licenses/by/2.0/,Greg Gjerdingen,https://www.flickr.com/people/greggjerdingen/\r\necaf98c8ce9f04c3,https://c1.staticflickr.com/9/8547/29609780616_f9dac2244e_o.jpg,0.5,0.25,1,1,561,697,https://www.flickr.com/photos/foran_hs_football/29609780616,https://creativecommons.org/licenses/by/2.0/,Foran Lions Football,https://www.flickr.com/people/foran_hs_football/\r\n093402dc35f59027,https://c1.staticflickr.com/9/8708/28528193204_59181c883b_o.jpg,0.4375,0,1,0.75,734,113,https://www.flickr.com/photos/kdghantous/28528193204,https://creativecommons.org/licenses/by/2.0/,Karim D. Ghantous,https://www.flickr.com/people/kdghantous/\r\n8bb7d10e97694eb5,https://c1.staticflickr.com/1/625/31772996983_1b99ed5012_o.jpg,0.42535545,0,1,0.75,143,393,https://www.flickr.com/photos/taylar/31772996983,https://creativecommons.org/licenses/by/2.0/,Ingrid Taylar,https://www.flickr.com/people/taylar/\r\n820ae900a943d605,https://c1.staticflickr.com/6/5486/30611700582_b302428c5d_o.jpg,0,0.267938931,0.75,0.803816794,888,701,http://www.flickr.com/photos/ecdq/30611700582/,https://creativecommons.org/licenses/by/2.0/,ECDQ,https://www.flickr.com/people/ecdq/\r\n8859efd46fab37ff,https://c1.staticflickr.com/6/5646/30929275820_fc732d7ae6_o.jpg,0.25,0,1,0.75,923,599,https://flickr.com/145862860@N07/30929275820,https://creativecommons.org/licenses/by/2.0/,Ky\\305\\215to B\\305\\215ken,https://www.flickr.com/people/kyotoboken/\r\ndde9f0aadcb02874,https://c1.staticflickr.com/9/8544/29151973710_8ab6e884c9_o.jpg,0.497426471,0,0.994852941,0.75,981,640,https://www.flickr.com/photos/n28307/29151973710,https://creativecommons.org/licenses/by/2.0/,Glenn Beltz,https://www.flickr.com/people/n28307/\r\n13810fcac72f12ba,https://c1.staticflickr.com/9/8302/28623258764_c87251b710_o.jpg,0,0.25,0.5,1,876,72,https://www.flickr.com/photos/usarmyband/28623258764/,https://creativecommons.org/licenses/by/2.0/,The United States Army Band,https://www.flickr.com/people/usarmyband/\r\nb1bf532ce2c4c3a1,https://c1.staticflickr.com/9/8037/28132989550_04e5549aec_o.jpg,0,0,0.495594714,0.75,875,557,https://www.flickr.com/photos/didbygraham/28132989550/,https://creativecommons.org/licenses/by/2.0/,Graham Richardson,https://www.flickr.com/people/didbygraham/\r\n9aba31a4b2404aa5,https://c1.staticflickr.com/1/458/31850811716_9a5882c71c_o.jpg,0.25,0.25,1,1,912,495,https://flickr.com/52066925@N00/31850811716,https://creativecommons.org/licenses/by/2.0/,anneheathen,https://www.flickr.com/people/annethelibrarian/\r\n5b85f98e0105d135,https://c1.staticflickr.com/8/7474/26833993115_f16054e784_o.jpg,0.248713235,0.25,0.746139706,1,570,933,https://www.flickr.com/photos/agenciabrasilia/26833993115,https://creativecommons.org/licenses/by/2.0/,Ag\\303\\252ncia Bras\\303\\255lia,https://www.flickr.com/people/agenciabrasilia/\r\n53c478ffc8ddb967,https://c1.staticflickr.com/9/8149/29031351804_3d4e158245_o.jpg,0.225840336,0,0.677521008,0.75,320,613,https://www.flickr.com/photos/gails_pictures/29031351804,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n92c3f5e99d8ac862,https://c1.staticflickr.com/8/7564/29085116541_43b6dfd46c_o.jpg,0,0,0.666666667,1,664,362,https://www.flickr.com/photos/ministeriodacultura/29085116541/,https://creativecommons.org/licenses/by/2.0/,Minist\\303\\251rio da Cultura,https://www.flickr.com/people/ministeriodacultura/\r\nfae45093bad0961a,https://c1.staticflickr.com/9/8802/30042939706_809382f8e8_o.jpg,0,0,0.499263623,0.75,529,528,https://www.flickr.com/photos/southbeachcars/30042939706,https://creativecommons.org/licenses/by/2.0/,Phillip Pessar,https://www.flickr.com/people/southbeachcars/\r\n0db0952967f6faa6,https://c1.staticflickr.com/6/5610/31207151522_a9d8ef4c85_o.jpg,0.28125,0,0.84375,0.75,816,552,https://flickr.com/29780474@N02/31207151522,https://creativecommons.org/licenses/by/2.0/,Phil Gayton,https://www.flickr.com/people/29780474@N02/\r\n9594ef7e0fc79ea8,https://c1.staticflickr.com/8/7720/26955054100_0d18b0786b_o.jpg,0.25,0.25,1,1,403,115,https://www.flickr.com/photos/87249144@N08/26955054100,https://creativecommons.org/licenses/by/2.0/,FolsomNatural,https://www.flickr.com/people/87249144@N08/\r\n634d7f18d098cccb,https://c1.staticflickr.com/6/5599/30385514465_e5bb5b1ec4_o.jpg,0,0.332106038,1,0.996318115,539,766,https://www.flickr.com/photos/londonmatt/30385514465,https://creativecommons.org/licenses/by/2.0/,Matt Brown,https://www.flickr.com/people/londonmatt/\r\n05cac1419419ce29,https://c1.staticflickr.com/3/2904/33166177915_2cf2b8ed63_o.jpg,0,0.25,0.5,1,854,550,https://www.flickr.com/photos/yangchen6712/33166177915,https://creativecommons.org/licenses/by/2.0/,YangChen(TW),https://www.flickr.com/people/yangchen6712/\r\n868af4a2e614f5ac,https://c1.staticflickr.com/9/8284/29515310871_0dd41b4840_o.jpg,0.25,0,0.75,0.75,718,467,https://www.flickr.com/photos/greggjerdingen/29515310871,https://creativecommons.org/licenses/by/2.0/,Greg Gjerdingen,https://www.flickr.com/people/greggjerdingen/\r\n6898f889c17f8e16,https://c1.staticflickr.com/6/5683/31789036413_516df80786_o.jpg,0.25,0.250923191,1,0.752769572,888,730,https://www.flickr.com/photos/eltb/31789036413,https://creativecommons.org/licenses/by/2.0/,Catedrales e Iglesias/Cathedrals and Churches,https://www.flickr.com/people/eltb/\r\n8e3324c208eb9a3b,https://c1.staticflickr.com/1/358/32563010712_d1417f0485_o.jpg,0,0,0.561328125,0.75,822,741,https://www.flickr.com/photos/linsinchen/32563010712,https://creativecommons.org/licenses/by/2.0/,Sinchen.Lin,https://www.flickr.com/people/linsinchen/\r\n7d7a8e41b8d0577e,https://c1.staticflickr.com/6/5518/30395077313_019dca3b01_o.jpg,0.25,0,1,0.75,965,182,https://flickr.com/83202873@N00/30395077313,https://creativecommons.org/licenses/by/2.0/,smittenkittenorig,https://www.flickr.com/people/smittenkittenoriginals/\r\n5ff168f7d006f69c,https://c1.staticflickr.com/1/506/31115729834_fc93bbea0f_o.jpg,0,0.25,0.5625,1,683,106,https://flickr.com/45457437@N00/31115729834,https://creativecommons.org/licenses/by/2.0/,David Jones,https://www.flickr.com/people/davidcjones/\r\nf348adf7ea1967f7,https://c1.staticflickr.com/9/8769/29760593836_4e7c8ebbda_o.jpg,0.5,0.25,1,1,909,415,https://www.flickr.com/photos/fun_flying/29760593836/,https://creativecommons.org/licenses/by/2.0/,D. Miller,https://www.flickr.com/people/fun_flying/\r\n40f79488e3c2d6c4,https://c1.staticflickr.com/8/7433/27742701310_08f79aaeb0_o.jpg,0,0.247434018,0.75,0.742302053,918,749,https://www.flickr.com/photos/mdpettitt/27742701310,https://creativecommons.org/licenses/by/2.0/,Martin Pettitt,https://www.flickr.com/people/mdpettitt/\r\n3e9496f7652e1968,https://c1.staticflickr.com/6/5503/30144665264_862fd303dc_o.jpg,0,0,0.666666667,1,469,834,http://www.flickr.com/photos/27148401@N06/30144665264,https://creativecommons.org/licenses/by/2.0/,shipley43,https://www.flickr.com/people/27148401@N06/\r\n2f7d5ff922af55c4,https://c1.staticflickr.com/9/8171/28748561034_d77c5db28e_o.jpg,0.475453172,0.25,1,1,481,633,https://www.flickr.com/photos/jeepersmedia/28748561034,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n79a5625805a08253,https://c1.staticflickr.com/8/7622/26950982570_224d340d99_o.jpg,0,0.25,0.498159057,1,766,805,https://www.flickr.com/photos/vastateparksstaff/26950982570,https://creativecommons.org/licenses/by/2.0/,Virginia State Parks,https://www.flickr.com/people/vastateparksstaff/\r\n412311ea17528b7c,https://c1.staticflickr.com/6/5553/31345581012_51f411cd78_o.jpg,0.25,0,0.75,0.75,764,440,https://www.flickr.com/photos/quinet/31345581012/,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n00312c7e7196baf4,https://c1.staticflickr.com/9/8539/29256709000_ca71fa227f_o.jpg,0.25,0.28125,1,0.84375,327,220,http://www.flickr.com/photos/27305106@N06/29256709000,https://creativecommons.org/licenses/by/2.0/,Charlie Jackson,https://www.flickr.com/people/chaz_pics/\r\n51d1c77f80467b59,https://c1.staticflickr.com/6/5555/31034796165_c8996cacff_o.jpg,0.329178886,0,0.987536657,1,977,538,https://www.flickr.com/photos/zemzina/31034796165/,https://creativecommons.org/licenses/by/2.0/,Olga Khomitsevich,https://www.flickr.com/people/zemzina/\r\n82179b1d313e0e23,https://c1.staticflickr.com/1/268/32350804376_10b1f40d6e_o.jpg,0,0,0.5625,0.75,830,902,https://www.flickr.com/photos/76969036@N02/32350804376/,https://creativecommons.org/licenses/by/2.0/,David Guo,https://www.flickr.com/people/76969036@N02/\r\nb0e0e06e647b478f,https://c1.staticflickr.com/6/5769/29834734204_b8897fa3f7_o.jpg,0,0,1,0.751173709,484,915,https://www.flickr.com/photos/sky_hlv/29834734204,https://creativecommons.org/licenses/by/2.0/,Jose A.,https://www.flickr.com/people/sky_hlv/\r\nc609f86fa706bb9e,https://c1.staticflickr.com/9/8552/29601011844_a32bf06fd2_o.jpg,0,0,0.5625,0.75,113,889,https://www.flickr.com/photos/fdecomite/29601011844/,https://creativecommons.org/licenses/by/2.0/,fdecomite,https://www.flickr.com/people/fdecomite/\r\n55a2e48a87c16998,https://c1.staticflickr.com/1/660/31687904441_5798b661c4_o.jpg,0.25,0,0.75,0.75,911,768,http://www.flickr.com/photos/wastes/31687904441/,https://creativecommons.org/licenses/by/2.0/,Charlotta Wasteson,https://www.flickr.com/people/wastes/\r\n7d833ac8ab852bd2,https://c1.staticflickr.com/1/536/31504100213_b2b24c3f0d_o.jpg,0.25,0.442757009,1,1,52,151,https://www.flickr.com/photos/jsjgeology/31504100213,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\nfe92a88132b5d67b,https://c1.staticflickr.com/6/5127/30101893166_db47d4701a_o.jpg,0.25,0.25,0.75,1,963,247,https://flickr.com/112540417@N03/30101893166,https://creativecommons.org/licenses/by/2.0/,Enrico BBQness,https://www.flickr.com/people/bbqnessdotcom/\r\n6ab876a15af539c4,https://c1.staticflickr.com/9/8869/28304826635_f4c6b9f7d7_o.jpg,0.25,0.28125,1,0.84375,984,611,https://www.flickr.com/photos/noaaphotolib/28304826635,https://creativecommons.org/licenses/by/2.0/,NOAA Photo Library,https://www.flickr.com/people/noaaphotolib/\r\n4fff343244f212cc,https://c1.staticflickr.com/6/5774/31301114735_677b0f08fa_o.jpg,0,0.305872757,0.75,0.917618271,821,403,https://www.flickr.com/photos/45131642@N00/31301114735/,https://creativecommons.org/licenses/by/2.0/,Phil Parker,https://www.flickr.com/people/45131642@N00/\r\n22a9c318b6e36853,https://c1.staticflickr.com/9/8016/29443140000_36cbb76243_o.jpg,0.249079529,0.25,0.747238586,1,620,211,https://www.flickr.com/photos/51715344@N07/29443140000/,https://creativecommons.org/licenses/by/2.0/,kami rao,https://www.flickr.com/people/51715344@N07/\r\n71d368782c72e313,https://c1.staticflickr.com/8/7431/27233705735_358064269d_o.jpg,0,0,0.599757674,0.75,913,487,https://www.flickr.com/photos/mspearman/27233705735,https://creativecommons.org/licenses/by/2.0/,Mark Spearman,https://www.flickr.com/people/mspearman/\r\n65b8c8b065069802,https://c1.staticflickr.com/8/7116/26898609074_1eab8aeb4b_o.jpg,0.25,0.25,0.75,1,666,379,https://www.flickr.com/photos/sg2012/26898609074/,https://creativecommons.org/licenses/by/2.0/,Steve Glover,https://www.flickr.com/people/sg2012/\r\nb0619a61cf135f5e,https://c1.staticflickr.com/8/7408/27239703215_11f55dc09e_o.jpg,0.345588235,0,0.691176471,0.75,985,420,https://www.flickr.com/photos/126816719@N03/27239703215/,https://creativecommons.org/licenses/by/2.0/,Jon Brinn,https://www.flickr.com/people/126816719@N03/\r\n93394c1f9c4c0053,https://c1.staticflickr.com/9/8065/27937528884_57f0526c15_o.jpg,0.5,0.25,1,1,320,883,https://www.flickr.com/photos/52450054@N04/27937528884,https://creativecommons.org/licenses/by/2.0/,Judy Gallagher,https://www.flickr.com/people/52450054@N04/\r\naa90f93ed6fae139,https://c1.staticflickr.com/1/577/32316961601_cd6d1ed035_o.jpg,0,0,0.75,0.75,310,828,https://www.flickr.com/photos/orangeaurochs/32316961601,https://creativecommons.org/licenses/by/2.0/,Orangeaurochs,https://www.flickr.com/people/orangeaurochs/\r\naa07ba1587ec047e,https://c1.staticflickr.com/9/8576/29046638172_6468dd6b89_o.jpg,0.4375,0,1,0.75,469,900,https://www.flickr.com/photos/stephenk1977/29046638172,https://creativecommons.org/licenses/by/2.0/,stephenk1977,https://www.flickr.com/people/stephenk1977/\r\n5d8085ae68e07d72,https://c1.staticflickr.com/8/7049/27394318535_a8c7ccdeb8_o.jpg,0,0.28125,0.75,0.84375,709,643,https://www.flickr.com/photos/zongo/27394318535,https://creativecommons.org/licenses/by/2.0/,DAVID HOLT,https://www.flickr.com/people/zongo/\r\n749ed3502293d019,https://c1.staticflickr.com/9/8356/29415192905_173fbf26da_o.jpg,0,0.330871492,1,1,475,578,https://www.flickr.com/photos/romitagirl67/29415192905/,https://creativecommons.org/licenses/by/2.0/,RomitaGirl67,https://www.flickr.com/people/romitagirl67/\r\n791c3d333052ca64,https://c1.staticflickr.com/9/8440/29197626174_134d6fd15f_o.jpg,0,0,0.5625,0.75,964,908,https://www.flickr.com/photos/cbroders/29197626174,https://creativecommons.org/licenses/by/2.0/,Counse,https://www.flickr.com/people/cbroders/\r\n90e9a43bfb6a68b7,https://c1.staticflickr.com/9/8865/28740349711_5758bc82f4_o.jpg,0,0,0.5,0.75,867,476,https://www.flickr.com/photos/iowapipe/28740349711,https://creativecommons.org/licenses/by/2.0/,IowaPipe,https://www.flickr.com/people/iowapipe/\r\n72e6ba0cc7fb30a6,https://c1.staticflickr.com/9/8270/29053097670_b2b9f42c1f_o.jpg,0.28125,0,0.84375,0.75,875,834,https://www.flickr.com/photos/aswans1234567/29053097670,https://creativecommons.org/licenses/by/2.0/,Alex Swanston's Bus Photos,https://www.flickr.com/people/aswans1234567/\r\n8c5b5bc0f39e2e06,https://c1.staticflickr.com/9/8767/28247360751_7c466f8efc_o.jpg,0.421177267,0.25,0.842354533,1,898,253,https://www.flickr.com/photos/brandedbulltank/28247360751/,https://creativecommons.org/licenses/by/2.0/,Tank,https://www.flickr.com/people/brandedbulltank/\r\ncc43200264a62126,https://c1.staticflickr.com/8/7447/27965290502_73c8417ef8_o.jpg,0.25,0,0.75,0.75,981,590,https://www.flickr.com/photos/naql/27965290502,https://creativecommons.org/licenses/by/2.0/,naql,https://www.flickr.com/people/naql/\r\nbe95f5be87f6fadd,https://c1.staticflickr.com/6/5721/30814461816_67e70a5169_o.jpg,0.28125,0,0.84375,0.75,459,603,https://www.flickr.com/photos/davidwilson1949/30814461816/,https://creativecommons.org/licenses/by/2.0/,David Wilson,https://www.flickr.com/people/davidwilson1949/\r\n7b6e963025eaeb66,https://c1.staticflickr.com/1/316/31606962904_b2666a8780_o.jpg,0.331617647,0,0.994852941,1,994,660,https://flickr.com/62788663@N04/31606962904,https://creativecommons.org/licenses/by/2.0/,Martin Jambon,https://www.flickr.com/people/mjambon/\r\n7f842d0c1e747334,https://c1.staticflickr.com/8/7634/27141449325_4f1a76c3ed_o.jpg,0,0,1,1,305,796,https://www.flickr.com/photos/78428166@N00/27141449325,https://creativecommons.org/licenses/by/2.0/,Tony Alter,https://www.flickr.com/people/78428166@N00/\r\ned3cb7f5154ae02d,https://c1.staticflickr.com/9/8072/29900375851_914813808d_o.jpg,0.245255474,0.25,0.735766423,1,980,747,https://www.flickr.com/photos/col_and_tasha/29900375851,https://creativecommons.org/licenses/by/2.0/,Col Ford and Natasha de Vere,https://www.flickr.com/people/col_and_tasha/\r\n6ecbd7d19f5b9147,https://c1.staticflickr.com/9/8737/28811659522_a0d4210243_o.jpg,0,0,0.483695652,0.75,596,423,https://www.flickr.com/photos/robdammers/28811659522/,https://creativecommons.org/licenses/by/2.0/,Rob Dammers,https://www.flickr.com/people/robdammers/\r\nd016b3204e7ca96e,https://c1.staticflickr.com/9/8558/29643762441_5e78f77661_o.jpg,0,0,0.75,1,792,778,https://flickr.com/39160147@N03/29643762441,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n8400742ba78ac009,https://c1.staticflickr.com/9/8544/28451617274_58e56a03de_o.jpg,0,0,0.666666667,1,971,671,https://www.flickr.com/photos/blackwing_de/28451617274,https://creativecommons.org/licenses/by/2.0/,Sebastian Werner,https://www.flickr.com/people/blackwing_de/\r\n121c87c3c822e2c5,https://c1.staticflickr.com/1/750/31463784832_7a3c075877_o.jpg,0.25,0.25,0.75,1,459,386,https://www.flickr.com/photos/quinet/31463784832,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n76886996d4ab4614,https://c1.staticflickr.com/9/8192/28493603004_1b628ae8e9_o.jpg,0,0.25,0.5625,1,975,763,https://www.flickr.com/photos/jsjgeology/28493603004,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n9d5ef11ec55123e5,https://c1.staticflickr.com/9/8872/27883604363_8cf68f0178_o.jpg,0,0.25,0.577654517,1,100,513,http://www.flickr.com/photos/dangrothe/27883604363/,https://creativecommons.org/licenses/by/2.0/,Daniel Grothe,https://www.flickr.com/people/dangrothe/\r\n1dfbeffd999f280e,https://c1.staticflickr.com/9/8433/28830160583_6bc4405a04_o.jpg,0.181375628,0,0.544126884,0.75,582,569,https://www.flickr.com/photos/mauritsverbiest/28830160583,https://creativecommons.org/licenses/by/2.0/,Maurits Verbiest,https://www.flickr.com/people/mauritsverbiest/\r\ne3f5dbed7bd31314,https://c1.staticflickr.com/8/7682/27343823144_eeaa9065cb_o.jpg,0.25,0.5,1,1,547,864,https://www.flickr.com/photos/barnimages/27343823144,https://creativecommons.org/licenses/by/2.0/,barnimages.com,https://www.flickr.com/people/barnimages/\r\neb8ce5ba5c47e344,https://c1.staticflickr.com/1/684/33351191226_792c561fd4_o.jpg,0,0,0.75,0.5625,669,893,https://flickr.com/126397594@N03/33351191226,https://creativecommons.org/licenses/by/2.0/,G\\303\\274ldem \\303\\234st\\303\\274n,https://www.flickr.com/people/guldem/\r\nf5dffe28fc19b6e7,https://c1.staticflickr.com/6/5529/31290944551_78ff48d9a3_o.jpg,0.266071429,0.25,1,1,918,730,https://flickr.com/91873856@N00/31290944551,https://creativecommons.org/licenses/by/2.0/,Phil Venditti,https://www.flickr.com/people/venditti_min_min-venditti/\r\n1ebff1b3c8de66de,https://c1.staticflickr.com/9/8551/29684108315_b9b08f8035_o.jpg,0.464122137,0.25,1,1,337,810,https://www.flickr.com/photos/ruthanddave/29684108315,https://creativecommons.org/licenses/by/2.0/,Ruth Hartnup,https://www.flickr.com/people/ruthanddave/\r\n0546b47c13f154d3,https://c1.staticflickr.com/1/744/31706873925_eb85a4c572_o.jpg,0,0,0.976785714,1,108,318,https://www.flickr.com/photos/92252798@N07/31706873925,https://creativecommons.org/licenses/by/2.0/,Dick Culbert,https://www.flickr.com/people/92252798@N07/\r\nf52db15134c1a74e,https://c1.staticflickr.com/6/5628/30300134543_71291f4050_o.jpg,0,0,1,1,638,776,https://flickr.com/13894201@N00/30300134543,https://creativecommons.org/licenses/by/2.0/,\\303\\201d\\303\\241m Szedl\\303\\241k,https://www.flickr.com/people/kelt/\r\n8094d4088a08cf09,https://c1.staticflickr.com/1/725/31560908483_4276d9241c_o.jpg,0,0,0.666666667,1,441,615,https://www.flickr.com/photos/allagashbrewing/31560908483/,https://creativecommons.org/licenses/by/2.0/,Allagash Brewing,https://www.flickr.com/people/allagashbrewing/\r\nd9f624188d16fe93,https://c1.staticflickr.com/8/7402/27067928760_efcb66d26f_o.jpg,0.25,0.25,0.75,1,40,602,https://www.flickr.com/photos/juliofromero/27067928760,https://creativecommons.org/licenses/by/2.0/,Julio Romero,https://www.flickr.com/people/juliofromero/\r\n631b6820e91a9e3e,https://c1.staticflickr.com/3/2183/32694281652_62b0c63d26_o.jpg,0.333333333,0,1,1,13,303,https://www.flickr.com/photos/wildreturn/32694281652/,https://creativecommons.org/licenses/by/2.0/,Andy Reago & Chrissy McClarren,https://www.flickr.com/people/wildreturn/\r\n611f453c4bb616b9,https://c1.staticflickr.com/8/7494/27095667650_2158e5d0a3_o.jpg,0.25,0.25,0.75,1,310,65,https://www.flickr.com/photos/gaby1/27095667650,https://creativecommons.org/licenses/by/2.0/,Gabriel Gonz\\303\\241lez,https://www.flickr.com/people/gaby1/\r\n5792d4a42d2a82c8,https://c1.staticflickr.com/9/8655/30162044361_628a8758b1_o.jpg,0,0,1,1,988,419,https://flickr.com/51966661@N00/30162044361,https://creativecommons.org/licenses/by/2.0/,valerialaura,https://www.flickr.com/people/valerialaura/\r\nc0e0c11350b8740e,https://c1.staticflickr.com/9/8316/28819944676_84ab998606_o.jpg,0,0,0.741835148,1,319,698,https://www.flickr.com/photos/gails_pictures/28819944676,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\nc663670c06c06ea4,https://c1.staticflickr.com/6/5671/30978227515_c8c43aa172_o.jpg,0,0.248826291,1,1,909,903,http://www.flickr.com/photos/verzo/30978227515/,https://creativecommons.org/licenses/by/2.0/,Roberto Verzo,https://www.flickr.com/people/verzo/\r\n49591319edeb6b3a,https://c1.staticflickr.com/8/7379/28056205761_56dc8a13a9_o.jpg,0.333333333,0,1,1,309,416,https://www.flickr.com/photos/104342908@N08/28056205761,https://creativecommons.org/licenses/by/2.0/,Michael Mueller,https://www.flickr.com/people/104342908@N08/\r\n02eb799d147d5f91,https://c1.staticflickr.com/8/7461/27924109056_3f35c44ec8_o.jpg,0,0,0.5,0.75,625,773,https://www.flickr.com/photos/vvbad/27924109056/,https://creativecommons.org/licenses/by/2.0/,VVBAD,https://www.flickr.com/people/vvbad/\r\nf2d36d089779166a,https://c1.staticflickr.com/8/7528/27618132235_3bd6c10788_o.jpg,0,0.25,0.503698225,1,349,805,https://www.flickr.com/photos/eljay/27618132235/,https://creativecommons.org/licenses/by/2.0/,Eljay,https://www.flickr.com/people/eljay/\r\ncc8d850717026796,https://c1.staticflickr.com/8/7304/26869810723_660f0642a5_o.jpg,0,0,1,1,662,791,https://www.flickr.com/photos/29233640@N07/26869810723,https://creativecommons.org/licenses/by/2.0/,Robert Couse-Baker,https://www.flickr.com/people/29233640@N07/\r\nab3fef142d031784,https://c1.staticflickr.com/1/17/30695559644_ea51d4a001_o.jpg,0,0.25,0.5625,1,657,737,https://www.flickr.com/photos/27665395@N05/30695559644,https://creativecommons.org/licenses/by/2.0/,Jason Lawrence,https://www.flickr.com/people/27665395@N05/\r\nc84cbc2c05227bee,https://c1.staticflickr.com/1/778/32130016155_b5c1709d28_o.jpg,0.28125,0,0.84375,0.75,437,979,https://www.flickr.com/photos/nzcarfreak/32130016155/,https://creativecommons.org/licenses/by/2.0/,Riley,https://www.flickr.com/people/nzcarfreak/\r\n9d55997552acf0de,https://c1.staticflickr.com/4/3827/33042042362_c6d848f625_o.jpg,0,0.25,0.498159057,1,260,199,https://flickr.com/145431051@N05/33042042362,https://creativecommons.org/licenses/by/2.0/,Natalia King,https://www.flickr.com/people/inataliaking/\r\n8da3593258ae27cb,https://c1.staticflickr.com/8/7304/27230857426_6398e3ff70_o.jpg,0.257672156,0.25,0.773016467,1,581,399,https://www.flickr.com/photos/eu2016nl/27230857426,https://creativecommons.org/licenses/by/2.0/,EU2016 NL,https://www.flickr.com/people/eu2016nl/\r\nfccab516c111d524,https://c1.staticflickr.com/1/578/31833204592_df37b2fd5f_o.jpg,0.5,0,1,0.75,650,757,https://www.flickr.com/photos/shadowgate/31833204592,https://creativecommons.org/licenses/by/2.0/,Shadowgate,https://www.flickr.com/people/shadowgate/\r\n315e55af9615abd2,https://c1.staticflickr.com/1/779/31725794312_e202c64123_o.jpg,0.256726457,0.25,0.770179372,1,347,106,https://flickr.com/13476480@N07/31725794312,https://creativecommons.org/licenses/by/2.0/,manhhai,https://www.flickr.com/people/13476480@N07/\r\nc135337ca7a575ae,https://c1.staticflickr.com/9/8239/28468432034_845b238b0a_o.jpg,0.28125,0,0.84375,0.75,348,178,https://www.flickr.com/photos/sedoglia/28468432034,https://creativecommons.org/licenses/by/2.0/,Sergio,https://www.flickr.com/people/sedoglia/\r\n3bfe9ff7b488485d,https://c1.staticflickr.com/8/7526/27213158786_4411c71bb4_o.jpg,0,0.25,0.5,1,706,710,https://www.flickr.com/photos/45220772@N06/27213158786,https://creativecommons.org/licenses/by/2.0/,Chris GBNL,https://www.flickr.com/people/45220772@N06/\r\n1e33aa336b89dcff,https://c1.staticflickr.com/9/8069/29343726103_76a4d6550d_o.jpg,0.455384615,0,1,0.75,539,588,https://www.flickr.com/photos/hisgett/29343726103/,https://creativecommons.org/licenses/by/2.0/,Tony Hisgett,https://www.flickr.com/people/hisgett/\r\n39a483e9a23716ed,https://c1.staticflickr.com/6/5803/30224386731_129a573cb6_o.jpg,0,0.25,0.497426471,1,676,646,https://www.flickr.com/photos/thenationalguard/30224386731,https://creativecommons.org/licenses/by/2.0/,The National Guard,https://www.flickr.com/people/thenationalguard/\r\n5191cfba868056a5,https://c1.staticflickr.com/9/8057/28990247303_f5235fa9ae_o.jpg,0.250923191,0.25,0.752769572,1,971,793,http://www.flickr.com/photos/pekkatamminen/28990247303/,https://creativecommons.org/licenses/by/2.0/,Pekka Tamminen,https://www.flickr.com/people/pekkatamminen/\r\n4572c8986c28343b,https://c1.staticflickr.com/9/8672/28241882454_babaf6f88f_o.jpg,0.334315169,0,1,1,437,535,https://www.flickr.com/photos/greggjerdingen/28241882454,https://creativecommons.org/licenses/by/2.0/,Greg Gjerdingen,https://www.flickr.com/people/greggjerdingen/\r\n7700d0ba053dc4bd,https://c1.staticflickr.com/9/8881/28767707555_05e533e1a1_o.jpg,0.25,0.249079529,1,0.747238586,937,292,https://www.flickr.com/photos/scotnelson/28767707555,https://creativecommons.org/licenses/by/2.0/,Scot Nelson,https://www.flickr.com/people/scotnelson/\r\neab9755117160864,https://c1.staticflickr.com/8/7509/29649911804_fbed1c6dac_o.jpg,0.25,0,1,0.618442623,611,510,https://flickr.com/128305985@N02/29649911804,https://creativecommons.org/licenses/by/2.0/,Varevo Designs,https://www.flickr.com/people/128305985@N02/\r\n8b47d72b39b2bd33,https://c1.staticflickr.com/8/7458/27617552740_586605d6d1_o.jpg,0.33113069,0,0.99339207,1,526,97,https://flickr.com/77742560@N06/27617552740,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n7d2a7bca22967f29,https://c1.staticflickr.com/8/7525/27211199483_b103fe3789_o.jpg,0,0,0.512686567,0.75,735,600,https://www.flickr.com/photos/hisgett/27211199483,https://creativecommons.org/licenses/by/2.0/,Tony Hisgett,https://www.flickr.com/people/hisgett/\r\n8e89734ed940e222,https://c1.staticflickr.com/1/510/31839325642_621a361097_o.jpg,0.304216867,0,1,1,657,794,https://www.flickr.com/photos/autohistorian/31839325642,https://creativecommons.org/licenses/by/2.0/,Alden Jewell,https://www.flickr.com/people/autohistorian/\r\n30b922dbfdbffd11,https://c1.staticflickr.com/3/2431/32476279840_81a638927d_o.jpg,0.28125,0,0.84375,0.75,610,972,https://www.flickr.com/photos/karen_roe/32476279840,https://creativecommons.org/licenses/by/2.0/,Karen Roe,https://www.flickr.com/people/karen_roe/\r\nef0e34b8ccb6d42f,https://c1.staticflickr.com/9/8841/28756457540_dd42170f61_o.jpg,0.25,0.25,0.75,1,982,712,https://www.flickr.com/photos/87533529@N00/28756457540/,https://creativecommons.org/licenses/by/2.0/,Joel Dinda,https://www.flickr.com/people/mwlguide/\r\n506b87b78daa499c,https://c1.staticflickr.com/8/7427/26651646184_52fdea24f9_o.jpg,0,0,0.575553797,0.75,821,244,https://www.flickr.com/photos/jsjgeology/26651646184,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n62e0e0efc6deeef4,https://c1.staticflickr.com/8/7292/26212453023_25d43a954f_o.jpg,0,0,0.5,0.75,971,772,https://www.flickr.com/photos/akashadorable/26212453023,https://creativecommons.org/licenses/by/2.0/,reverie_rambler,https://www.flickr.com/people/akashadorable/\r\n28652355cd12a1e7,https://c1.staticflickr.com/8/7281/27719555664_9d2eee5a25_o.jpg,0.276162791,0,0.828488372,0.75,308,63,https://www.flickr.com/photos/86548370@N00/27719555664/,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\n0ed355357757528d,https://c1.staticflickr.com/1/771/30766164923_65f81b0a16_o.jpg,0,0,0.666666667,1,763,795,https://www.flickr.com/photos/dalbera/30766164923,https://creativecommons.org/licenses/by/2.0/,Jean-Pierre Dalb\\303\\251ra,https://www.flickr.com/people/dalbera/\r\nabe9504895f6e8dd,https://c1.staticflickr.com/8/7719/27005746036_e3c2d66824_o.jpg,0.204773031,0.25,0.614319092,1,566,607,https://www.flickr.com/photos/kansasscanner/27005746036/,https://creativecommons.org/licenses/by/2.0/,Tyler Silvest,https://www.flickr.com/people/kansasscanner/\r\ndd984311bff04342,https://c1.staticflickr.com/6/5505/30880758442_b260c358d1_o.jpg,0.578252033,0,1,0.75,118,524,https://flickr.com/149229065@N07/30880758442,https://creativecommons.org/licenses/by/2.0/,bready_fv,https://www.flickr.com/people/149229065@N07/\r\nf7947af6545be760,https://c1.staticflickr.com/9/8088/29622519706_0b2d2626e1_o.jpg,0,0,0.7484375,1,418,722,https://www.flickr.com/photos/xingxiyang/29622519706/,https://creativecommons.org/licenses/by/2.0/,Kiefer,https://www.flickr.com/people/xingxiyang/\r\n01244097ca8ffdfa,https://c1.staticflickr.com/9/8420/29072932140_428884b149_o.jpg,0,0,0.666666667,1,91,563,https://www.flickr.com/photos/bigtallguy/29072932140,https://creativecommons.org/licenses/by/2.0/,Paul Reynolds,https://www.flickr.com/people/bigtallguy/\r\n3cf4bb0637d0f182,https://c1.staticflickr.com/6/5336/31376083036_f61bd4c093_o.jpg,0,0.4375,0.75,1,862,52,http://www.flickr.com/photos/subzonica/31376083036/,https://creativecommons.org/licenses/by/2.0/,Hern\\303\\241n Garc\\303\\255a Crespo,https://www.flickr.com/people/subzonica/\r\n1d5361e858fe8462,https://c1.staticflickr.com/1/252/31161493730_01da9f540b_o.jpg,0,0.25,0.377400768,1,499,523,https://www.flickr.com/photos/usnavalwarcollegeri/31161493730/,https://creativecommons.org/licenses/by/2.0/,U.S. Naval War College,https://www.flickr.com/people/usnavalwarcollegeri/\r\n6a50fbe8723a6a4a,https://c1.staticflickr.com/1/571/31590985202_b888dd2600_o.jpg,0.5,0.25,1,1,563,597,https://www.flickr.com/photos/rollercoasterphilosophy/31590985202/,https://creativecommons.org/licenses/by/2.0/,Jeremy Thompson,https://www.flickr.com/people/rollercoasterphilosophy/\r\n001b5e6f1b89fd3b,https://c1.staticflickr.com/6/5342/31290416685_7145707d8a_o.jpg,0.4375,0.25,1,1,577,909,https://www.flickr.com/photos/127226743@N02/31290416685,https://creativecommons.org/licenses/by/2.0/,Dimitris Kamaras,https://www.flickr.com/people/127226743@N02/\r\n25eef91443ded954,https://c1.staticflickr.com/6/5495/29933353114_e345092c4c_o.jpg,0.471030043,0,0.942060086,0.75,437,728,https://www.flickr.com/photos/autohistorian/29933353114,https://creativecommons.org/licenses/by/2.0/,Alden Jewell,https://www.flickr.com/people/autohistorian/\r\ned3f9b6a081a4802,https://c1.staticflickr.com/8/7432/27519628793_d12a4b1349_o.jpg,0,0,0.666666667,1,818,210,https://www.flickr.com/photos/thecarspy/27519628793,https://creativecommons.org/licenses/by/2.0/,The Car Spy,https://www.flickr.com/people/thecarspy/\r\n95fbd20d0aaf319a,https://c1.staticflickr.com/1/292/31801536095_12b2b4d011_o.jpg,0.333333333,0,1,1,826,300,https://www.flickr.com/photos/shankaronline/31801536095,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n2bc6c55a0ab24f86,https://c1.staticflickr.com/8/7578/28472228121_b58f3263f6_o.jpg,0.28125,0,0.84375,0.75,786,697,https://www.flickr.com/photos/merfam/28472228121/,https://creativecommons.org/licenses/by/2.0/,Jason Meredith,https://www.flickr.com/people/merfam/\r\n319e8b63d6a194aa,https://c1.staticflickr.com/9/8203/29608037100_b9c0377cb0_o.jpg,0.4375,0.25,1,1,826,529,https://www.flickr.com/photos/127226743@N02/29608037100,https://creativecommons.org/licenses/by/2.0/,Dimitris Kamaras,https://www.flickr.com/people/127226743@N02/\r\n4fcc9c949d66a4e0,https://c1.staticflickr.com/9/8273/29520584994_c6a87b78b7_o.jpg,0.176470588,0,0.529411765,0.75,771,104,https://flickr.com/35080385@N08/29520584994,https://creativecommons.org/licenses/by/2.0/,Daniel Max,https://www.flickr.com/people/35080385@N08/\r\n5c847b72d1e1713f,https://c1.staticflickr.com/8/7402/27581042815_97ab7892e1_o.jpg,0.25,0,1,0.75,556,522,https://www.flickr.com/photos/13152419@N05/27581042815/,https://creativecommons.org/licenses/by/2.0/,A.J. WOODSON,https://www.flickr.com/people/13152419@N05/\r\n707bce947d54bcd8,https://c1.staticflickr.com/6/5533/31047975071_6e1e9857a2_o.jpg,0.436619718,0.25,1,1,947,941,https://www.flickr.com/photos/amantedar/31047975071/,https://creativecommons.org/licenses/by/2.0/,Amante Darmanin,https://www.flickr.com/people/amantedar/\r\n43fa89c17e628c06,https://c1.staticflickr.com/6/5669/30471849463_58158aea97_o.jpg,0.28125,0,0.84375,0.75,133,885,https://www.flickr.com/photos/ddebold/30471849463,https://creativecommons.org/licenses/by/2.0/,Don DeBold,https://www.flickr.com/people/ddebold/\r\n6612fd36e6dd9534,https://c1.staticflickr.com/8/7446/27992527025_02e1093865_o.jpg,0.248826291,0,1,1,309,866,https://www.flickr.com/photos/treegrow/27992527025,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\n4146a745f9b2f3cb,https://c1.staticflickr.com/1/453/32533674871_37d9a42b5c_o.jpg,0.28125,0,0.84375,0.75,51,418,https://www.flickr.com/photos/kjfnjy/32533674871,https://creativecommons.org/licenses/by/2.0/,Ken and Nyetta,https://www.flickr.com/people/kjfnjy/\r\n74bfc95412b2f824,https://c1.staticflickr.com/8/7046/26382321234_683510e9d6_o.jpg,0,0,1,1,672,655,https://www.flickr.com/photos/salvolg/26382321234,https://creativecommons.org/licenses/by/2.0/,Salvatore,https://www.flickr.com/people/salvolg/\r\n5fc906f27c8975d2,https://c1.staticflickr.com/9/8029/29167405853_4e4009a5d5_o.jpg,0,0,0.5625,0.75,979,367,https://www.flickr.com/photos/marsupilami92/29167405853,https://creativecommons.org/licenses/by/2.0/,patrick janicek,https://www.flickr.com/people/marsupilami92/\r\n2fc17ec1fad5d947,https://c1.staticflickr.com/2/1618/26152633944_15768856b8_o.jpg,0,0.25,0.579761905,1,874,929,https://www.flickr.com/photos/sergejf/26152633944,https://creativecommons.org/licenses/by/2.0/,sergejf,https://www.flickr.com/people/sergejf/\r\ncb63df64516985c1,https://c1.staticflickr.com/8/7226/27305715702_336cd34ebb_o.jpg,0.334315169,0,1,1,635,772,https://www.flickr.com/photos/sarahvain/27305715702,https://creativecommons.org/licenses/by/2.0/,Sarah Stierch,https://www.flickr.com/people/sarahvain/\r\n7b8413ef241118df,https://c1.staticflickr.com/6/5627/31291332596_e0705f6670_o.jpg,0.4375,0.25,1,1,548,485,https://www.flickr.com/photos/davidwilson1949/31291332596,https://creativecommons.org/licenses/by/2.0/,David Wilson,https://www.flickr.com/people/davidwilson1949/\r\n2e854fca826e54b3,https://c1.staticflickr.com/1/307/31234607810_e2d5b4b1c8_o.jpg,0.333333333,0,1,1,858,596,https://www.flickr.com/photos/dalbera/31234607810/,https://creativecommons.org/licenses/by/2.0/,Jean-Pierre Dalb\\303\\251ra,https://www.flickr.com/people/dalbera/\r\nf0add7f552d2d895,https://c1.staticflickr.com/6/5800/30295930850_0c40713818_o.jpg,0,0,0.5625,0.75,650,543,https://www.flickr.com/photos/technoevangelist/30295930850,https://creativecommons.org/licenses/by/2.0/,Fee Plumley,https://www.flickr.com/people/technoevangelist/\r\n9d4674c86096e4bd,https://c1.staticflickr.com/9/8410/29866310754_8c13093feb_o.jpg,0.25,0.470789074,1,1,662,614,https://www.flickr.com/photos/autohistorian/29866310754,https://creativecommons.org/licenses/by/2.0/,Alden Jewell,https://www.flickr.com/people/autohistorian/\r\n05437214d1ba9580,https://c1.staticflickr.com/9/8829/28380794150_706cf5d693_o.jpg,0,0.25,0.5625,1,810,118,https://www.flickr.com/photos/68147320@N02/28380794150,https://creativecommons.org/licenses/by/2.0/,T.Tseng,https://www.flickr.com/people/68147320@N02/\r\ndacb07f80166a7fd,https://c1.staticflickr.com/6/5813/30874086821_c5bcdcb048_o.jpg,0.5,0,1,0.75,449,108,http://www.flickr.com/photos/boykoblagoev/30874086821/,https://creativecommons.org/licenses/by/2.0/,Boyko Blagoev,https://www.flickr.com/people/boykoblagoev/\r\n82b6f0bc71f5d0af,https://c1.staticflickr.com/4/3943/15734053815_ccb8cb6852_o.jpg,0,0,1,1,563,578,https://www.flickr.com/photos/zedzap/15734053815/,https://creativecommons.org/licenses/by/2.0/,Nick Kenrick,https://www.flickr.com/people/zedzap/\r\n0f36316247c03427,https://c1.staticflickr.com/1/504/31690529866_dc4e0a80a7_o.jpg,0.500736377,0.25,1,1,753,681,https://www.flickr.com/photos/squashdev/31690529866,https://creativecommons.org/licenses/by/2.0/,England Squash Masters,https://www.flickr.com/people/squashdev/\r\nefeabdd8a05e91f8,https://c1.staticflickr.com/9/8810/28183540151_f7108aa9a2_o.jpg,0,0,0.5625,0.75,498,311,https://www.flickr.com/photos/131810676@N03/28183540151/,https://creativecommons.org/licenses/by/2.0/,baka_neko_baka,https://www.flickr.com/people/131810676@N03/\r\n7453471febf1db6f,https://c1.staticflickr.com/9/8870/28331839720_9de4810a16_o.jpg,0,0.333333333,1,1,595,90,https://www.flickr.com/photos/stevenpisano/28331839720,https://creativecommons.org/licenses/by/2.0/,Steven Pisano,https://www.flickr.com/people/stevenpisano/\r\ne6f2de0c605f1ed3,https://c1.staticflickr.com/9/8392/29438273202_8ba928f59f_o.jpg,0.471590909,0,1,0.75,132,994,https://www.flickr.com/photos/dkeats/29438273202,https://creativecommons.org/licenses/by/2.0/,Derek Keats,https://www.flickr.com/people/dkeats/\r\n648231d9393250fd,https://c1.staticflickr.com/1/436/31994770622_64e6478fe4_o.jpg,0.578822733,0.25,1,1,548,650,https://www.flickr.com/photos/belurashok/31994770622/,https://creativecommons.org/licenses/by/2.0/,Belur Ashok,https://www.flickr.com/people/belurashok/\r\n864fffaf7a83b3f4,https://c1.staticflickr.com/8/7471/27863056826_bc9e568c4c_o.jpg,0,0,0.75,1,581,373,https://www.flickr.com/photos/juggernautco/27863056826,https://creativecommons.org/licenses/by/2.0/,Daniel X. O'Neil,https://www.flickr.com/people/juggernautco/\r\nc179282bd977c6e0,https://c1.staticflickr.com/3/2907/32124474993_c63801fabf_o.jpg,0,0,0.556376361,0.75,672,105,https://www.flickr.com/photos/glorycycles/32124474993,https://creativecommons.org/licenses/by/2.0/,Glory Cycles,https://www.flickr.com/people/glorycycles/\r\n40a7f086461190f3,https://c1.staticflickr.com/2/1514/26690160211_eda12ace1e_o.jpg,0.42862776,0.25,1,1,493,841,https://www.flickr.com/photos/martius/26690160211,https://creativecommons.org/licenses/by/2.0/,M a n u e l,https://www.flickr.com/people/martius/\r\n90e1aa56cae80a60,https://c1.staticflickr.com/6/5068/30114102505_bcc8ec61db_o.jpg,0,0,1,1,918,417,https://flickr.com/64648244@N04/30114102505,https://creativecommons.org/licenses/by/2.0/,Ulf K,https://www.flickr.com/people/ulfk/\r\n839aa3d166b020cf,https://c1.staticflickr.com/8/7104/26514437520_58ea29102e_o.jpg,0,0.25,0.72079646,1,14,211,https://www.flickr.com/photos/nrgoodrum/26514437520,https://creativecommons.org/licenses/by/2.0/,nick goodrum,https://www.flickr.com/people/nrgoodrum/\r\n38bdd950c1d08be6,https://c1.staticflickr.com/1/467/32470039506_2c9d8f3c2c_o.jpg,0,0,0.75,1,133,891,https://www.flickr.com/photos/rusty_clark/32470039506/,https://creativecommons.org/licenses/by/2.0/,Rusty Clark ~ 100K Photos,https://www.flickr.com/people/rusty_clark/\r\n2d95297e468bdf9b,https://c1.staticflickr.com/8/7358/27718678441_390843b0a4_o.jpg,0.28125,0.25,0.84375,1,866,526,https://www.flickr.com/photos/jeepersmedia/27718678441,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\nef5b1e5a76ae66dc,https://c1.staticflickr.com/9/8028/29706482742_e07ecf46e1_o.jpg,0.497426471,0,0.994852941,0.75,540,725,https://www.flickr.com/photos/emilysnuffer/29706482742,https://creativecommons.org/licenses/by/2.0/,Emily May,https://www.flickr.com/people/emilysnuffer/\r\n7a11ae4c1f5e790c,https://c1.staticflickr.com/8/7680/28813894581_61251a80fa_o.jpg,0.249079529,0.25,0.747238586,1,689,500,https://www.flickr.com/photos/zigazou76/28813894581,https://creativecommons.org/licenses/by/2.0/,Fr\\303\\251d\\303\\251ric BISSON,https://www.flickr.com/people/zigazou76/\r\n477bf90e7413df23,https://c1.staticflickr.com/8/7338/27067101244_7942b5769e_o.jpg,0,0,0.578296703,1,542,836,https://www.flickr.com/photos/fifeanddrum/27067101244,https://creativecommons.org/licenses/by/2.0/,U. S. Army Old Guard Fife and Drum Corps,https://www.flickr.com/people/fifeanddrum/\r\n7060ae82e5b96a77,https://c1.staticflickr.com/8/7414/27849533291_23d512cc58_o.jpg,0,0,0.75,1,522,495,https://www.flickr.com/photos/jeepersmedia/27849533291,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\n42e55ac6782d4d99,https://c1.staticflickr.com/9/8841/28851410876_d279c09ca0_o.jpg,0,0,0.5625,0.75,974,819,https://flickr.com/131967103@N02/28851410876,https://creativecommons.org/licenses/by/2.0/,Ryan McMinds,https://www.flickr.com/people/131967103@N02/\r\n8868879c2acd5d9e,https://c1.staticflickr.com/8/7477/27387672956_baa0cdeb89_o.jpg,0.455384615,0,1,0.75,322,130,https://www.flickr.com/photos/usfwsmidwest/27387672956,https://creativecommons.org/licenses/by/2.0/,USFWSmidwest,https://www.flickr.com/people/usfwsmidwest/\r\n5b3a8c63e41802e7,https://c1.staticflickr.com/9/8883/27865729823_3db56871fa_o.jpg,0.25,0.25,1,1,750,758,https://flickr.com/38315261@N00/27865729823,https://creativecommons.org/licenses/by/2.0/,Sharon Mollerus,https://www.flickr.com/people/clairity/\r\n659ca4ac6cc57e89,https://c1.staticflickr.com/1/371/31418192282_a6fa2c7a51_o.jpg,0.25,0,0.75,0.75,778,590,https://www.flickr.com/photos/quinet/31418192282,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n9534108b0e8dceb1,https://c1.staticflickr.com/1/668/31683586152_9c8d97ef8f_o.jpg,0.282719436,0.25,0.848158307,1,647,822,https://www.flickr.com/photos/arnehendriks/31683586152,https://creativecommons.org/licenses/by/2.0/,Arne Hendriks,https://www.flickr.com/people/arnehendriks/\r\nfbc6b98b13c768e2,https://c1.staticflickr.com/9/8435/29171057840_4acaf34304_o.jpg,0,0.25,0.5625,1,459,566,https://www.flickr.com/photos/rheba/29171057840,https://creativecommons.org/licenses/by/2.0/,Rheba,https://www.flickr.com/people/rheba/\r\n7c3b0d93654b39cd,https://c1.staticflickr.com/6/5128/29789328711_639198fc6f_o.jpg,0.25,0,1,0.5625,708,428,https://www.flickr.com/photos/playamongfriends/29789328711,https://creativecommons.org/licenses/by/2.0/,Play Among Friends Paf,https://www.flickr.com/people/playamongfriends/\r\n43694caa12d5f823,https://c1.staticflickr.com/1/445/31614966550_d5fc6f7f3d_o.jpg,0.28125,0.25,0.84375,1,979,757,https://www.flickr.com/photos/cambodia4kidsorg/31614966550/,https://creativecommons.org/licenses/by/2.0/,Cambodia4kids.org Beth Kanter,https://www.flickr.com/people/cambodia4kidsorg/\r\n69788a29e7436808,https://c1.staticflickr.com/6/5332/30849701300_a85871c357_o.jpg,0.12962963,0,1,1,310,50,https://www.flickr.com/photos/gails_pictures/30849701300/,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\ndaa43d65c3b18a84,https://c1.staticflickr.com/9/8253/29381659730_52bcbf8b59_o.jpg,0,0,1,0.75,822,54,https://www.flickr.com/photos/39551170@N02/29381659730,https://creativecommons.org/licenses/by/2.0/,Simon_sees,https://www.flickr.com/people/39551170@N02/\r\na8c3363bbcc7015c,https://c1.staticflickr.com/9/8823/29472376703_be7d867566_o.jpg,0,0,0.912068966,1,739,840,https://www.flickr.com/photos/tomronworldwide/29472376703,https://creativecommons.org/licenses/by/2.0/,Ron Frazier,https://www.flickr.com/people/tomronworldwide/\r\n93a0857ff81c25ce,https://c1.staticflickr.com/6/5733/30469514315_1d8ce7f4c2_o.jpg,0,0,0.5,0.75,548,959,https://www.flickr.com/photos/rhemkes/30469514315,https://creativecommons.org/licenses/by/2.0/,Roel Hemkes,https://www.flickr.com/people/rhemkes/\r\nb11fa0f292c0178f,https://c1.staticflickr.com/1/780/32263240380_ee5c7a9637_o.jpg,0.25,0,0.75,0.75,702,901,http://www.flickr.com/photos/124790945@N06/32263240380,https://creativecommons.org/licenses/by/2.0/,Phuket@photographer.net,https://www.flickr.com/people/linvoyage/\r\n1dc0533f87996962,https://c1.staticflickr.com/8/7704/26497012284_30e830cc10_o.jpg,0.409855769,0.25,1,1,566,853,https://www.flickr.com/photos/jsjgeology/26497012284,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n589319c64adf5fce,https://c1.staticflickr.com/8/7460/27398216034_b731c9e68a_o.jpg,0.333333333,0,1,1,699,588,https://www.flickr.com/photos/libereurope/27398216034/,https://creativecommons.org/licenses/by/2.0/,LIBER Europe,https://www.flickr.com/people/libereurope/\r\ncebd45d7200f6111,https://c1.staticflickr.com/9/8339/29164491280_b7d3bdd713_o.jpg,0.25,0,1,0.719444444,71,947,https://www.flickr.com/photos/garlandcannon/29164491280,https://creativecommons.org/licenses/by/2.0/,garlandcannon,https://www.flickr.com/people/garlandcannon/\r\n7e73c7adc520e406,https://c1.staticflickr.com/8/7416/26859649600_fb2cbaf86b_o.jpg,0.25,0,1,0.64807047,133,86,https://www.flickr.com/photos/lakeworth/26859649600,https://creativecommons.org/licenses/by/2.0/,Henry,https://www.flickr.com/people/lakeworth/\r\n6de6432b0f23ae75,https://c1.staticflickr.com/1/579/32182435510_e9fab3e93c_o.jpg,0.333333333,0,1,1,105,407,https://www.flickr.com/photos/rexness/32182435510,https://creativecommons.org/licenses/by/2.0/,Rexness,https://www.flickr.com/people/rexness/\r\nebb4dad78cfd67a9,https://c1.staticflickr.com/8/7293/27173553526_fe84eeeec9_o.jpg,0.25,0,1,1,738,730,https://www.flickr.com/photos/jeepersmedia/27173553526,https://creativecommons.org/licenses/by/2.0/,Mike Mozart,https://www.flickr.com/people/jeepersmedia/\r\ne8c6f75ad26bae84,https://c1.staticflickr.com/9/8393/29734167231_b03b36f04b_o.jpg,0.28125,0.25,0.84375,1,973,929,https://www.flickr.com/photos/jsjgeology/29734167231,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n1e408a8d1e2cdc74,https://c1.staticflickr.com/6/5509/30181862974_0a8d40efb1_o.jpg,0.4375,0,1,0.75,769,779,https://www.flickr.com/photos/stmunchins/30181862974/,https://creativecommons.org/licenses/by/2.0/,St Munchin's College,https://www.flickr.com/people/stmunchins/\r\n11afa51200515f9a,https://c1.staticflickr.com/8/7654/28347963084_4a9385edd0_o.jpg,0.263089005,0,0.789267016,1,826,567,https://www.flickr.com/photos/anandajoti/28347963084/,https://creativecommons.org/licenses/by/2.0/,Photo Dharma,https://www.flickr.com/people/anandajoti/\r\n9f6fe669276f04cf,https://c1.staticflickr.com/9/8760/28674693282_7eec755334_o.jpg,0,0,0.7484375,1,920,697,https://www.flickr.com/photos/zimpenfish/28674693282,https://creativecommons.org/licenses/by/2.0/,rjp,https://www.flickr.com/people/zimpenfish/\r\n408bae58c26a8076,https://c1.staticflickr.com/6/5773/29699525424_ea6336aaa8_o.jpg,0.462155963,0,1,0.75,133,761,https://www.flickr.com/photos/taylar/29699525424,https://creativecommons.org/licenses/by/2.0/,Ingrid Taylar,https://www.flickr.com/people/taylar/\r\nb5548ecffa20aac8,https://c1.staticflickr.com/1/713/32057543825_ded4304f3b_o.jpg,0,0,0.5,0.75,934,108,http://www.flickr.com/photos/143850343@N06/32057543825/,https://creativecommons.org/licenses/by/2.0/,Sam-H-A,https://www.flickr.com/people/143850343@N06/\r\n9ee13ecd8d6727c0,https://c1.staticflickr.com/6/5708/30313299300_68025e82ed_o.jpg,0,0.498153619,0.75,1,498,554,https://www.flickr.com/photos/quinet/30313299300,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\nd1a723656dc08b5e,https://c1.staticflickr.com/8/7407/27895921745_908d139fdd_o.jpg,0.503303965,0.25,1,1,822,814,https://www.flickr.com/photos/shankaronline/27895921745,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\nd1709eec353a6f93,https://c1.staticflickr.com/8/7421/27144655490_a81fccf9ca_o.jpg,0,0.25,0.75,1,297,556,https://www.flickr.com/photos/rjbailey/27144655490/,https://creativecommons.org/licenses/by/2.0/,Ron Bailey,https://www.flickr.com/people/rjbailey/\r\nb89dbfcc49dcf156,https://c1.staticflickr.com/8/7448/27273015613_f2767e3f98_o.jpg,0.210588633,0,0.6317659,0.75,889,467,https://www.flickr.com/photos/spencerhall/27273015613,https://creativecommons.org/licenses/by/2.0/,Spencer Hall,https://www.flickr.com/people/spencerhall/\r\n7368fe0ff47343a0,https://c1.staticflickr.com/8/7520/26512614104_b123959864_o.jpg,0.28125,0.25,0.84375,1,965,447,https://www.flickr.com/photos/rainsong/26512614104,https://creativecommons.org/licenses/by/2.0/,Ali - \\330\\271\\331\\204\\331\\212 from Riyadh - \\331\\205\\331\\206 \\330\\247\\331\\204\\330\\261\\331\\212\\330\\247\\330\\266,https://www.flickr.com/people/rainsong/\r\nd23fd2119c4f5dec,https://c1.staticflickr.com/9/8165/28790897384_31003e57ee_o.jpg,0.249631811,0,0.748895434,0.75,329,123,https://www.flickr.com/photos/fhgitarre/28790897384,https://creativecommons.org/licenses/by/2.0/,FHgitarre,https://www.flickr.com/people/fhgitarre/\r\n5a0e21d54b9c4954,https://c1.staticflickr.com/8/7243/27968268826_09285822b1_o.jpg,0,0,0.715608466,0.75,328,146,https://www.flickr.com/photos/noaaphotolib/27968268826,https://creativecommons.org/licenses/by/2.0/,NOAA Photo Library,https://www.flickr.com/people/noaaphotolib/\r\n4b8517e927f32962,https://c1.staticflickr.com/9/8140/29892684285_e0633bea72_o.jpg,0.25,0.25,1,1,866,271,https://flickr.com/10307586@N03/29892684285,https://creativecommons.org/licenses/by/2.0/,Mark Pegrum,https://www.flickr.com/people/ozmark17/\r\n38deb4830d12358f,https://c1.staticflickr.com/6/5650/31182154435_7dc98f5dba_o.jpg,0.5,0,1,0.75,980,936,https://www.flickr.com/photos/henry_hemming/31182154435,https://creativecommons.org/licenses/by/2.0/,Henry Hemming,https://www.flickr.com/people/henry_hemming/\r\n81804435ad51df95,https://c1.staticflickr.com/1/255/31787555660_d866250efd_o.jpg,0,0,0.5,0.75,437,241,https://www.flickr.com/photos/nielsautos/31787555660,https://creativecommons.org/licenses/by/2.0/,Niels de Wit,https://www.flickr.com/people/nielsautos/\r\ne4941a341ace171a,https://c1.staticflickr.com/9/8006/28587599723_947de8d314_o.jpg,0.25,0.25,1,0.75,709,298,https://www.flickr.com/photos/asrusch/28587599723,https://creativecommons.org/licenses/by/2.0/,Andy Rusch,https://www.flickr.com/people/asrusch/\r\n558a7b66b78ba65b,https://c1.staticflickr.com/6/5797/30174833703_4e5f61ded4_o.jpg,0,0,1,0.745709828,422,668,https://www.flickr.com/photos/tusnelda/30174833703/,https://creativecommons.org/licenses/by/2.0/,storebukkebruse,https://www.flickr.com/people/tusnelda/\r\n435137d9d4f4125e,https://c1.staticflickr.com/8/7535/28182425121_c71755b34e_o.jpg,0,0,0.75,0.75,2,269,https://www.flickr.com/photos/aidanmorgan/28182425121/,https://creativecommons.org/licenses/by/2.0/,John Morgan,https://www.flickr.com/people/aidanmorgan/\r\nc43077b2283479be,https://c1.staticflickr.com/6/5522/30653895580_2c136ce8c7_o.jpg,0,0,0.75,0.588,14,765,https://flickr.com/93242958@N00/30653895580,https://creativecommons.org/licenses/by/2.0/,Derek Keats,https://www.flickr.com/people/dkeats/\r\n77c72d0a01e1af70,https://c1.staticflickr.com/9/8214/28891107093_72da720a1d_o.jpg,0.5,0,1,0.75,428,836,https://www.flickr.com/photos/katiethebeau/28891107093/,https://creativecommons.org/licenses/by/2.0/,KatieThebeau,https://www.flickr.com/people/katiethebeau/\r\n72ac2a5af79e26d5,https://c1.staticflickr.com/6/5613/31220748780_31a537d326_o.jpg,0.28125,0,0.84375,0.75,718,862,https://www.flickr.com/photos/hugo90/31220748780,https://creativecommons.org/licenses/by/2.0/,JOHN LLOYD,https://www.flickr.com/people/hugo90/\r\n32c0e4647ce5a812,https://c1.staticflickr.com/9/8344/29851489791_076d3494a0_o.jpg,0.486899563,0,0.973799127,0.75,327,743,https://www.flickr.com/photos/yogendra174/29851489791,https://creativecommons.org/licenses/by/2.0/,Yogendra Joshi,https://www.flickr.com/people/yogendra174/\r\n54153d17b11b495c,https://c1.staticflickr.com/6/5500/30389728162_23cdff8ba8_o.jpg,0.421177267,0,0.842354533,0.75,657,551,https://www.flickr.com/photos/micsworld/30389728162,https://creativecommons.org/licenses/by/2.0/,Mic,https://www.flickr.com/people/micsworld/\r\nfb1efda0b27d5289,https://c1.staticflickr.com/1/460/31194422753_24550f0615_o.jpg,0.5,0,1,0.75,848,203,https://www.flickr.com/photos/quinet/31194422753,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n7c5ea6f89e6e2733,https://c1.staticflickr.com/1/454/32710889631_00f86c1d71_o.jpg,0.500736377,0,1,0.75,820,587,https://www.flickr.com/photos/dominik_junger/32710889631/in/album-72157676671651883/,https://creativecommons.org/licenses/by/2.0/,Dominik Junger,https://www.flickr.com/people/dominik_junger/\r\n721804d1d3a60206,https://c1.staticflickr.com/8/7790/27456693684_f07ebe3aea_o.jpg,0,0,0.666666667,1,909,34,https://www.flickr.com/photos/presidentialoffice/27456693684,https://creativecommons.org/licenses/by/2.0/,\\347\\270\\275\\347\\265\\261\\345\\272\\234,https://www.flickr.com/people/presidentialoffice/\r\ndcb6b32cd80f2b1f,https://c1.staticflickr.com/8/7560/29512813082_6efd1c1b56_o.jpg,0.28125,0,0.84375,0.75,556,832,https://www.flickr.com/photos/27665395@N05/29512813082/,https://creativecommons.org/licenses/by/2.0/,Jason Lawrence,https://www.flickr.com/people/27665395@N05/\r\n49c27b439d39a00c,https://c1.staticflickr.com/1/653/33034701806_914058174b_o.jpg,0.335787923,0,1,1,490,418,https://flickr.com/38144691@N00/33034701806,https://creativecommons.org/licenses/by/2.0/,August Kelm,https://www.flickr.com/people/augustksphotos/\r\n2c065d2dc6b2cf44,https://c1.staticflickr.com/6/5441/30476315260_213dbcbdf3_o.jpg,0.439547582,0,1,0.75,934,463,https://www.flickr.com/photos/hmk/30476315260,https://creativecommons.org/licenses/by/2.0/,H. Michael Karshis,https://www.flickr.com/people/hmk/\r\nb3f200c672c5ca4f,https://c1.staticflickr.com/8/7539/28925572132_ff6fc934e0_o.jpg,0,0.25,0.5,1,976,275,https://www.flickr.com/photos/gabrielgm/28925572132/,https://creativecommons.org/licenses/by/2.0/,Gabriel Garcia Marengo,https://www.flickr.com/people/gabrielgm/\r\n3f40eba168e5d385,https://c1.staticflickr.com/6/5717/31106619226_f60452e9f2_o.jpg,0.299878837,0,0.899636511,0.75,515,936,https://www.flickr.com/photos/85546319@N04/31106619226,https://creativecommons.org/licenses/by/2.0/,Robert Sheie,https://www.flickr.com/people/85546319@N04/\r\n0e71473957df4793,https://c1.staticflickr.com/1/630/31560462464_df68f43299_o.jpg,0.502573529,0,1,0.75,878,348,https://www.flickr.com/photos/agenciabrasilia/31560462464,https://creativecommons.org/licenses/by/2.0/,Ag\\303\\252ncia Bras\\303\\255lia,https://www.flickr.com/people/agenciabrasilia/\r\n6c9f8901246faff2,https://c1.staticflickr.com/1/745/31509879035_78978e4f01_o.jpg,0,0,0.708206687,1,629,689,https://www.flickr.com/photos/guldem/31509879035,https://creativecommons.org/licenses/by/2.0/,G\\303\\274ldem \\303\\234st\\303\\274n,https://www.flickr.com/people/guldem/\r\na64ae1acd9d09a87,https://c1.staticflickr.com/1/717/31077768123_e0787a8de6_o.jpg,0,0,1,1,131,83,http://www.flickr.com/photos/paul_e_balfe/31077768123/,https://creativecommons.org/licenses/by/2.0/,Paul Balfe,https://www.flickr.com/people/paul_e_balfe/\r\n05f7c084329ff0c6,https://c1.staticflickr.com/6/5513/30518437054_80b12bbda5_o.jpg,0.5,0.25,1,1,973,910,https://flickr.com/14247646@N00/30518437054,https://creativecommons.org/licenses/by/2.0/,James,https://www.flickr.com/people/jametiks/\r\n5d77597361d06d2b,https://c1.staticflickr.com/1/478/31414603844_530a364312_o.jpg,0,0,1,0.714503817,904,92,https://www.flickr.com/photos/avedaphotos/31414603844/,https://creativecommons.org/licenses/by/2.0/,Aveda Corporation,https://www.flickr.com/people/avedaphotos/\r\nac2a8a4777dbe746,https://c1.staticflickr.com/8/7615/27545901420_caafb85d83_o.jpg,0,0,0.499263623,0.75,896,46,https://www.flickr.com/photos/621crw/27545901420,https://creativecommons.org/licenses/by/2.0/,The 621st Contingency Response Wing,https://www.flickr.com/people/621crw/\r\n3d75d9676ad6361c,https://c1.staticflickr.com/6/5589/30796308216_2ba813235e_o.jpg,0,0,0.601941748,0.75,151,517,https://www.flickr.com/photos/pandora_6666/30796308216,https://creativecommons.org/licenses/by/2.0/,Jo Naylor,https://www.flickr.com/people/pandora_6666/\r\ne3a740861f12f49e,https://c1.staticflickr.com/9/8453/29471934070_b712c35d65_o.jpg,0.210873984,0,0.632621951,0.75,820,900,https://www.flickr.com/photos/ceta_und_ttip_stoppen/29471934070/,https://creativecommons.org/licenses/by/2.0/,CETA und TTIP stoppen,https://www.flickr.com/people/ceta_und_ttip_stoppen/\r\n64577d3916d6dafb,https://c1.staticflickr.com/8/7774/26695019922_97087ac2c0_o.jpg,0.25,0,1,0.5625,606,754,https://www.flickr.com/photos/afropicmusing/26695019922,https://creativecommons.org/licenses/by/2.0/,afromusing,https://www.flickr.com/people/afropicmusing/\r\n3b32eca218ed3dbe,https://c1.staticflickr.com/8/7341/28021300931_378cfe5932_o.jpg,0,0,0.490510949,0.75,742,838,https://www.flickr.com/photos/jdlasica/28021300931,https://creativecommons.org/licenses/by/2.0/,JD Lasica,https://www.flickr.com/people/jdlasica/\r\n0707f3dfb7c29f5c,https://c1.staticflickr.com/6/5700/30555623662_42d83db0d3_o.jpg,0,0.25,0.498159057,1,606,740,https://www.flickr.com/photos/145304628@N02/30555623662,https://creativecommons.org/licenses/by/2.0/,Degree53 UK,https://www.flickr.com/people/145304628@N02/\r\n59e9bdbb7a3fc4ba,https://c1.staticflickr.com/9/8533/30011107125_37837842a9_o.jpg,0,0,1,0.666666667,518,407,https://www.flickr.com/photos/zigazou76/30011107125,https://creativecommons.org/licenses/by/2.0/,Fr\\303\\251d\\303\\251ric BISSON,https://www.flickr.com/people/zigazou76/\r\n686ea4ec999364ff,https://c1.staticflickr.com/8/7319/27523198984_f9d44bba3a_o.jpg,0,0.28125,0.75,0.84375,459,554,https://www.flickr.com/photos/nostri-imago/27523198984,https://creativecommons.org/licenses/by/2.0/,Cliff,https://www.flickr.com/people/nostri-imago/\r\ncb71a056a8e05bb2,https://c1.staticflickr.com/9/8001/29254381704_cc81ff8df9_o.jpg,0,0,0.563380282,0.75,320,941,https://flickr.com/9243304@N04/29254381704,https://creativecommons.org/licenses/by/2.0/,Amante Darmanin,https://www.flickr.com/people/amantedar/\r\nf63d58b5af16091f,https://c1.staticflickr.com/1/681/32382788261_b7554c250b_o.jpg,0,0,1,0.732612056,716,420,https://www.flickr.com/photos/jlcernadas/32382788261/,https://creativecommons.org/licenses/by/2.0/,Jose Luis Cernadas Iglesias,https://www.flickr.com/people/jlcernadas/\r\nd3aa8ee50458c020,https://c1.staticflickr.com/1/444/32463408492_cf7ab6f5fb_o.jpg,0,0,0.75,0.630794702,431,874,https://www.flickr.com/photos/viumariners/32463408492/,https://creativecommons.org/licenses/by/2.0/,VIU Mariners,https://www.flickr.com/people/viumariners/\r\n2891f4aa2279d369,https://c1.staticflickr.com/9/8658/28224525812_7301285f51_o.jpg,0.5,0,1,0.75,471,405,https://www.flickr.com/photos/rorals/28224525812,https://creativecommons.org/licenses/by/2.0/,Rory,https://www.flickr.com/people/rorals/\r\nae24c23192976a40,https://c1.staticflickr.com/6/5811/31350835411_b856c2e0f7_o.jpg,0,0,0.498159057,0.75,752,888,https://www.flickr.com/photos/davehamster/31350835411,https://creativecommons.org/licenses/by/2.0/,David Merrett,https://www.flickr.com/people/davehamster/\r\n348ac56def3ccd35,https://c1.staticflickr.com/9/8601/30053184852_b6c871293b_o.jpg,0.331617647,0,0.994852941,1,916,533,https://www.flickr.com/photos/alpesdehauteprovence-tourisme/30053184852,https://creativecommons.org/licenses/by/2.0/,ADT 04,https://www.flickr.com/people/alpesdehauteprovence-tourisme/\r\nd46216d31908163e,https://c1.staticflickr.com/9/8723/28309396545_349e5147bb_o.jpg,0.333333333,0,1,1,896,785,https://www.flickr.com/photos/ronmacphotos/28309396545,https://creativecommons.org/licenses/by/2.0/,Ronnie Macdonald,https://www.flickr.com/people/ronmacphotos/\r\na8dd9ba15f7b817b,https://c1.staticflickr.com/1/575/31572326940_d87b136e59_o.jpg,0.251849112,0.25,0.755547337,1,806,740,https://www.flickr.com/photos/prefeituradeolinda/31572326940,https://creativecommons.org/licenses/by/2.0/,Prefeitura de Olinda,https://www.flickr.com/people/prefeituradeolinda/\r\n5224fec9ba33b50d,https://c1.staticflickr.com/9/8081/29568549431_86f962156e_o.jpg,0,0,0.75,1,604,532,https://flickr.com/28834651@N07/29568549431,https://creativecommons.org/licenses/by/2.0/,Aul Rah,https://www.flickr.com/people/28834651@N07/\r\nb3c384a2d1a391c0,https://c1.staticflickr.com/6/5543/30815777131_13d5e7667c_o.jpg,0,0,0.599757674,0.75,631,456,https://www.flickr.com/photos/85546319@N04/30815777131,https://creativecommons.org/licenses/by/2.0/,Robert Sheie,https://www.flickr.com/people/85546319@N04/\r\n4a1a488e8a975f7c,https://c1.staticflickr.com/9/8420/29353940465_05ece9b92e_o.jpg,0.25,0.25,0.75,1,920,833,https://www.flickr.com/photos/harshlight/29353940465,https://creativecommons.org/licenses/by/2.0/,HarshLight,https://www.flickr.com/people/harshlight/\r\nd7c1d1632cf0acc7,https://c1.staticflickr.com/9/8653/28465398416_2c4202cb8d_o.jpg,0.25,0,0.75,0.75,404,873,https://www.flickr.com/photos/usnavy/28465398416,https://creativecommons.org/licenses/by/2.0/,Official U.S. Navy Page,https://www.flickr.com/people/usnavy/\r\nca50e3535eb6396c,https://c1.staticflickr.com/6/5527/29526798504_12061e1707_o.jpg,0,0,0.5,0.75,556,929,https://flickr.com/137837832@N07/29526798504,https://creativecommons.org/licenses/by/2.0/,Mark Holloway,https://www.flickr.com/people/137837832@N07/\r\n30e132285c2e960f,https://c1.staticflickr.com/6/5663/30553419670_2249584a31_o.jpg,0.25,0,1,1,912,427,https://www.flickr.com/photos/jonathanrolande/30553419670,https://creativecommons.org/licenses/by/2.0/,Jonathan Rolande,https://www.flickr.com/people/jonathanrolande/\r\nea65c0ff71e7ceb3,https://c1.staticflickr.com/4/3953/33320124506_2fbd93baa7_o.jpg,0.333333333,0,1,1,404,737,https://www.flickr.com/photos/usnavy/33320124506,https://creativecommons.org/licenses/by/2.0/,Official U.S. Navy Page,https://www.flickr.com/people/usnavy/\r\n200285e90078210b,https://c1.staticflickr.com/1/701/30876765453_574a6815b0_o.jpg,0.5,0.25,1,1,442,741,https://flickr.com/22539273@N00/30876765453,https://creativecommons.org/licenses/by/2.0/,Guilhem Vellut,https://www.flickr.com/people/o_0/\r\ncec46dbd1fa3473a,https://c1.staticflickr.com/9/8320/29351169445_b0b0090587_o.jpg,0.455696203,0.25,0.911392405,1,898,246,https://www.flickr.com/photos/lge/29351169445,https://creativecommons.org/licenses/by/2.0/,LG\\354\\240\\204\\354\\236\\220,https://www.flickr.com/people/lge/\r\n6314b953a2f66e36,https://c1.staticflickr.com/1/306/31686321825_ef08b80af5_o.jpg,0.281690141,0.25,0.845070423,1,311,774,https://www.flickr.com/photos/treegrow/31686321825,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\n1f3f949ccbc81c93,https://c1.staticflickr.com/9/8643/28179475136_e6925b8a4b_o.jpg,0,0.25,1,1,304,666,https://flickr.com/27305106@N06/28179475136,https://creativecommons.org/licenses/by/2.0/,Charlie Jackson,https://www.flickr.com/people/chaz_pics/\r\n04546cc5e537d7cd,https://c1.staticflickr.com/8/7347/27291550531_ee40bdf58a_o.jpg,0,0,0.495594714,0.75,426,649,https://www.flickr.com/photos/kansasscanner/27291550531/,https://creativecommons.org/licenses/by/2.0/,Tyler Silvest,https://www.flickr.com/people/kansasscanner/\r\n73064ceb466612d5,https://c1.staticflickr.com/6/5777/30172206492_407d68dc21_o.jpg,0.53125,0,1,0.75,719,673,https://flickr.com/97760755@N08/30172206492,https://creativecommons.org/licenses/by/2.0/,Viaggio Routard,https://www.flickr.com/people/viaggioroutard/\r\nae78688b32986638,https://c1.staticflickr.com/9/8290/29399680281_1b2c136428_o.jpg,0,0,0.5,0.75,913,879,https://www.flickr.com/photos/ddebold/29399680281/,https://creativecommons.org/licenses/by/2.0/,Don DeBold,https://www.flickr.com/people/ddebold/\r\n72d85232abd98067,https://c1.staticflickr.com/8/7655/28880045512_841b0c2d3a_o.jpg,0.25,0,0.75,0.75,471,599,https://www.flickr.com/photos/quinet/28880045512,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n89d8361c6bfe905c,https://c1.staticflickr.com/8/7290/27369075485_5c164b1839_o.jpg,0.25,0,0.75,0.75,604,51,https://www.flickr.com/photos/quinet/27369075485,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\n4ca02fd28a5e9703,https://c1.staticflickr.com/9/8453/29423857932_71a8a4a98f_o.jpg,0.25,0.302066451,1,0.906199352,996,239,https://www.flickr.com/photos/92252798@N07/29423857932,https://creativecommons.org/licenses/by/2.0/,Dick Culbert,https://www.flickr.com/people/92252798@N07/\r\n48762c2cd2ed47e1,https://c1.staticflickr.com/6/5546/30817355150_7f1f77f6e2_o.jpg,0.25,0,1,0.75,755,308,https://www.flickr.com/photos/hmk/30817355150,https://creativecommons.org/licenses/by/2.0/,H. Michael Karshis,https://www.flickr.com/people/hmk/\r\nb217644382dcd864,https://c1.staticflickr.com/9/8082/29684292096_85e6a733cf_o.jpg,0.182707993,0,1,1,512,708,https://www.flickr.com/photos/atoach/29684292096,https://creativecommons.org/licenses/by/2.0/,Tim Green,https://www.flickr.com/people/atoach/\r\nc4824f023903731a,https://c1.staticflickr.com/8/7777/26400069864_79fd47a2fe_o.jpg,0.25,0.25,1,1,738,817,https://www.flickr.com/photos/betsyweber/26400069864,https://creativecommons.org/licenses/by/2.0/,Betsy Weber,https://www.flickr.com/people/betsyweber/\r\n629138193e021cdf,https://c1.staticflickr.com/9/8768/28328433276_2d2146df81_o.jpg,0,0.5,0.75,1,885,739,https://www.flickr.com/photos/glenbowman/28328433276,https://creativecommons.org/licenses/by/2.0/,Glen Bowman,https://www.flickr.com/people/glenbowman/\r\n7d4186ce0332245d,https://c1.staticflickr.com/9/8578/28958405495_50b6230377_o.jpg,0,0,0.666666667,1,976,936,https://www.flickr.com/photos/o_0/28958405495,https://creativecommons.org/licenses/by/2.0/,Guilhem Vellut,https://www.flickr.com/people/o_0/\r\n90dfd75887f3e67b,https://c1.staticflickr.com/8/7683/26182602974_ddb15b42a7_o.jpg,0,0.25,1,1,986,45,https://www.flickr.com/photos/110035327@N08/26182602974,https://creativecommons.org/licenses/by/2.0/,Stefano Akram,https://www.flickr.com/people/110035327@N08/\r\ned56c7616a8ac010,https://c1.staticflickr.com/8/7479/27427361256_e341e77ffe_o.jpg,0.267938931,0,0.803816794,0.75,782,568,https://www.flickr.com/photos/aperturenut/27427361256,https://creativecommons.org/licenses/by/2.0/,Rick Briggs,https://www.flickr.com/people/aperturenut/\r\n5d6b9cadb6f9e1cd,https://c1.staticflickr.com/9/8790/28885989621_1e024392cc_o.jpg,0,0,0.5,0.75,896,746,https://www.flickr.com/photos/elsie/28885989621/,https://creativecommons.org/licenses/by/2.0/,Les Chatfield,https://www.flickr.com/people/elsie/\r\na033832d9ff8913b,https://c1.staticflickr.com/1/545/32710656176_55231d6054_o.jpg,0,0,0.713414634,1,101,357,https://www.flickr.com/photos/wildreturn/32710656176/,https://creativecommons.org/licenses/by/2.0/,Andy Reago & Chrissy McClarren,https://www.flickr.com/people/wildreturn/\r\n313ff0037b86260a,https://c1.staticflickr.com/6/5812/30890228762_9ffa0a8fcb_o.jpg,0,0.2515625,1,1,498,610,https://www.flickr.com/photos/southbeachcars/30890228762/,https://creativecommons.org/licenses/by/2.0/,Phillip Pessar,https://www.flickr.com/people/southbeachcars/\r\nafe045d68f1a60d0,https://c1.staticflickr.com/6/5496/30030140344_d59046fc37_o.jpg,0.281690141,0,0.845070423,0.75,990,956,https://flickr.com/72054607@N00/30030140344,https://creativecommons.org/licenses/by/2.0/,Mark Jones,https://www.flickr.com/people/mekonsrevenge/\r\nb19e2df06dbcccd9,https://c1.staticflickr.com/8/7511/27829151505_2172064741_o.jpg,0.421177267,0.25,0.842354533,1,673,816,https://www.flickr.com/photos/flickbergers/27829151505/,https://creativecommons.org/licenses/by/2.0/,Tim Berger,https://www.flickr.com/people/flickbergers/\r\n510e5062fef4a993,https://c1.staticflickr.com/6/5822/30267535413_2a7544b60d_o.jpg,0,0,0.599757674,0.75,631,293,https://www.flickr.com/photos/85546319@N04/30267535413/,https://creativecommons.org/licenses/by/2.0/,Robert Sheie,https://www.flickr.com/people/85546319@N04/\r\n943274babd06a662,https://c1.staticflickr.com/1/327/32913036916_44339626f4_o.jpg,0.28125,0,0.84375,0.75,669,678,https://www.flickr.com/photos/davidstanleytravel/32913036916,https://creativecommons.org/licenses/by/2.0/,David Stanley,https://www.flickr.com/people/davidstanleytravel/\r\n0e83f990a04ed67d,https://c1.staticflickr.com/8/7788/28316840171_4f69657675_o.jpg,0,0,0.642547033,1,822,52,https://www.flickr.com/photos/newsoresund/28316840171,https://creativecommons.org/licenses/by/2.0/,News Oresund,https://www.flickr.com/people/newsoresund/\r\n607702d7ef44d081,https://c1.staticflickr.com/9/8233/29529503870_9d1063daa4_o.jpg,0.28125,0,0.84375,0.75,865,347,https://www.flickr.com/photos/zombieite/29529503870,https://creativecommons.org/licenses/by/2.0/,zombieite,https://www.flickr.com/people/zombieite/\r\ne9811aebe8bed6e5,https://c1.staticflickr.com/9/8738/27986089450_329b7a8ebc_o.jpg,0.4375,0,1,0.75,336,410,https://www.flickr.com/photos/peter-trimming/27986089450,https://creativecommons.org/licenses/by/2.0/,Peter Trimming,https://www.flickr.com/people/peter-trimming/\r\n380053c506e32105,https://c1.staticflickr.com/8/7396/27441608895_889e301b2b_o.jpg,0,0.25,0.5625,1,780,514,https://www.flickr.com/photos/thoseguys119/27441608895,https://creativecommons.org/licenses/by/2.0/,ThoseGuys119,https://www.flickr.com/people/thoseguys119/\r\n668bc83a884d24f9,https://c1.staticflickr.com/6/5448/30684271290_40fd020a51_o.jpg,0,0,0.728813559,1,3,391,https://flickr.com/64084569@N02/30684271290,https://creativecommons.org/licenses/by/2.0/,Singa Hitam,https://www.flickr.com/people/64084569@N02/\r\nbe9a3bea2989a0b4,https://c1.staticflickr.com/6/5724/30572182090_20ab46d2fb_o.jpg,0.25,0,0.75,0.75,718,677,https://www.flickr.com/photos/42220226@N07/30572182090,https://creativecommons.org/licenses/by/2.0/,Sicnag,https://www.flickr.com/people/42220226@N07/\r\nb74c05ecff5f99d8,https://c1.staticflickr.com/9/8403/29162412302_d1940244a9_o.jpg,0,0,0.75,0.565438871,407,568,https://www.flickr.com/photos/eltb/29162412302,https://creativecommons.org/licenses/by/2.0/,Catedrales e Iglesias/Cathedrals and Churches,https://www.flickr.com/people/eltb/\r\n7a51c260ca0e83e7,https://c1.staticflickr.com/4/3912/33185518846_5c9ca0ff3a_o.jpg,0.4375,0,1,0.75,548,748,https://www.flickr.com/photos/belurashok/33185518846/,https://creativecommons.org/licenses/by/2.0/,Belur Ashok,https://www.flickr.com/people/belurashok/\r\ne89d48768917f563,https://c1.staticflickr.com/9/8043/29272187281_a1bf8985b6_o.jpg,0.437793427,0,1,0.75,459,308,https://www.flickr.com/photos/rheba/29272187281/,https://creativecommons.org/licenses/by/2.0/,Rheba,https://www.flickr.com/people/rheba/\r\n080bec22a6b1d37d,https://c1.staticflickr.com/8/7787/27498389575_2f7553fca7_o.jpg,0.322236181,0,0.966708543,0.75,760,818,https://www.flickr.com/photos/51764518@N02/27498389575,https://creativecommons.org/licenses/by/2.0/,Joe Haupt,https://www.flickr.com/people/51764518@N02/\r\ne37b556e8889d2ec,https://c1.staticflickr.com/6/5679/30117494085_f273295290_o.jpg,0.301659125,0,1,1,752,618,https://www.flickr.com/photos/karen_roe/30117494085,https://creativecommons.org/licenses/by/2.0/,Karen Roe,https://www.flickr.com/people/karen_roe/\r\n67dd47336dc89421,https://c1.staticflickr.com/8/7566/26809054634_395b8c8989_o.jpg,0,0,0.5,0.75,818,949,https://www.flickr.com/photos/curtsm/26809054634/,https://creativecommons.org/licenses/by/2.0/,Curt Smith,https://www.flickr.com/people/curtsm/\r\n25f85aadcf248cc1,https://c1.staticflickr.com/6/5503/30178488442_471bae40e3_o.jpg,0.25,0,0.75,0.75,854,890,https://www.flickr.com/photos/tanaka_juuyoh/30178488442,https://creativecommons.org/licenses/by/2.0/,TANAKA Juuyoh (\\347\\224\\260\\344\\270\\255\\345\\215\\201\\346\\264\\213),https://www.flickr.com/people/tanaka_juuyoh/\r\nc3e9a84e039df3ce,https://c1.staticflickr.com/9/8538/29956439481_7289ba54cc_o.jpg,0.25,0,1,1,469,765,https://www.flickr.com/photos/27665395@N05/29956439481,https://creativecommons.org/licenses/by/2.0/,Jason Lawrence,https://www.flickr.com/people/27665395@N05/\r\n0134ba03294509a1,https://c1.staticflickr.com/9/8489/29195271540_316f8c5315_o.jpg,0.25,0,0.75,0.75,982,355,https://www.flickr.com/photos/mindahaas/29195271540/,https://creativecommons.org/licenses/by/2.0/,Minda Haas Kuhlmann,https://www.flickr.com/people/mindahaas/\r\n53fe22426ff21243,https://c1.staticflickr.com/1/626/31511641894_19a2e1d2aa_o.jpg,0.496696035,0,0.99339207,0.75,536,779,https://www.flickr.com/photos/glorycycles/31511641894,https://creativecommons.org/licenses/by/2.0/,Glory Cycles,https://www.flickr.com/people/glorycycles/\r\n4053a6a3d0e63c01,https://c1.staticflickr.com/1/561/32656096985_6601815751_o.jpg,0,0,0.5625,0.75,392,460,https://www.flickr.com/photos/kjfnjy/32656096985/,https://creativecommons.org/licenses/by/2.0/,Ken and Nyetta,https://www.flickr.com/people/kjfnjy/\r\n3ece3c1f94e4141a,https://c1.staticflickr.com/1/484/31819662813_17ab59d903_o.jpg,0.422345483,0.25,1,1,332,15,https://www.flickr.com/photos/usfwsmtnprairie/31819662813,https://creativecommons.org/licenses/by/2.0/,USFWS Mountain-Prairie,https://www.flickr.com/people/usfwsmtnprairie/\r\n7d253f78ef6f029a,https://c1.staticflickr.com/6/5488/29967850914_d7e94fa9f2_o.jpg,0,0,1,1,683,92,https://flickr.com/23562593@N07/29967850914,https://creativecommons.org/licenses/by/2.0/,Benjamin Wunn,https://www.flickr.com/people/derzw3rg/\r\nfecd59b8de8ae6f7,https://c1.staticflickr.com/1/475/31566723602_ced86113ef_o.jpg,0.25,0.25,0.75,1,820,793,https://www.flickr.com/photos/stevendepolo/31566723602/,https://creativecommons.org/licenses/by/2.0/,Steven Depolo,https://www.flickr.com/people/stevendepolo/\r\n24159caf9c5e92be,https://c1.staticflickr.com/1/471/30661449994_6c0d5f9050_o.jpg,0.5,0,1,0.75,708,743,https://www.flickr.com/photos/146625745@N08/30661449994,https://creativecommons.org/licenses/by/2.0/,home thods,https://www.flickr.com/people/146625745@N08/\r\n44982b98ba192bb7,https://c1.staticflickr.com/9/8301/28642395174_40ebd0c8df_o.jpg,0.25,0,0.75,0.75,322,411,https://www.flickr.com/photos/wm_archiv/28642395174/,https://creativecommons.org/licenses/by/2.0/,Allie_Caulfield,https://www.flickr.com/people/wm_archiv/\r\n7ca714cdcd5d36d9,https://c1.staticflickr.com/8/7400/27242157102_b9cb18ffa1_o.jpg,0,0,0.5625,0.75,780,106,https://www.flickr.com/photos/thoseguys119/27242157102,https://creativecommons.org/licenses/by/2.0/,ThoseGuys119,https://www.flickr.com/people/thoseguys119/\r\ncbaff3255afbe83f,https://c1.staticflickr.com/9/8118/29896215326_352a9625cf_o.jpg,0,0.25,0.670648464,1,561,853,https://www.flickr.com/photos/dirkhansen/29896215326,https://creativecommons.org/licenses/by/2.0/,SD Dirk,https://www.flickr.com/people/dirkhansen/\r\nfafbad0b3093a886,https://c1.staticflickr.com/9/8699/27967786060_8e927f96b6_o.jpg,0.287776899,0.25,0.863330696,1,672,315,https://www.flickr.com/photos/glorycycles/27967786060,https://creativecommons.org/licenses/by/2.0/,Glory Cycles,https://www.flickr.com/people/glorycycles/\r\n45315ea51da415ab,https://c1.staticflickr.com/9/8855/27598466234_c924d2bdfc_o.jpg,0.25,0.25,0.75,1,356,189,https://www.flickr.com/photos/o_0/27598466234,https://creativecommons.org/licenses/by/2.0/,Guilhem Vellut,https://www.flickr.com/people/o_0/\r\n1e4bb9fd37654da6,https://c1.staticflickr.com/9/8270/29189744532_98cb04acea_o.jpg,0.25,0.25,1,0.75,820,538,https://www.flickr.com/photos/tim_uk/29189744532,https://creativecommons.org/licenses/by/2.0/,Tim Sheerman-Chase,https://www.flickr.com/people/tim_uk/\r\n5f3ea3c6b683fb57,https://c1.staticflickr.com/1/780/32585346961_4f8e8d97e3_o.jpg,0.5,0,1,0.75,773,62,https://flickr.com/28752865@N08/32585346961,https://creativecommons.org/licenses/by/2.0/,Karen Roe,https://www.flickr.com/people/karen_roe/\r\nccbcaf10176e78e0,https://c1.staticflickr.com/9/8397/27954967983_925cdd6886_o.jpg,0,0,0.5,0.75,923,858,https://www.flickr.com/photos/alan-light/27954967983,https://creativecommons.org/licenses/by/2.0/,Alan Light,https://www.flickr.com/people/alan-light/\r\n8d17aa69b4fd7b45,https://c1.staticflickr.com/6/5664/29822954894_e605bf0229_o.jpg,0,0,0.666666667,1,407,590,https://www.flickr.com/photos/quinet/29822954894,https://creativecommons.org/licenses/by/2.0/,Thomas Quine,https://www.flickr.com/people/quinet/\r\nc25da85d5d62b000,https://c1.staticflickr.com/8/7322/27388042450_19773a1bc9_o.jpg,0,0.25,0.5,1,404,219,https://www.flickr.com/photos/usnavy/27388042450,https://creativecommons.org/licenses/by/2.0/,Official U.S. Navy Page,https://www.flickr.com/people/usnavy/\r\nbeb3e55c385feb12,https://c1.staticflickr.com/9/8817/28251101564_5f9eb85574_o.jpg,0.177484472,0.25,0.532453416,1,815,881,https://www.flickr.com/photos/watts_photos/28251101564,https://creativecommons.org/licenses/by/2.0/,C Watts,https://www.flickr.com/people/watts_photos/\r\ned3573d0c197fbaf,https://c1.staticflickr.com/9/8337/28698753904_d31322462d_o.jpg,0.498159057,0.25,0.996318115,1,848,190,https://www.flickr.com/photos/exercitobrasileiro/28698753904,https://creativecommons.org/licenses/by/2.0/,Ex\\303\\251rcito Brasileiro,https://www.flickr.com/people/exercitobrasileiro/\r\n95d1a6be6786190f,https://c1.staticflickr.com/6/5580/31026020991_dbbbedc4f6_o.jpg,0.5,0.25,1,1,891,983,https://www.flickr.com/photos/swisstchoukball/31026020991/,https://creativecommons.org/licenses/by/2.0/,Swiss Tchoukball,https://www.flickr.com/people/swisstchoukball/\r\nc8abefc975e94b10,https://c1.staticflickr.com/8/7451/27275536125_1b48f36807_o.jpg,0,0,0.563380282,0.75,875,675,https://www.flickr.com/photos/southbeachcars/27275536125,https://creativecommons.org/licenses/by/2.0/,Phillip Pessar,https://www.flickr.com/people/southbeachcars/\r\nbe97d1e63c32043e,https://c1.staticflickr.com/9/8268/29636815873_a756a0aaec_o.jpg,0,0.25,0.670648464,1,95,230,https://www.flickr.com/photos/sucriertt/29636815873/,https://creativecommons.org/licenses/by/2.0/,Feroze Omardeen,https://www.flickr.com/people/sucriertt/\r\ndf2cd1b82d5dec14,https://c1.staticflickr.com/6/5499/31112840340_73557ca884_o.jpg,0.5,0.25,1,1,519,414,https://www.flickr.com/photos/box_repsol/31112840340,https://creativecommons.org/licenses/by/2.0/,Box Repsol,https://www.flickr.com/people/box_repsol/\r\n024f73e53c3c1f1d,https://c1.staticflickr.com/8/7784/27062661262_9732ff0e24_o.jpg,0.333333333,0,1,1,580,638,https://www.flickr.com/photos/dalbera/27062661262,https://creativecommons.org/licenses/by/2.0/,Jean-Pierre Dalb\\303\\251ra,https://www.flickr.com/people/dalbera/\r\ne957c417fcd20c5b,https://c1.staticflickr.com/6/5728/31046143176_16800f95c8_o.jpg,0.199868074,0.25,0.599604222,1,820,791,https://www.flickr.com/photos/usembassyta/31046143176/,https://creativecommons.org/licenses/by/2.0/,U.S. Embassy Tel Aviv,https://www.flickr.com/people/usembassyta/\r\n795f5aef9bef8134,https://c1.staticflickr.com/8/7246/27059914000_5331c7a27c_o.jpg,0.231382979,0.25,0.694148936,1,826,29,https://www.flickr.com/photos/sanp/27059914000,https://creativecommons.org/licenses/by/2.0/,San P,https://www.flickr.com/people/sanp/\r\n422d08407aeed1be,https://c1.staticflickr.com/9/8891/28070937640_a79a8cbfa5_o.jpg,0.4375,0,1,0.75,114,44,https://www.flickr.com/photos/yutakaseki/28070937640,https://creativecommons.org/licenses/by/2.0/,Yutaka Seki,https://www.flickr.com/people/yutakaseki/\r\n69a39406cb1d2ce5,https://c1.staticflickr.com/8/7179/27561675775_4977abbb7b_o.jpg,0,0.25,0.75,1,476,46,https://www.flickr.com/photos/44156646@N06/27561675775,https://creativecommons.org/licenses/by/2.0/,Frederick Lang Jr.,https://www.flickr.com/people/44156646@N06/\r\n564de663adeb8cd5,https://c1.staticflickr.com/8/7207/27043384543_1ee00dc2e1_o.jpg,0.494868035,0,0.98973607,0.75,558,133,https://www.flickr.com/photos/governortomwolf/27043384543,https://creativecommons.org/licenses/by/2.0/,Governor Tom Wolf,https://www.flickr.com/people/governortomwolf/\r\n04c1bde5affb7b5b,https://c1.staticflickr.com/1/473/31566239624_128d26f762_o.jpg,0.285496183,0,1,1,653,680,https://www.flickr.com/photos/rdecom/31566239624,https://creativecommons.org/licenses/by/2.0/,U.S. Army RDECOM,https://www.flickr.com/people/rdecom/\r\n90f5385a77a49f77,https://c1.staticflickr.com/8/7393/26525400000_f5d0ae4309_o.jpg,0.248713235,0.25,0.746139706,1,704,905,https://www.flickr.com/photos/panr/26525400000,https://creativecommons.org/licenses/by/2.0/,Robert Cutts,https://www.flickr.com/people/panr/\r\n604b54a3b7424d9b,https://c1.staticflickr.com/9/8358/29207069984_4f7b859b65_o.jpg,0,0.25,0.501846381,1,939,430,https://flickr.com/25797459@N06/29207069984,https://creativecommons.org/licenses/by/2.0/,zaphad1,https://www.flickr.com/people/25797459@N06/\r\na6535c039bfeba2c,https://c1.staticflickr.com/1/737/32082026800_fdf44c3c27_o.jpg,0,0,0.561328125,0.75,536,292,https://www.flickr.com/photos/glorycycles/32082026800,https://creativecommons.org/licenses/by/2.0/,Glory Cycles,https://www.flickr.com/people/glorycycles/\r\n35d67f3f70aeef92,https://c1.staticflickr.com/8/7324/28047778902_e23628dc50_o.jpg,0.25,0.25,0.75,1,980,582,https://www.flickr.com/photos/coreyheitzmd/28047778902,https://creativecommons.org/licenses/by/2.0/,CHeitz,https://www.flickr.com/people/coreyheitzmd/\r\n3efa3df701898afd,https://c1.staticflickr.com/9/8783/28492814412_129982444e_o.jpg,0,0,0.905498282,1,737,1000,https://www.flickr.com/photos/stevendepolo/28492814412,https://creativecommons.org/licenses/by/2.0/,Steven Depolo,https://www.flickr.com/people/stevendepolo/\r\nf08f34c258228251,https://c1.staticflickr.com/9/8256/28174054160_790b304fc4_o.jpg,0.28125,0.25,0.84375,1,699,608,https://www.flickr.com/photos/rubenholthuijsen/28174054160/,https://creativecommons.org/licenses/by/2.0/,Ruben Holthuijsen,https://www.flickr.com/people/rubenholthuijsen/\r\nd47c9aabecda0e12,https://c1.staticflickr.com/8/7441/27270121096_cc1007eb2f_o.jpg,0.5,0,1,0.75,986,199,https://www.flickr.com/photos/56218409@N03/27270121096,https://creativecommons.org/licenses/by/2.0/,Matthias Ripp,https://www.flickr.com/people/56218409@N03/\r\nd97c1f3f1c211d1a,https://c1.staticflickr.com/9/8448/28945025323_6e7cdf207a_o.jpg,0,0,0.75,0.75,979,467,https://www.flickr.com/photos/edans/28945025323,https://creativecommons.org/licenses/by/2.0/,Enrique Dans,https://www.flickr.com/people/edans/\r\n62949c357e429690,https://c1.staticflickr.com/1/764/31219968184_cf7294c955_o.jpg,0.25,0.25,0.75,1,973,116,https://www.flickr.com/photos/jlascar/31219968184,https://creativecommons.org/licenses/by/2.0/,Jorge L\\303\\241scar,https://www.flickr.com/people/jlascar/\r\n9311a098cfb4b946,https://c1.staticflickr.com/8/7373/26786081873_73a09aab2f_o.jpg,0.25,0.25,0.75,1,821,192,https://www.flickr.com/photos/ben_salter/26786081873/,https://creativecommons.org/licenses/by/2.0/,Ben Salter,https://www.flickr.com/people/ben_salter/\r\n38d8d105d5e70c80,https://c1.staticflickr.com/8/7457/27675995271_5b6dc8a928_o.jpg,0,0,1,1,422,755,https://www.flickr.com/photos/arbron/27675995271,https://creativecommons.org/licenses/by/2.0/,Jeff Hitchcock,https://www.flickr.com/people/arbron/\r\na28beac14b9037da,https://c1.staticflickr.com/8/7408/26281652233_6667df70cb_o.jpg,0.333333333,0,1,1,405,351,https://www.flickr.com/photos/km30192002/26281652233/,https://creativecommons.org/licenses/by/2.0/,km30192002,https://www.flickr.com/people/km30192002/\r\nfa3aae44c2675e88,https://c1.staticflickr.com/9/8776/28284921904_800c76d8ac_o.jpg,0.496696035,0,0.99339207,0.75,771,375,https://www.flickr.com/photos/spera-designerschuhe/28284921904,https://creativecommons.org/licenses/by/2.0/,\"SPERA.de Designerschuhe, Taschen und Accessoires\",https://www.flickr.com/people/spera-designerschuhe/\r\n44aa69f642b86628,https://c1.staticflickr.com/8/7413/27678489230_1d967f5b58_o.jpg,0,0,0.634532374,1,880,549,https://www.flickr.com/photos/45909111@N00/27678489230,https://creativecommons.org/licenses/by/2.0/,Gwydion M. Williams,https://www.flickr.com/people/45909111@N00/\r\n6cc289dc7a8fce86,https://c1.staticflickr.com/9/8671/28656397400_d4db68df02_o.jpg,0.501840943,0.25,1,1,158,1000,https://www.flickr.com/photos/naql/28656397400,https://creativecommons.org/licenses/by/2.0/,naql,https://www.flickr.com/people/naql/\r\na92babc0a6436d3b,https://c1.staticflickr.com/8/7582/28982049520_eddc2d650d_o.jpg,0,0.25,0.5,1,760,655,https://www.flickr.com/photos/ulift/28982049520/in/album-72157672059230641/,https://creativecommons.org/licenses/by/2.0/,WEi WEi,https://www.flickr.com/people/ulift/\r\n9fbeb797de1e1639,https://c1.staticflickr.com/9/8083/29921495296_991d4559c0_o.jpg,0.421747967,0,0.843495935,0.75,763,511,https://www.flickr.com/photos/vuhung/29921495296,https://creativecommons.org/licenses/by/2.0/,Nguyen Hung Vu,https://www.flickr.com/people/vuhung/\r\n16179baa1513c701,https://c1.staticflickr.com/1/722/32761564425_d122c3838d_o.jpg,0,0,0.75,1,95,505,https://flickr.com/95098864@N08/32761564425,https://creativecommons.org/licenses/by/2.0/,S. Rohrlach,https://www.flickr.com/people/95098864@N08/\r\n5a8d5bc9bf40e832,https://c1.staticflickr.com/6/5692/30199498082_eba90ef2e1_o.jpg,0,0,0.666666667,1,519,634,https://www.flickr.com/photos/box_repsol/30199498082,https://creativecommons.org/licenses/by/2.0/,Box Repsol,https://www.flickr.com/people/box_repsol/\r\naa4e1cf9cb6e7801,https://c1.staticflickr.com/9/8183/29036345831_4da7d925c9_o.jpg,0.151376147,0.25,0.45412844,1,896,840,https://www.flickr.com/photos/ronmacphotos/29036345831/,https://creativecommons.org/licenses/by/2.0/,Ronnie Macdonald,https://www.flickr.com/people/ronmacphotos/\r\n200735f93ea764f2,https://c1.staticflickr.com/8/7582/29431396080_570340a5ef_o.jpg,0.310344828,0,0.931034483,0.75,309,285,https://www.flickr.com/photos/gails_pictures/29431396080,https://creativecommons.org/licenses/by/2.0/,gailhampshire,https://www.flickr.com/people/gails_pictures/\r\n5d53f9bc389e98ad,https://c1.staticflickr.com/1/722/30784991794_6af0ab8176_o.jpg,0.25,0.25,0.75,1,265,468,https://www.flickr.com/photos/unirodlibrary/30784991794/,https://creativecommons.org/licenses/by/2.0/,Rod Library,https://www.flickr.com/people/unirodlibrary/\r\n5b974695c09ad165,https://c1.staticflickr.com/9/8453/29350958465_8426fed375_o.jpg,0.25,0.484655689,1,1,991,393,https://www.flickr.com/photos/clairity/29350958465/,https://creativecommons.org/licenses/by/2.0/,Sharon Mollerus,https://www.flickr.com/people/clairity/\r\nbecf3b004177ded6,https://c1.staticflickr.com/8/7565/28346247573_c430ef5720_o.jpg,0.193181818,0,1,1,986,457,https://www.flickr.com/photos/34905030@N00/28346247573/,https://creativecommons.org/licenses/by/2.0/,Orangeaurochs,https://www.flickr.com/people/orangeaurochs/\r\n4f04ca75d7aafc0a,https://c1.staticflickr.com/8/7126/27594041042_ef94e651db_o.jpg,0.25,0.249631811,1,0.748895434,840,633,https://www.flickr.com/photos/girlguidesofcan/27594041042/,https://creativecommons.org/licenses/by/2.0/,Girl Guides of Canada,https://www.flickr.com/people/girlguidesofcan/\r\n8764eecf6e898c7f,https://c1.staticflickr.com/8/7393/27564047580_b1667768cf_o.jpg,0.33773862,0,1,1,484,711,https://www.flickr.com/photos/sybarite48/27564047580,https://creativecommons.org/licenses/by/2.0/,Daniel Jolivet,https://www.flickr.com/people/sybarite48/\r\nc5f5e09f4f5fd41a,https://c1.staticflickr.com/9/8134/28977820041_e39c5c15c4_o.jpg,0.559602649,0.25,0.839403974,1,404,421,https://flickr.com/7489441@N06/28977820041,https://creativecommons.org/licenses/by/2.0/,Clemens Vasters,https://www.flickr.com/people/clemensv/\r\n17f3b8a057bc01be,https://c1.staticflickr.com/8/7537/28104131410_e13ab8f9ef_o.jpg,0,0,1,1,859,575,https://www.flickr.com/photos/v1ctor/28104131410,https://creativecommons.org/licenses/by/2.0/,Victor,https://www.flickr.com/people/v1ctor/\r\nc10f6e0689efbc1c,https://c1.staticflickr.com/1/258/32241427822_ac7330299e_o.jpg,0.27818818,0.25,0.834564541,1,888,297,https://www.flickr.com/photos/guldem/32241427822,https://creativecommons.org/licenses/by/2.0/,G\\303\\274ldem \\303\\234st\\303\\274n,https://www.flickr.com/people/guldem/\r\n0e9bdc3ab80b1a71,https://c1.staticflickr.com/8/7415/27313160706_a2440a83b0_o.jpg,0.5,0,1,0.75,367,335,https://www.flickr.com/photos/ross_elliott/27313160706,https://creativecommons.org/licenses/by/2.0/,Ross Elliott,https://www.flickr.com/people/ross_elliott/\r\nc17d39b1eba4cc03,https://c1.staticflickr.com/8/7468/27818709730_d10459c415_o.jpg,0,0.5,0.75,1,669,732,https://www.flickr.com/photos/jdlasica/27818709730,https://creativecommons.org/licenses/by/2.0/,JD Lasica,https://www.flickr.com/people/jdlasica/\r\nec346009f286ffca,https://c1.staticflickr.com/1/571/30740221514_3c4fa602a0_o.jpg,0,0,0.5,0.75,120,783,https://www.flickr.com/photos/conchur/30740221514/,https://creativecommons.org/licenses/by/2.0/,Conor Lawless,https://www.flickr.com/people/conchur/\r\n01fa4be27dd11b5d,https://c1.staticflickr.com/1/268/31600495093_a46c0835d2_o.jpg,0.4375,0,1,0.75,451,829,https://www.flickr.com/photos/ufv/31600495093/,https://creativecommons.org/licenses/by/2.0/,University of the Fraser Valley,https://www.flickr.com/people/ufv/\r\n490df9e021386d29,https://c1.staticflickr.com/6/5800/30150475824_b342db70fa_o.jpg,0.150773196,0,0.452319588,0.75,405,351,https://flickr.com/37804979@N00/30150475824,https://creativecommons.org/licenses/by/2.0/,Tony Hisgett,https://www.flickr.com/people/hisgett/\r\n08fbf75e60a6efb7,https://c1.staticflickr.com/9/8270/29799045164_ab072918cb_o.jpg,0,0,0.66226138,1,484,666,https://www.flickr.com/photos/sybarite48/29799045164,https://creativecommons.org/licenses/by/2.0/,Daniel Jolivet,https://www.flickr.com/people/sybarite48/\r\nfc583c08026e3d74,https://c1.staticflickr.com/6/5595/30252932446_a7c00f3a97_o.jpg,0.333333333,0,1,1,881,588,https://www.flickr.com/photos/mobikefed/30252932446,https://creativecommons.org/licenses/by/2.0/,MoBikeFed,https://www.flickr.com/people/mobikefed/\r\ndb6c6671a6924eba,https://c1.staticflickr.com/9/8269/30416458805_dc9a3ef437_o.jpg,0.25,0,1,0.591346154,407,412,https://www.flickr.com/photos/eltb/30416458805,https://creativecommons.org/licenses/by/2.0/,Catedrales e Iglesias/Cathedrals and Churches,https://www.flickr.com/people/eltb/\r\n5a6439b10c1f6761,https://c1.staticflickr.com/6/5703/29661528724_1ccb11fae8_o.jpg,0,0.25,0.561328125,1,320,871,https://flickr.com/86548370@N00/29661528724,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\n218e02b6cfa0466e,https://c1.staticflickr.com/1/702/32609631415_d7bf443a2c_o.jpg,0.5,0.25,1,1,699,217,https://www.flickr.com/photos/renaissancechambara/32609631415/,https://creativecommons.org/licenses/by/2.0/,Ged Carroll,https://www.flickr.com/people/renaissancechambara/\r\n3f8bf287f20333eb,https://c1.staticflickr.com/9/8895/29333257946_c53911b432_o.jpg,0,0,0.5,0.75,973,713,https://www.flickr.com/photos/aschaf/29333257946/,https://creativecommons.org/licenses/by/2.0/,Andrea Schaffer,https://www.flickr.com/people/aschaf/\r\nbe5543dc68f0b970,https://c1.staticflickr.com/6/5835/30148396554_e2ec6fbf0d_o.jpg,0.441600679,0,0.736001131,0.75,780,636,https://www.flickr.com/photos/thoseguys119/30148396554,https://creativecommons.org/licenses/by/2.0/,ThoseGuys119,https://www.flickr.com/people/thoseguys119/\r\n87fcc5f70673d3d2,https://c1.staticflickr.com/8/7455/27040942130_44a032d0fd_o.jpg,0.421177267,0,0.842354533,0.75,699,159,https://www.flickr.com/photos/nat507/27040942130,https://creativecommons.org/licenses/by/2.0/,Nathan Hughes Hamilton,https://www.flickr.com/people/nat507/\r\n99f175132cd1c6e2,https://c1.staticflickr.com/8/7045/27716560671_79806e498e_o.jpg,0.25,0,1,0.75,934,459,https://www.flickr.com/photos/danielchownet/27716560671,https://creativecommons.org/licenses/by/2.0/,Daniel Chow,https://www.flickr.com/people/danielchownet/\r\na4f718e554dd18dd,https://c1.staticflickr.com/6/5445/29774026800_5e37689427_o.jpg,0.5,0.25,1,1,976,95,https://www.flickr.com/photos/wm_archiv/29774026800,https://creativecommons.org/licenses/by/2.0/,Allie_Caulfield,https://www.flickr.com/people/wm_archiv/\r\n46a489dcff5471a4,https://c1.staticflickr.com/3/2634/32267074504_a395babafd_o.jpg,0,0,0.75,0.75,920,590,https://www.flickr.com/photos/83015819@N00/32267074504/,https://creativecommons.org/licenses/by/2.0/,Jon Worth,https://www.flickr.com/people/83015819@N00/\r\n38c6124f40b8d845,https://c1.staticflickr.com/8/7589/27543292150_7375ef6b66_o.jpg,0.249631811,0,0.748895434,0.75,896,19,https://www.flickr.com/photos/ministreiodadefesa/27543292150/,https://creativecommons.org/licenses/by/2.0/,Minist\\303\\251rio da Defesa,https://www.flickr.com/people/ministreiodadefesa/\r\n6740e1e89b5bbab4,https://c1.staticflickr.com/1/439/32456945576_0d93348b0e_o.jpg,0.25,0.25,0.75,1,981,685,https://www.flickr.com/photos/30478819@N08/32456945576,https://creativecommons.org/licenses/by/2.0/,Marco Verch,https://www.flickr.com/people/30478819@N08/\r\n54ff02d9dfbf8540,https://c1.staticflickr.com/6/5703/30663546512_038de017f8_o.jpg,0.249079529,0,0.747238586,0.75,884,370,https://www.flickr.com/photos/reliv/30663546512/,https://creativecommons.org/licenses/by/2.0/,Reliv International,https://www.flickr.com/people/reliv/\r\n2c9398f44d099e0f,https://c1.staticflickr.com/3/2051/32840993165_29d0cb8a1e_o.jpg,0,0,1,0.75,522,482,http://www.flickr.com/photos/23119666@N03/32840993165/,https://creativecommons.org/licenses/by/2.0/,Mark Bonica,https://www.flickr.com/people/23119666@N03/\r\n6896a8b78f404dc8,https://c1.staticflickr.com/9/8358/29790317621_d6a35e5136_o.jpg,0.4375,0.25,1,1,980,75,https://www.flickr.com/photos/kameyama-mie/29790317621,https://creativecommons.org/licenses/by/2.0/,Kasadera,https://www.flickr.com/people/kameyama-mie/\r\nd2d6ae6e0b8a0c31,https://c1.staticflickr.com/6/5600/30243986500_e184ef72fb_o.jpg,0,0,0.5,0.75,986,601,https://www.flickr.com/photos/blumenbiene/30243986500,https://creativecommons.org/licenses/by/2.0/,Maja Dumat,https://www.flickr.com/people/blumenbiene/\r\nfccfc3563cd562e1,https://c1.staticflickr.com/9/8645/27729536774_b56a633841_o.jpg,0.25,0.25,0.75,1,977,656,https://flickr.com/111347625@N06/27729536774,https://creativecommons.org/licenses/by/2.0/,Heribert Bechen ... bin dann mal weg!,https://www.flickr.com/people/hb1248/\r\nd3a88c50958b1b5f,https://c1.staticflickr.com/9/8746/27853631254_508e610bc1_o.jpg,0,0.25,0.5,1,977,1000,https://www.flickr.com/photos/celteverett/27853631254,https://creativecommons.org/licenses/by/2.0/,Gregorio Puga Bail\\303\\263n,https://www.flickr.com/people/celteverett/\r\n513b3032ed1f388c,https://c1.staticflickr.com/8/7514/27572579621_930aecf52c_o.jpg,0.28125,0.25,0.84375,1,114,542,https://www.flickr.com/photos/viewpix69/27572579621/,https://creativecommons.org/licenses/by/2.0/,Torsten Scholz,https://www.flickr.com/people/viewpix69/\r\n6217866896fbe2b7,https://c1.staticflickr.com/6/5719/30923267621_c221b094fe_o.jpg,0,0,0.5,0.75,699,867,https://www.flickr.com/photos/mondo79/30923267621/,https://creativecommons.org/licenses/by/2.0/,Mondo79,https://www.flickr.com/people/mondo79/\r\n21554e755b9706d2,https://c1.staticflickr.com/1/641/32391235981_071803d033_o.jpg,0,0,0.75,0.529210926,342,76,https://www.flickr.com/photos/ilri/32391235981,https://creativecommons.org/licenses/by/2.0/,ILRI,https://www.flickr.com/people/ilri/\r\n87a2147620d5e1cb,https://c1.staticflickr.com/6/5603/30536597041_14c0e333ab_o.jpg,0.4375,0,1,0.75,780,290,https://www.flickr.com/photos/thoseguys119/30536597041/,https://creativecommons.org/licenses/by/2.0/,ThoseGuys119,https://www.flickr.com/people/thoseguys119/\r\nb86d604067f50613,https://c1.staticflickr.com/6/5791/30885593836_dda0ff2d59_o.jpg,0,0.248826291,1,1,490,998,https://www.flickr.com/photos/southbeachcars/30885593836/,https://creativecommons.org/licenses/by/2.0/,Phillip Pessar,https://www.flickr.com/people/southbeachcars/\r\naf3736cca01a2d28,https://c1.staticflickr.com/1/636/32603914891_d687da69f1_o.jpg,0.25,0,1,0.75,318,766,https://www.flickr.com/photos/treegrow/32603914891/,https://creativecommons.org/licenses/by/2.0/,Katja Schulz,https://www.flickr.com/people/treegrow/\r\n5adb3d985cf217a9,https://c1.staticflickr.com/3/2183/32995484255_f1c2125a38_o.jpg,0,0.25,0.421177267,1,405,875,https://www.flickr.com/photos/n-bphotography/32995484255/,https://creativecommons.org/licenses/by/2.0/,Nicky Boogaard,https://www.flickr.com/people/n-bphotography/\r\n60566d5680497389,https://c1.staticflickr.com/8/7117/27308045115_c9860d768d_o.jpg,0.5,0,1,0.75,473,544,https://www.flickr.com/photos/vastateparksstaff/27308045115,https://creativecommons.org/licenses/by/2.0/,Virginia State Parks,https://www.flickr.com/people/vastateparksstaff/\r\n1cfa5381359bff9f,https://c1.staticflickr.com/6/5669/30198502892_8d507f5277_o.jpg,0.33773862,0,1,1,422,852,https://www.flickr.com/photos/parktatar/30198502892,https://creativecommons.org/licenses/by/2.0/,\\320\\237\\320\\260\\321\\200\\320\\272\\320\\270 \\320\\242\\320\\260\\321\\202\\320\\260\\321\\200\\321\\201\\321\\202\\320\\260\\320\\275\\320\\260,https://www.flickr.com/people/parktatar/\r\ncd40b0e46c9da574,https://c1.staticflickr.com/8/7705/27405552721_fa059a91db_o.jpg,0,0.25,0.75,1,662,739,https://www.flickr.com/photos/29233640@N07/27405552721,https://creativecommons.org/licenses/by/2.0/,Robert Couse-Baker,https://www.flickr.com/people/29233640@N07/\r\ne9ecdf28dc1d5a85,https://c1.staticflickr.com/8/7293/27771548652_1bbed73d9a_o.jpg,0.25,0.25,1,0.75,609,192,https://www.flickr.com/photos/66755335@N05/27771548652,https://creativecommons.org/licenses/by/2.0/,Geneva Vanderzeil apairandasparediy.com,https://www.flickr.com/people/66755335@N05/\r\n821aae58dc96b4b2,https://c1.staticflickr.com/8/7668/26973339655_c7771d9f37_o.jpg,0.447698745,0.25,0.89539749,1,341,575,http://www.flickr.com/photos/125382597@N08/26973339655/,https://creativecommons.org/licenses/by/2.0/,Roger Hsu,https://www.flickr.com/people/125382597@N08/\r\n715335d5ddfca1f2,https://c1.staticflickr.com/6/5443/30684352681_e443895cc6_o.jpg,0,0.25,0.497426471,1,481,915,https://www.flickr.com/photos/agenciasenado/30684352681,https://creativecommons.org/licenses/by/2.0/,Senado Federal,https://www.flickr.com/people/agenciasenado/\r\n1be47d60e0156bbb,https://c1.staticflickr.com/6/5828/29885217483_5d5c3b4561_o.jpg,0.25,0,1,1,650,912,https://flickr.com/8558643@N06/29885217483,https://creativecommons.org/licenses/by/2.0/,Christophe Delaere,https://www.flickr.com/people/delaere/\r\n8724748782cb6c31,https://c1.staticflickr.com/1/505/31480812833_d9a37e0bee_o.jpg,0,0,0.944554455,1,76,647,https://flickr.com/55893585@N08/31480812833,https://creativecommons.org/licenses/by/2.0/,BLM Nevada,https://www.flickr.com/people/blmnevada/\r\n2ba52bcf59097dc8,https://c1.staticflickr.com/9/8703/29072678195_c8c916e504_o.jpg,0,0,0.563380282,0.75,971,850,https://www.flickr.com/photos/blackwing_de/29072678195,https://creativecommons.org/licenses/by/2.0/,Sebastian Werner,https://www.flickr.com/people/blackwing_de/\r\n2ed718f6db9785f8,https://c1.staticflickr.com/9/8109/29184695352_02765c0138_o.jpg,0.25,0,0.75,0.75,519,843,https://www.flickr.com/photos/box_repsol/29184695352,https://creativecommons.org/licenses/by/2.0/,Box Repsol,https://www.flickr.com/people/box_repsol/\r\n4a251fb6d80b5137,https://c1.staticflickr.com/8/7735/27313853085_2df82462d3_o.jpg,0.439547582,0,1,0.75,800,82,https://flickr.com/21461098@N00/27313853085,https://creativecommons.org/licenses/by/2.0/,Owen Allen,https://www.flickr.com/people/owen59/\r\n3617bfc5d06bc196,https://c1.staticflickr.com/6/5608/29603872360_115b742b20_o.jpg,0.210588633,0.25,0.6317659,1,885,554,https://www.flickr.com/photos/ankurp/29603872360,https://creativecommons.org/licenses/by/2.0/,Ankur P,https://www.flickr.com/people/ankurp/\r\n143c74a57326fd40,https://c1.staticflickr.com/1/362/31438358143_d5b15f28de_o.jpg,0.333333333,0,1,1,422,423,https://www.flickr.com/photos/robin-schmitt/31438358143/,https://creativecommons.org/licenses/by/2.0/,Siris,https://www.flickr.com/people/robin-schmitt/\r\nd051e0bd46125123,https://c1.staticflickr.com/6/5748/31131978085_c9153efb9d_o.jpg,0.4375,0.25,1,1,669,776,https://www.flickr.com/photos/96223380@N02/31131978085,https://creativecommons.org/licenses/by/2.0/,Mighty Travels,https://www.flickr.com/people/96223380@N02/\r\n84af93c72781b7f0,https://c1.staticflickr.com/2/1616/26473779220_95c074ef92_o.jpg,0,0.443623639,0.75,1,438,782,https://www.flickr.com/photos/kansallisarkisto/26473779220,https://creativecommons.org/licenses/by/2.0/,Kansallisarkisto,https://www.flickr.com/people/kansallisarkisto/\r\na52570066ecc61fd,https://c1.staticflickr.com/1/660/31595952701_5eab9952f3_o.jpg,0,0,0.75,1,976,583,https://www.flickr.com/photos/elzey/31595952701,https://creativecommons.org/licenses/by/2.0/,Richard Elzey,https://www.flickr.com/people/elzey/\r\n3e8bf1eee7083f6e,https://c1.staticflickr.com/6/5325/30267534593_e497c0a76a_o.jpg,0,0,0.599757674,0.75,631,683,https://www.flickr.com/photos/85546319@N04/30267534593,https://creativecommons.org/licenses/by/2.0/,Robert Sheie,https://www.flickr.com/people/85546319@N04/\r\nf6d5c95ce7908a58,https://c1.staticflickr.com/9/8103/29317853454_b2dec062c2_o.jpg,0.493768328,0,0.987536657,0.75,561,872,https://www.flickr.com/photos/codnewsroom/29317853454,https://creativecommons.org/licenses/by/2.0/,COD Newsroom,https://www.flickr.com/people/codnewsroom/\r\nbfb2aaf48d7fc08c,https://c1.staticflickr.com/6/5347/30848584970_9d7dd68247_o.jpg,0,0.25,0.5,1,979,953,https://www.flickr.com/photos/deaconmacmillan/30848584970/,https://creativecommons.org/licenses/by/2.0/,Deacon MacMillan,https://www.flickr.com/people/deaconmacmillan/\r\n5528f7330399485f,https://c1.staticflickr.com/6/5821/30665663016_f164e1bbae_o.jpg,0.248826291,0,1,1,74,212,http://www.flickr.com/photos/cbroders/30665663016/,https://creativecommons.org/licenses/by/2.0/,Counse,https://www.flickr.com/people/cbroders/\r\n635ea505879178ed,https://c1.staticflickr.com/6/5787/29740591494_abecb27341_o.jpg,0,0.35413643,1,1,539,923,https://www.flickr.com/photos/22711505@N05/29740591494,https://creativecommons.org/licenses/by/2.0/,Ron Cogswell,https://www.flickr.com/people/22711505@N05/\r\na8631b077e74f7aa,https://c1.staticflickr.com/9/8521/28560392972_e2f7a70fbd_o.jpg,0,0.4375,0.75,1,480,486,https://www.flickr.com/photos/jonnymooshoo/28560392972/,https://creativecommons.org/licenses/by/2.0/,Jonathan Leung,https://www.flickr.com/people/jonnymooshoo/\r\nbb03be3702584eeb,https://c1.staticflickr.com/9/8825/28189849211_8747a0ee87_o.jpg,0,0,0.85475793,1,672,624,https://www.flickr.com/photos/glorycycles/28189849211,https://creativecommons.org/licenses/by/2.0/,Glory Cycles,https://www.flickr.com/people/glorycycles/\r\n8090736c8299ab28,https://c1.staticflickr.com/8/7301/27600935412_70c27a16d2_o.jpg,0.25,0.4375,1,1,980,894,https://www.flickr.com/photos/kitkaphotogirl/27600935412/,https://creativecommons.org/licenses/by/2.0/,Christine Majul,https://www.flickr.com/people/kitkaphotogirl/\r\n74b13710bcf4f7f3,https://c1.staticflickr.com/8/7271/27059072776_6f5d6d0919_o.jpg,0.503303965,0,1,0.75,775,918,https://www.flickr.com/photos/spera-designerschuhe/27059072776,https://creativecommons.org/licenses/by/2.0/,\"SPERA.de Designerschuhe, Taschen und Accessoires\",https://www.flickr.com/people/spera-designerschuhe/\r\necd3a969dd466349,https://c1.staticflickr.com/8/7423/28019501745_3975453d32_o.jpg,0.25,0.25,1,1,285,693,https://www.flickr.com/photos/editor/28019501745,https://creativecommons.org/licenses/by/2.0/,Bart Everson,https://www.flickr.com/people/editor/\r\n2abcab3d77d4767e,https://c1.staticflickr.com/6/5673/30925169246_bcb5ac8949_o.jpg,0,0.248826291,1,1,443,520,https://www.flickr.com/photos/eltb/30925169246,https://creativecommons.org/licenses/by/2.0/,Catedrales e Iglesias/Cathedrals and Churches,https://www.flickr.com/people/eltb/\r\n8c975de17f3247d1,https://c1.staticflickr.com/1/712/32281377125_d669c5337a_o.jpg,0.28125,0,0.84375,0.75,650,526,https://www.flickr.com/photos/stonehenge-stone-circle/32281377125/,https://creativecommons.org/licenses/by/2.0/,Stonehenge Stone Circle,https://www.flickr.com/people/stonehenge-stone-circle/\r\n8d86fae18afef4e0,https://c1.staticflickr.com/6/5834/29905668306_e17a9b35ed_o.jpg,0.238170347,0,1,1,38,79,https://www.flickr.com/photos/2ndpeter/29905668306,https://creativecommons.org/licenses/by/2.0/,Peter Paplanus,https://www.flickr.com/people/2ndpeter/\r\n619a9d0eb2acdd82,https://c1.staticflickr.com/9/8079/28742831150_8cd69d4202_o.jpg,0,0,0.666666667,1,885,843,https://www.flickr.com/photos/zigazou76/28742831150/,https://creativecommons.org/licenses/by/2.0/,Fr\\303\\251d\\303\\251ric BISSON,https://www.flickr.com/people/zigazou76/\r\n9ab953cce5c7f879,https://c1.staticflickr.com/3/2872/33227757875_2a76bc808e_o.jpg,0.333333333,0,1,1,443,367,https://www.flickr.com/photos/dany13/33227757875/,https://creativecommons.org/licenses/by/2.0/,dany13,https://www.flickr.com/people/dany13/\r\n1c2e9fe8b0b2fdf2,https://c1.staticflickr.com/8/7633/27140101841_c1f8aa3b4d_o.jpg,0,0,0.666666667,1,58,905,https://www.flickr.com/photos/gregthebusker/27140101841,https://creativecommons.org/licenses/by/2.0/,Greg Schechter,https://www.flickr.com/people/gregthebusker/\r\na414ccf5c51454e0,https://c1.staticflickr.com/8/7609/27446813796_e349a63c2f_o.jpg,0,0,0.665684831,1,830,866,https://www.flickr.com/photos/fotosagenciabrasil/27446813796,https://creativecommons.org/licenses/by/2.0/,Ag\\303\\252ncia Brasil Fotografias,https://www.flickr.com/people/fotosagenciabrasil/\r\n26aa2e6ee43387e7,https://c1.staticflickr.com/1/378/31616355823_06b59a862a_o.jpg,0.25,0,1,0.75,913,741,https://www.flickr.com/photos/12463666@N03/31616355823/,https://creativecommons.org/licenses/by/2.0/,Bob M ~,https://www.flickr.com/people/12463666@N03/\r\n3302a8c24caf20f4,https://c1.staticflickr.com/8/7553/27852122740_a93432645b_o.jpg,0,0,0.666666667,1,704,548,https://flickr.com/53301297@N00/27852122740,https://creativecommons.org/licenses/by/2.0/,Paul Sableman,https://www.flickr.com/people/pasa/\r\ndd722682a9f2b9dd,https://c1.staticflickr.com/6/5773/29781416180_6cd73114d5_o.jpg,0,0,0.5,0.75,437,458,https://www.flickr.com/photos/greggjerdingen/29781416180,https://creativecommons.org/licenses/by/2.0/,Greg Gjerdingen,https://www.flickr.com/people/greggjerdingen/\r\n6afe58efa85756dc,https://c1.staticflickr.com/8/7406/27477875915_19bb21bd54_o.jpg,0,0.333333333,1,1,929,849,https://www.flickr.com/photos/insatiablemunchies/27477875915,https://creativecommons.org/licenses/by/2.0/,insatiablemunch,https://www.flickr.com/people/insatiablemunchies/\r\n0ceee9a38409e30c,https://c1.staticflickr.com/1/412/31684872936_8e1df2a992_o.jpg,0.4375,0,1,0.75,484,712,https://www.flickr.com/photos/16801915@N06/31684872936/,https://creativecommons.org/licenses/by/2.0/,Reading Tom,https://www.flickr.com/people/16801915@N06/\r\n1df5e1adad21d618,https://c1.staticflickr.com/1/532/32376482310_86efb9eec4_o.jpg,0.25,0,0.75,0.75,405,556,https://www.flickr.com/photos/thejointstaff/32376482310,https://creativecommons.org/licenses/by/2.0/,Chairman of the Joint Chiefs of Staff,https://www.flickr.com/people/thejointstaff/\r\n57e62606f02d0251,https://c1.staticflickr.com/8/7307/27798850402_87d7558850_o.jpg,0.496696035,0,0.99339207,0.75,548,646,https://www.flickr.com/photos/shankaronline/27798850402,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\nd0fe8afb55c3c7cd,https://c1.staticflickr.com/9/8107/28396972686_381dae0505_o.jpg,0,0,1,0.75,498,778,https://www.flickr.com/photos/duncanh1/28396972686,https://creativecommons.org/licenses/by/2.0/,Dun.can,https://www.flickr.com/people/duncanh1/\r\n91d18b353d5b3be2,https://c1.staticflickr.com/4/3935/33250842731_598bbc7d1e_o.jpg,0,0,0.75,0.75,931,26,https://www.flickr.com/photos/dylan20/33250842731,https://creativecommons.org/licenses/by/2.0/,Dylan Tweney,https://www.flickr.com/people/dylan20/\r\n21b0803c965f76ef,https://c1.staticflickr.com/6/5667/31229598566_3fca43f16f_o.jpg,0,0.25,0.403311258,1,920,977,https://www.flickr.com/photos/crackdog/31229598566,https://creativecommons.org/licenses/by/2.0/,William Creswell,https://www.flickr.com/people/crackdog/\r\nf20783fc38473514,https://c1.staticflickr.com/9/8550/28448778474_c095f2264b_o.jpg,0.248713235,0.25,0.746139706,1,386,774,https://www.flickr.com/photos/shankaronline/28448778474,https://creativecommons.org/licenses/by/2.0/,shankar s.,https://www.flickr.com/people/shankaronline/\r\n3d6e530955c60759,https://c1.staticflickr.com/8/7293/26333233883_fb2b616f95_o.jpg,0,0,0.577654517,0.75,512,892,https://www.flickr.com/photos/hugo90/26333233883,https://creativecommons.org/licenses/by/2.0/,JOHN LLOYD,https://www.flickr.com/people/hugo90/\r\n7e5c847bafc4d50c,https://c1.staticflickr.com/6/5557/30353091180_fe0a06de72_o.jpg,0.551833123,0.25,0.919721871,1,909,529,https://www.flickr.com/photos/flyingsinger/30353091180,https://creativecommons.org/licenses/by/2.0/,Bruce Irving,https://www.flickr.com/people/flyingsinger/\r\n0aebe24fc257286e,https://c1.staticflickr.com/8/7343/27235728323_4d87753e80_o.jpg,0,0.25,0.535877863,1,561,315,https://www.flickr.com/photos/143049688@N08/27235728323,https://creativecommons.org/licenses/by/2.0/,MR MAO PICS,https://www.flickr.com/people/143049688@N08/\r\n3c970c84b598f2d9,https://c1.staticflickr.com/6/5693/25380221549_fd25f3d570_o.jpg,0.223849372,0,0.671548117,0.75,780,597,https://www.flickr.com/photos/thoseguys119/25380221549,https://creativecommons.org/licenses/by/2.0/,ThoseGuys119,https://www.flickr.com/people/thoseguys119/\r\nc6166621198a3de4,https://c1.staticflickr.com/6/5563/31119538711_0c79469ee9_o.jpg,0,0.161983471,1,1,986,747,https://www.flickr.com/photos/mauroguanandi/31119538711/,https://creativecommons.org/licenses/by/2.0/,mauroguanandi,https://www.flickr.com/people/mauroguanandi/\r\n134bfd93a04b88b2,https://c1.staticflickr.com/8/7284/28111216916_c5cd478e8a_o.jpg,0.333333333,0,1,1,581,962,https://www.flickr.com/photos/yourmildura/28111216916,https://creativecommons.org/licenses/by/2.0/,Your Mildura,https://www.flickr.com/people/yourmildura/\r\n4227a6e105cc0588,https://c1.staticflickr.com/6/5659/30998499956_ec5656fd7f_o.jpg,0.464993395,0,1,1,971,141,https://www.flickr.com/photos/apardavila/30998499956,https://creativecommons.org/licenses/by/2.0/,Arturo Pardavila III,https://www.flickr.com/people/apardavila/\r\n405dd60729a8c18d,https://c1.staticflickr.com/6/5604/31248667090_381a827f4b_o.jpg,0,0,0.5625,0.75,848,496,http://www.flickr.com/photos/mohit_s/31248667090/,https://creativecommons.org/licenses/by/2.0/,Mohit S,https://www.flickr.com/people/mohit_s/\r\n0d29832e5da60f5f,https://c1.staticflickr.com/9/8160/29190648723_bcf0356317_o.jpg,0.25,0,1,1,973,181,https://www.flickr.com/photos/jsjgeology/29190648723,https://creativecommons.org/licenses/by/2.0/,James St. John,https://www.flickr.com/people/jsjgeology/\r\n90ab38bb74020d5a,https://c1.staticflickr.com/9/8072/28451660015_5953f09d2a_o.jpg,0.5,0,1,0.75,647,141,https://www.flickr.com/photos/tipsfortravellers/28451660015,https://creativecommons.org/licenses/by/2.0/,Gary Bembridge,https://www.flickr.com/people/tipsfortravellers/\r\n068c5d66662a837a,https://c1.staticflickr.com/8/7330/26495171353_a3fff0f136_o.jpg,0,0.25,0.573459716,1,964,773,https://www.flickr.com/photos/lafoodie/26495171353,https://creativecommons.org/licenses/by/2.0/,L.A. Foodie,https://www.flickr.com/people/lafoodie/\r\ne5f15610a22098bb,https://c1.staticflickr.com/9/8407/29490211104_10c2f48229_o.jpg,0.303658537,0.25,0.91097561,1,672,575,https://www.flickr.com/photos/ophilos/29490211104,https://creativecommons.org/licenses/by/2.0/,Christos Loufopoulos,https://www.flickr.com/people/ophilos/\r\n4cd90a6459ce9571,https://c1.staticflickr.com/8/7513/29088533881_89782643f3_o.jpg,0.234308131,0,0.702924394,0.75,719,622,https://www.flickr.com/photos/tdlucas5000/29088533881,https://creativecommons.org/licenses/by/2.0/,tdlucas5000,https://www.flickr.com/people/tdlucas5000/\r\n4a0bac33f8e81e60,https://c1.staticflickr.com/9/8551/29475088050_cbbd9ea8a2_o.jpg,0,0,0.669128508,1,407,857,https://www.flickr.com/photos/eltb/29475088050,https://creativecommons.org/licenses/by/2.0/,Catedrales e Iglesias/Cathedrals and Churches,https://www.flickr.com/people/eltb/\r\ncc283b8f7f46fbf0,https://c1.staticflickr.com/9/8026/29591945975_73a66c762b_o.jpg,0.4375,0,1,0.75,574,962,https://www.flickr.com/photos/mauroguanandi/29591945975,https://creativecommons.org/licenses/by/2.0/,mauroguanandi,https://www.flickr.com/people/mauroguanandi/\r\n70b9b8ae11327472,https://c1.staticflickr.com/9/8138/30013050040_04d3fdd00b_o.jpg,0,0,1,1,11,983,https://www.flickr.com/photos/46183897@N00/30013050040,https://creativecommons.org/licenses/by/2.0/,Robert Nunnally,https://www.flickr.com/people/46183897@N00/\r\n5dd8e133e66ecd3a,https://c1.staticflickr.com/1/477/32384785852_132146bba9_o.jpg,0,0,0.643394649,0.75,142,346,https://flickr.com/11076453@N00/32384785852,https://creativecommons.org/licenses/by/2.0/,Steve Childs,https://www.flickr.com/people/steve_childs/\r\n4dc360d125c9238c,https://c1.staticflickr.com/8/7516/26892006331_d8fff23f32_o.jpg,0.25,0,0.75,0.75,527,840,https://www.flickr.com/photos/johovac/26892006331,https://creativecommons.org/licenses/by/2.0/,Ed S. Johovac,https://www.flickr.com/people/johovac/\r\n8805b0b20d808c49,https://c1.staticflickr.com/8/7728/27175361700_dd1959dd51_o.jpg,0.5,0.25,1,1,704,927,https://www.flickr.com/photos/faceme/27175361700,https://creativecommons.org/licenses/by/2.0/,FaceMePLS,https://www.flickr.com/people/faceme/\r\n15d27dbdc9eeda7f,https://c1.staticflickr.com/1/704/32389054391_5ac6fee7e7_o.jpg,0,0,0.671597633,1,469,765,https://www.flickr.com/photos/54144402@N03/32389054391,https://creativecommons.org/licenses/by/2.0/,Bob Dass,https://www.flickr.com/people/54144402@N03/\r\n448a738f6dcba07e,https://c1.staticflickr.com/8/7755/28840807203_405a538ce8_o.jpg,0,0.25,0.5,1,563,416,https://www.flickr.com/photos/harshlight/28840807203/,https://creativecommons.org/licenses/by/2.0/,HarshLight,https://www.flickr.com/people/harshlight/\r\ne5948d6294eb8ca3,https://c1.staticflickr.com/6/5480/31214754031_7eae5eeede_o.jpg,0,0,0.5625,0.75,887,676,https://www.flickr.com/photos/davidstanleytravel/31214754031,https://creativecommons.org/licenses/by/2.0/,David Stanley,https://www.flickr.com/people/davidstanleytravel/"
},
{
"path": "dataset/download_images.py",
"content": "\"\"\"Script which downloads dataset images.\n\nUsage:\n python download_images.py --input_file=INPUT_FILE --output_dir=IMAGES_DIR\n\nwhere:\n INPUT_FILE is input csv file with dataset description, i.e. dev_dataset.csv\n IMAGES_DIR is output directory where all images should be downloaded\n\nExample:\n # create directory for images\n mkdir images\n # download images declared in dev_dataset.csv\n python download_images.py --input_file=dev_dataset.csv --output_dir=images\n\n\nDependencies:\n Python 2.7 or higher.\n Pillow library: https://python-pillow.org/\n\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport argparse\nimport csv\nimport os\nimport sys\n\nfrom PIL import Image\nfrom io import BytesIO\n\ntry:\n from urllib.request import urlopen\nexcept ImportError:\n from urllib2 import urlopen\n\ndef parse_args():\n \"\"\"Parses command line arguments.\"\"\"\n parser = argparse.ArgumentParser(\n description='Tool to download dataset images.')\n parser.add_argument('--input_file', required=True,\n help='Location of dataset.csv')\n parser.add_argument('--output_dir', required=True,\n help='Output path to download images')\n args = parser.parse_args()\n return args.input_file, args.output_dir\n\n\ndef download_image(image_id, url, x1, y1, x2, y2, output_dir):\n \"\"\"Downloads one image, crops it, resizes it and saves it locally.\"\"\"\n output_filename = os.path.join(output_dir, image_id + '.png')\n if os.path.exists(output_filename):\n # Don't download image if it's already there\n return True\n try:\n # Download image\n url_file = urlopen(url)\n if url_file.getcode() != 200:\n return False\n image_buffer = url_file.read()\n # Crop, resize and save image\n image = Image.open(BytesIO(image_buffer)).convert('RGB')\n w = image.size[0]\n h = image.size[1]\n image = image.crop((int(x1*w), int(y1*h), int(x2*w), int(y2*h)))\n image = image.resize((299, 299), resample=Image.ANTIALIAS)\n image.save(output_filename)\n except IOError:\n return False\n return True\n\n\ndef main():\n input_filename, output_dir = parse_args()\n failed_to_download = set()\n with open(input_filename) as input_file:\n reader = csv.reader(input_file)\n header_row = next(reader)\n try:\n row_idx_image_id = header_row.index('ImageId')\n row_idx_url = header_row.index('URL')\n row_idx_x1 = header_row.index('x1')\n row_idx_y1 = header_row.index('y1')\n row_idx_x2 = header_row.index('x2')\n row_idx_y2 = header_row.index('y2')\n except ValueError as e:\n print('One of the columns was not found in the source file: ', e.message)\n\n for idx, row in enumerate(reader):\n if len(row) < len(header_row):\n # skip partial or empty lines\n continue\n if not download_image(image_id=row[row_idx_image_id],\n url=row[row_idx_url],\n x1=float(row[row_idx_x1]),\n y1=float(row[row_idx_y1]),\n x2=float(row[row_idx_x2]),\n y2=float(row[row_idx_y2]),\n output_dir=output_dir):\n failed_to_download.add(row[row_idx_image_id])\n sys.stdout.write('\\rDownloaded {0} images'.format(idx + 1))\n sys.stdout.flush()\n\n print()\n if failed_to_download:\n print('\\nUnable to download images with the following IDs:')\n for image_id in failed_to_download:\n print(image_id)\n\n\nif __name__ == '__main__':\n main()\n"
},
{
"path": "metadata.json",
"content": "{\n \"type\": \"attack\",\n \"container\": \"tensorflow/tensorflow:1.6.0\",\n \"container_gpu\": \"tensorflow/tensorflow:1.6.0-gpu\",\n \"entry_point\": \"run_attack.sh\"\n}\n"
},
{
"path": "nets/__init__.py",
"content": "\n"
},
{
"path": "nets/alexnet.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains a model definition for AlexNet.\n\nThis work was first described in:\n ImageNet Classification with Deep Convolutional Neural Networks\n Alex Krizhevsky, Ilya Sutskever and Geoffrey E. Hinton\n\nand later refined in:\n One weird trick for parallelizing convolutional neural networks\n Alex Krizhevsky, 2014\n\nHere we provide the implementation proposed in \"One weird trick\" and not\n\"ImageNet Classification\", as per the paper, the LRN layers have been removed.\n\nUsage:\n with slim.arg_scope(alexnet.alexnet_v2_arg_scope()):\n outputs, end_points = alexnet.alexnet_v2(inputs)\n\n@@alexnet_v2\n\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\ntrunc_normal = lambda stddev: tf.truncated_normal_initializer(0.0, stddev)\n\n\ndef alexnet_v2_arg_scope(weight_decay=0.0005):\n with slim.arg_scope([slim.conv2d, slim.fully_connected],\n activation_fn=tf.nn.relu,\n biases_initializer=tf.constant_initializer(0.1),\n weights_regularizer=slim.l2_regularizer(weight_decay)):\n with slim.arg_scope([slim.conv2d], padding='SAME'):\n with slim.arg_scope([slim.max_pool2d], padding='VALID') as arg_sc:\n return arg_sc\n\n\ndef alexnet_v2(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.5,\n spatial_squeeze=True,\n scope='alexnet_v2'):\n \"\"\"AlexNet version 2.\n\n Described in: http://arxiv.org/pdf/1404.5997v2.pdf\n Parameters from:\n github.com/akrizhevsky/cuda-convnet2/blob/master/layers/\n layers-imagenet-1gpu.cfg\n\n Note: All the fully_connected layers have been transformed to conv2d layers.\n To use in classification mode, resize input to 224x224. To use in fully\n convolutional mode, set spatial_squeeze to false.\n The LRN layers have been removed and change the initializers from\n random_normal_initializer to xavier_initializer.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether or not the model is being trained.\n dropout_keep_prob: the probability that activations are kept in the dropout\n layers during training.\n spatial_squeeze: whether or not should squeeze the spatial dimensions of the\n outputs. Useful to remove unnecessary dimensions for classification.\n scope: Optional scope for the variables.\n\n Returns:\n the last op containing the log predictions and end_points dict.\n \"\"\"\n with tf.variable_scope(scope, 'alexnet_v2', [inputs]) as sc:\n end_points_collection = sc.name + '_end_points'\n # Collect outputs for conv2d, fully_connected and max_pool2d.\n with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.max_pool2d],\n outputs_collections=[end_points_collection]):\n net = slim.conv2d(inputs, 64, [11, 11], 4, padding='VALID',\n scope='conv1')\n net = slim.max_pool2d(net, [3, 3], 2, scope='pool1')\n net = slim.conv2d(net, 192, [5, 5], scope='conv2')\n net = slim.max_pool2d(net, [3, 3], 2, scope='pool2')\n net = slim.conv2d(net, 384, [3, 3], scope='conv3')\n net = slim.conv2d(net, 384, [3, 3], scope='conv4')\n net = slim.conv2d(net, 256, [3, 3], scope='conv5')\n net = slim.max_pool2d(net, [3, 3], 2, scope='pool5')\n\n # Use conv2d instead of fully_connected layers.\n with slim.arg_scope([slim.conv2d],\n weights_initializer=trunc_normal(0.005),\n biases_initializer=tf.constant_initializer(0.1)):\n net = slim.conv2d(net, 4096, [5, 5], padding='VALID',\n scope='fc6')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout6')\n net = slim.conv2d(net, 4096, [1, 1], scope='fc7')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout7')\n net = slim.conv2d(net, num_classes, [1, 1],\n activation_fn=None,\n normalizer_fn=None,\n biases_initializer=tf.zeros_initializer(),\n scope='fc8')\n\n # Convert end_points_collection into a end_point dict.\n end_points = slim.utils.convert_collection_to_dict(end_points_collection)\n if spatial_squeeze:\n net = tf.squeeze(net, [1, 2], name='fc8/squeezed')\n end_points[sc.name + '/fc8'] = net\n return net, end_points\nalexnet_v2.default_image_size = 224\n"
},
{
"path": "nets/alexnet_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for slim.nets.alexnet.\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import alexnet\n\nslim = tf.contrib.slim\n\n\nclass AlexnetV2Test(tf.test.TestCase):\n\n def testBuild(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = alexnet.alexnet_v2(inputs, num_classes)\n self.assertEquals(logits.op.name, 'alexnet_v2/fc8/squeezed')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testFullyConvolutional(self):\n batch_size = 1\n height, width = 300, 400\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = alexnet.alexnet_v2(inputs, num_classes, spatial_squeeze=False)\n self.assertEquals(logits.op.name, 'alexnet_v2/fc8/BiasAdd')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, 4, 7, num_classes])\n\n def testEndPoints(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = alexnet.alexnet_v2(inputs, num_classes)\n expected_names = ['alexnet_v2/conv1',\n 'alexnet_v2/pool1',\n 'alexnet_v2/conv2',\n 'alexnet_v2/pool2',\n 'alexnet_v2/conv3',\n 'alexnet_v2/conv4',\n 'alexnet_v2/conv5',\n 'alexnet_v2/pool5',\n 'alexnet_v2/fc6',\n 'alexnet_v2/fc7',\n 'alexnet_v2/fc8'\n ]\n self.assertSetEqual(set(end_points.keys()), set(expected_names))\n\n def testModelVariables(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n alexnet.alexnet_v2(inputs, num_classes)\n expected_names = ['alexnet_v2/conv1/weights',\n 'alexnet_v2/conv1/biases',\n 'alexnet_v2/conv2/weights',\n 'alexnet_v2/conv2/biases',\n 'alexnet_v2/conv3/weights',\n 'alexnet_v2/conv3/biases',\n 'alexnet_v2/conv4/weights',\n 'alexnet_v2/conv4/biases',\n 'alexnet_v2/conv5/weights',\n 'alexnet_v2/conv5/biases',\n 'alexnet_v2/fc6/weights',\n 'alexnet_v2/fc6/biases',\n 'alexnet_v2/fc7/weights',\n 'alexnet_v2/fc7/biases',\n 'alexnet_v2/fc8/weights',\n 'alexnet_v2/fc8/biases',\n ]\n model_variables = [v.op.name for v in slim.get_model_variables()]\n self.assertSetEqual(set(model_variables), set(expected_names))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = alexnet.alexnet_v2(eval_inputs, is_training=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n predictions = tf.argmax(logits, 1)\n self.assertListEqual(predictions.get_shape().as_list(), [batch_size])\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 2\n eval_batch_size = 1\n train_height, train_width = 224, 224\n eval_height, eval_width = 300, 400\n num_classes = 1000\n with self.test_session():\n train_inputs = tf.random_uniform(\n (train_batch_size, train_height, train_width, 3))\n logits, _ = alexnet.alexnet_v2(train_inputs)\n self.assertListEqual(logits.get_shape().as_list(),\n [train_batch_size, num_classes])\n tf.get_variable_scope().reuse_variables()\n eval_inputs = tf.random_uniform(\n (eval_batch_size, eval_height, eval_width, 3))\n logits, _ = alexnet.alexnet_v2(eval_inputs, is_training=False,\n spatial_squeeze=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [eval_batch_size, 4, 7, num_classes])\n logits = tf.reduce_mean(logits, [1, 2])\n predictions = tf.argmax(logits, 1)\n self.assertEquals(predictions.get_shape().as_list(), [eval_batch_size])\n\n def testForward(self):\n batch_size = 1\n height, width = 224, 224\n with self.test_session() as sess:\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = alexnet.alexnet_v2(inputs)\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits)\n self.assertTrue(output.any())\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/cifarnet.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains a variant of the CIFAR-10 model definition.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\n\ntrunc_normal = lambda stddev: tf.truncated_normal_initializer(stddev=stddev)\n\n\ndef cifarnet(images, num_classes=10, is_training=False,\n dropout_keep_prob=0.5,\n prediction_fn=slim.softmax,\n scope='CifarNet'):\n \"\"\"Creates a variant of the CifarNet model.\n\n Note that since the output is a set of 'logits', the values fall in the\n interval of (-infinity, infinity). Consequently, to convert the outputs to a\n probability distribution over the characters, one will need to convert them\n using the softmax function:\n\n logits = cifarnet.cifarnet(images, is_training=False)\n probabilities = tf.nn.softmax(logits)\n predictions = tf.argmax(logits, 1)\n\n Args:\n images: A batch of `Tensors` of size [batch_size, height, width, channels].\n num_classes: the number of classes in the dataset.\n is_training: specifies whether or not we're currently training the model.\n This variable will determine the behaviour of the dropout layer.\n dropout_keep_prob: the percentage of activation values that are retained.\n prediction_fn: a function to get predictions out of logits.\n scope: Optional variable_scope.\n\n Returns:\n logits: the pre-softmax activations, a tensor of size\n [batch_size, `num_classes`]\n end_points: a dictionary from components of the network to the corresponding\n activation.\n \"\"\"\n end_points = {}\n\n with tf.variable_scope(scope, 'CifarNet', [images, num_classes]):\n net = slim.conv2d(images, 64, [5, 5], scope='conv1')\n end_points['conv1'] = net\n net = slim.max_pool2d(net, [2, 2], 2, scope='pool1')\n end_points['pool1'] = net\n net = tf.nn.lrn(net, 4, bias=1.0, alpha=0.001/9.0, beta=0.75, name='norm1')\n net = slim.conv2d(net, 64, [5, 5], scope='conv2')\n end_points['conv2'] = net\n net = tf.nn.lrn(net, 4, bias=1.0, alpha=0.001/9.0, beta=0.75, name='norm2')\n net = slim.max_pool2d(net, [2, 2], 2, scope='pool2')\n end_points['pool2'] = net\n net = slim.flatten(net)\n end_points['Flatten'] = net\n net = slim.fully_connected(net, 384, scope='fc3')\n end_points['fc3'] = net\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout3')\n net = slim.fully_connected(net, 192, scope='fc4')\n end_points['fc4'] = net\n logits = slim.fully_connected(net, num_classes,\n biases_initializer=tf.zeros_initializer(),\n weights_initializer=trunc_normal(1/192.0),\n weights_regularizer=None,\n activation_fn=None,\n scope='logits')\n\n end_points['Logits'] = logits\n end_points['Predictions'] = prediction_fn(logits, scope='Predictions')\n\n return logits, end_points\ncifarnet.default_image_size = 32\n\n\ndef cifarnet_arg_scope(weight_decay=0.004):\n \"\"\"Defines the default cifarnet argument scope.\n\n Args:\n weight_decay: The weight decay to use for regularizing the model.\n\n Returns:\n An `arg_scope` to use for the inception v3 model.\n \"\"\"\n with slim.arg_scope(\n [slim.conv2d],\n weights_initializer=tf.truncated_normal_initializer(stddev=5e-2),\n activation_fn=tf.nn.relu):\n with slim.arg_scope(\n [slim.fully_connected],\n biases_initializer=tf.constant_initializer(0.1),\n weights_initializer=trunc_normal(0.04),\n weights_regularizer=slim.l2_regularizer(weight_decay),\n activation_fn=tf.nn.relu) as sc:\n return sc\n"
},
{
"path": "nets/inception.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Brings all inception models under one namespace.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\n# pylint: disable=unused-import\nfrom nets.inception_resnet_v2 import inception_resnet_v2\nfrom nets.inception_resnet_v2 import inception_resnet_v2_arg_scope\nfrom nets.inception_resnet_v2 import inception_resnet_v2_base\nfrom nets.inception_v1 import inception_v1\nfrom nets.inception_v1 import inception_v1_arg_scope\nfrom nets.inception_v1 import inception_v1_base\nfrom nets.inception_v2 import inception_v2\nfrom nets.inception_v2 import inception_v2_arg_scope\nfrom nets.inception_v2 import inception_v2_base\nfrom nets.inception_v3 import inception_v3\nfrom nets.inception_v3 import inception_v3_arg_scope\nfrom nets.inception_v3 import inception_v3_base\nfrom nets.inception_v4 import inception_v4\nfrom nets.inception_v4 import inception_v4_arg_scope\nfrom nets.inception_v4 import inception_v4_base\n# pylint: enable=unused-import\n"
},
{
"path": "nets/inception_resnet_v2.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains the definition of the Inception Resnet V2 architecture.\n\nAs described in http://arxiv.org/abs/1602.07261.\n\n Inception-v4, Inception-ResNet and the Impact of Residual Connections\n on Learning\n Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, Alex Alemi\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\n\n\ndef block35(net, scale=1.0, activation_fn=tf.nn.relu, scope=None, reuse=None):\n \"\"\"Builds the 35x35 resnet block.\"\"\"\n with tf.variable_scope(scope, 'Block35', [net], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n tower_conv = slim.conv2d(net, 32, 1, scope='Conv2d_1x1')\n with tf.variable_scope('Branch_1'):\n tower_conv1_0 = slim.conv2d(net, 32, 1, scope='Conv2d_0a_1x1')\n tower_conv1_1 = slim.conv2d(tower_conv1_0, 32, 3, scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n tower_conv2_0 = slim.conv2d(net, 32, 1, scope='Conv2d_0a_1x1')\n tower_conv2_1 = slim.conv2d(tower_conv2_0, 48, 3, scope='Conv2d_0b_3x3')\n tower_conv2_2 = slim.conv2d(tower_conv2_1, 64, 3, scope='Conv2d_0c_3x3')\n mixed = tf.concat(axis=3, values=[tower_conv, tower_conv1_1, tower_conv2_2])\n up = slim.conv2d(mixed, net.get_shape()[3], 1, normalizer_fn=None,\n activation_fn=None, scope='Conv2d_1x1')\n net += scale * up\n if activation_fn:\n net = activation_fn(net)\n return net\n\n\ndef block17(net, scale=1.0, activation_fn=tf.nn.relu, scope=None, reuse=None):\n \"\"\"Builds the 17x17 resnet block.\"\"\"\n with tf.variable_scope(scope, 'Block17', [net], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n tower_conv = slim.conv2d(net, 192, 1, scope='Conv2d_1x1')\n with tf.variable_scope('Branch_1'):\n tower_conv1_0 = slim.conv2d(net, 128, 1, scope='Conv2d_0a_1x1')\n tower_conv1_1 = slim.conv2d(tower_conv1_0, 160, [1, 7],\n scope='Conv2d_0b_1x7')\n tower_conv1_2 = slim.conv2d(tower_conv1_1, 192, [7, 1],\n scope='Conv2d_0c_7x1')\n mixed = tf.concat(axis=3, values=[tower_conv, tower_conv1_2])\n up = slim.conv2d(mixed, net.get_shape()[3], 1, normalizer_fn=None,\n activation_fn=None, scope='Conv2d_1x1')\n net += scale * up\n if activation_fn:\n net = activation_fn(net)\n return net\n\n\ndef block8(net, scale=1.0, activation_fn=tf.nn.relu, scope=None, reuse=None):\n \"\"\"Builds the 8x8 resnet block.\"\"\"\n with tf.variable_scope(scope, 'Block8', [net], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n tower_conv = slim.conv2d(net, 192, 1, scope='Conv2d_1x1')\n with tf.variable_scope('Branch_1'):\n tower_conv1_0 = slim.conv2d(net, 192, 1, scope='Conv2d_0a_1x1')\n tower_conv1_1 = slim.conv2d(tower_conv1_0, 224, [1, 3],\n scope='Conv2d_0b_1x3')\n tower_conv1_2 = slim.conv2d(tower_conv1_1, 256, [3, 1],\n scope='Conv2d_0c_3x1')\n mixed = tf.concat(axis=3, values=[tower_conv, tower_conv1_2])\n up = slim.conv2d(mixed, net.get_shape()[3], 1, normalizer_fn=None,\n activation_fn=None, scope='Conv2d_1x1')\n net += scale * up\n if activation_fn:\n net = activation_fn(net)\n return net\n\n\ndef inception_resnet_v2_base(inputs,\n final_endpoint='Conv2d_7b_1x1',\n output_stride=16,\n align_feature_maps=False,\n scope=None):\n \"\"\"Inception model from http://arxiv.org/abs/1602.07261.\n\n Constructs an Inception Resnet v2 network from inputs to the given final\n endpoint. This method can construct the network up to the final inception\n block Conv2d_7b_1x1.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n final_endpoint: specifies the endpoint to construct the network up to. It\n can be one of ['Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3',\n 'MaxPool_3a_3x3', 'Conv2d_3b_1x1', 'Conv2d_4a_3x3', 'MaxPool_5a_3x3',\n 'Mixed_5b', 'Mixed_6a', 'PreAuxLogits', 'Mixed_7a', 'Conv2d_7b_1x1']\n output_stride: A scalar that specifies the requested ratio of input to\n output spatial resolution. Only supports 8 and 16.\n align_feature_maps: When true, changes all the VALID paddings in the network\n to SAME padding so that the feature maps are aligned.\n scope: Optional variable_scope.\n\n Returns:\n tensor_out: output tensor corresponding to the final_endpoint.\n end_points: a set of activations for external use, for example summaries or\n losses.\n\n Raises:\n ValueError: if final_endpoint is not set to one of the predefined values,\n or if the output_stride is not 8 or 16, or if the output_stride is 8 and\n we request an end point after 'PreAuxLogits'.\n \"\"\"\n if output_stride != 8 and output_stride != 16:\n raise ValueError('output_stride must be 8 or 16.')\n\n padding = 'SAME' if align_feature_maps else 'VALID'\n\n end_points = {}\n\n def add_and_check_final(name, net):\n end_points[name] = net\n return name == final_endpoint\n\n with tf.variable_scope(scope, 'InceptionResnetV2', [inputs]):\n with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d],\n stride=1, padding='SAME'):\n # 149 x 149 x 32\n net = slim.conv2d(inputs, 32, 3, stride=2, padding=padding,\n scope='Conv2d_1a_3x3')\n if add_and_check_final('Conv2d_1a_3x3', net): return net, end_points\n\n # 147 x 147 x 32\n net = slim.conv2d(net, 32, 3, padding=padding,\n scope='Conv2d_2a_3x3')\n if add_and_check_final('Conv2d_2a_3x3', net): return net, end_points\n # 147 x 147 x 64\n net = slim.conv2d(net, 64, 3, scope='Conv2d_2b_3x3')\n if add_and_check_final('Conv2d_2b_3x3', net): return net, end_points\n # 73 x 73 x 64\n net = slim.max_pool2d(net, 3, stride=2, padding=padding,\n scope='MaxPool_3a_3x3')\n if add_and_check_final('MaxPool_3a_3x3', net): return net, end_points\n # 73 x 73 x 80\n net = slim.conv2d(net, 80, 1, padding=padding,\n scope='Conv2d_3b_1x1')\n if add_and_check_final('Conv2d_3b_1x1', net): return net, end_points\n # 71 x 71 x 192\n net = slim.conv2d(net, 192, 3, padding=padding,\n scope='Conv2d_4a_3x3')\n if add_and_check_final('Conv2d_4a_3x3', net): return net, end_points\n # 35 x 35 x 192\n net = slim.max_pool2d(net, 3, stride=2, padding=padding,\n scope='MaxPool_5a_3x3')\n if add_and_check_final('MaxPool_5a_3x3', net): return net, end_points\n\n # 35 x 35 x 320\n with tf.variable_scope('Mixed_5b'):\n with tf.variable_scope('Branch_0'):\n tower_conv = slim.conv2d(net, 96, 1, scope='Conv2d_1x1')\n with tf.variable_scope('Branch_1'):\n tower_conv1_0 = slim.conv2d(net, 48, 1, scope='Conv2d_0a_1x1')\n tower_conv1_1 = slim.conv2d(tower_conv1_0, 64, 5,\n scope='Conv2d_0b_5x5')\n with tf.variable_scope('Branch_2'):\n tower_conv2_0 = slim.conv2d(net, 64, 1, scope='Conv2d_0a_1x1')\n tower_conv2_1 = slim.conv2d(tower_conv2_0, 96, 3,\n scope='Conv2d_0b_3x3')\n tower_conv2_2 = slim.conv2d(tower_conv2_1, 96, 3,\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n tower_pool = slim.avg_pool2d(net, 3, stride=1, padding='SAME',\n scope='AvgPool_0a_3x3')\n tower_pool_1 = slim.conv2d(tower_pool, 64, 1,\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n [tower_conv, tower_conv1_1, tower_conv2_2, tower_pool_1], 3)\n\n if add_and_check_final('Mixed_5b', net): return net, end_points\n # TODO(alemi): Register intermediate endpoints\n net = slim.repeat(net, 10, block35, scale=0.17)\n\n # 17 x 17 x 1088 if output_stride == 8,\n # 33 x 33 x 1088 if output_stride == 16\n use_atrous = output_stride == 8\n\n with tf.variable_scope('Mixed_6a'):\n with tf.variable_scope('Branch_0'):\n tower_conv = slim.conv2d(net, 384, 3, stride=1 if use_atrous else 2,\n padding=padding,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n tower_conv1_0 = slim.conv2d(net, 256, 1, scope='Conv2d_0a_1x1')\n tower_conv1_1 = slim.conv2d(tower_conv1_0, 256, 3,\n scope='Conv2d_0b_3x3')\n tower_conv1_2 = slim.conv2d(tower_conv1_1, 384, 3,\n stride=1 if use_atrous else 2,\n padding=padding,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_2'):\n tower_pool = slim.max_pool2d(net, 3, stride=1 if use_atrous else 2,\n padding=padding,\n scope='MaxPool_1a_3x3')\n net = tf.concat([tower_conv, tower_conv1_2, tower_pool], 3)\n\n if add_and_check_final('Mixed_6a', net): return net, end_points\n\n # TODO(alemi): register intermediate endpoints\n with slim.arg_scope([slim.conv2d], rate=2 if use_atrous else 1):\n net = slim.repeat(net, 20, block17, scale=0.10)\n if add_and_check_final('PreAuxLogits', net): return net, end_points\n\n if output_stride == 8:\n # TODO(gpapan): Properly support output_stride for the rest of the net.\n raise ValueError('output_stride==8 is only supported up to the '\n 'PreAuxlogits end_point for now.')\n\n # 8 x 8 x 2080\n with tf.variable_scope('Mixed_7a'):\n with tf.variable_scope('Branch_0'):\n tower_conv = slim.conv2d(net, 256, 1, scope='Conv2d_0a_1x1')\n tower_conv_1 = slim.conv2d(tower_conv, 384, 3, stride=2,\n padding=padding,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n tower_conv1 = slim.conv2d(net, 256, 1, scope='Conv2d_0a_1x1')\n tower_conv1_1 = slim.conv2d(tower_conv1, 288, 3, stride=2,\n padding=padding,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_2'):\n tower_conv2 = slim.conv2d(net, 256, 1, scope='Conv2d_0a_1x1')\n tower_conv2_1 = slim.conv2d(tower_conv2, 288, 3,\n scope='Conv2d_0b_3x3')\n tower_conv2_2 = slim.conv2d(tower_conv2_1, 320, 3, stride=2,\n padding=padding,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_3'):\n tower_pool = slim.max_pool2d(net, 3, stride=2,\n padding=padding,\n scope='MaxPool_1a_3x3')\n net = tf.concat(\n [tower_conv_1, tower_conv1_1, tower_conv2_2, tower_pool], 3)\n\n if add_and_check_final('Mixed_7a', net): return net, end_points\n\n # TODO(alemi): register intermediate endpoints\n net = slim.repeat(net, 9, block8, scale=0.20)\n net = block8(net, activation_fn=None)\n\n # 8 x 8 x 1536\n net = slim.conv2d(net, 1536, 1, scope='Conv2d_7b_1x1')\n if add_and_check_final('Conv2d_7b_1x1', net): return net, end_points\n\n raise ValueError('final_endpoint (%s) not recognized', final_endpoint)\n\n\ndef inception_resnet_v2(inputs, num_classes=1001, is_training=True,\n dropout_keep_prob=0.8,\n reuse=None,\n scope='InceptionResnetV2',\n create_aux_logits=True):\n \"\"\"Creates the Inception Resnet V2 model.\n\n Args:\n inputs: a 4-D tensor of size [batch_size, height, width, 3].\n num_classes: number of predicted classes.\n is_training: whether is training or not.\n dropout_keep_prob: float, the fraction to keep before final layer.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n scope: Optional variable_scope.\n create_aux_logits: Whether to include the auxilliary logits.\n\n Returns:\n logits: the logits outputs of the model.\n end_points: the set of end_points from the inception model.\n \"\"\"\n end_points = {}\n\n with tf.variable_scope(scope, 'InceptionResnetV2', [inputs, num_classes],\n reuse=reuse) as scope:\n with slim.arg_scope([slim.batch_norm, slim.dropout],\n is_training=is_training):\n\n net, end_points = inception_resnet_v2_base(inputs, scope=scope)\n\n if create_aux_logits:\n with tf.variable_scope('AuxLogits'):\n aux = end_points['PreAuxLogits']\n aux = slim.avg_pool2d(aux, 5, stride=3, padding='VALID',\n scope='Conv2d_1a_3x3')\n aux = slim.conv2d(aux, 128, 1, scope='Conv2d_1b_1x1')\n aux = slim.conv2d(aux, 768, [5, 5],\n padding='VALID', scope='Conv2d_2a_5x5')\n aux = tf.reshape(aux, [aux.get_shape()[0], aux.get_shape()[-1]])\n aux = slim.fully_connected(aux, num_classes, activation_fn=None,\n scope='Logits')\n end_points['AuxLogits'] = aux\n\n with tf.variable_scope('Logits'):\n net = slim.avg_pool2d(net, [8, 8], padding='VALID',\n scope='AvgPool_1a_8x8')\n net = tf.reshape(net, [net.get_shape()[0], net.get_shape()[-1]])\n\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='Dropout')\n\n end_points['PreLogitsFlatten'] = net\n logits = slim.fully_connected(net, num_classes, activation_fn=None,\n scope='Logits')\n end_points['Logits'] = logits\n end_points['Predictions'] = tf.nn.softmax(logits, name='Predictions')\n\n return logits, end_points\ninception_resnet_v2.default_image_size = 299\n\n\ndef inception_resnet_v2_arg_scope(weight_decay=0.00004,\n batch_norm_decay=0.9997,\n batch_norm_epsilon=0.001):\n \"\"\"Returns the scope with the default parameters for inception_resnet_v2.\n\n Args:\n weight_decay: the weight decay for weights variables.\n batch_norm_decay: decay for the moving average of batch_norm momentums.\n batch_norm_epsilon: small float added to variance to avoid dividing by zero.\n\n Returns:\n a arg_scope with the parameters needed for inception_resnet_v2.\n \"\"\"\n # Set weight_decay for weights in conv2d and fully_connected layers.\n with slim.arg_scope([slim.conv2d, slim.fully_connected],\n weights_regularizer=slim.l2_regularizer(weight_decay),\n biases_regularizer=slim.l2_regularizer(weight_decay)):\n\n batch_norm_params = {\n 'decay': batch_norm_decay,\n 'epsilon': batch_norm_epsilon,\n }\n # Set activation_fn and parameters for batch_norm.\n with slim.arg_scope([slim.conv2d], activation_fn=tf.nn.relu,\n normalizer_fn=slim.batch_norm,\n normalizer_params=batch_norm_params) as scope:\n return scope\n"
},
{
"path": "nets/inception_resnet_v2_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for slim.inception_resnet_v2.\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import inception\n\n\nclass InceptionTest(tf.test.TestCase):\n\n def testBuildLogits(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, endpoints = inception.inception_resnet_v2(inputs, num_classes)\n self.assertTrue('AuxLogits' in endpoints)\n auxlogits = endpoints['AuxLogits']\n self.assertTrue(\n auxlogits.op.name.startswith('InceptionResnetV2/AuxLogits'))\n self.assertListEqual(auxlogits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue(logits.op.name.startswith('InceptionResnetV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testBuildWithoutAuxLogits(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, endpoints = inception.inception_resnet_v2(inputs, num_classes,\n create_aux_logits=False)\n self.assertTrue('AuxLogits' not in endpoints)\n self.assertTrue(logits.op.name.startswith('InceptionResnetV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testBuildEndPoints(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_resnet_v2(inputs, num_classes)\n self.assertTrue('Logits' in end_points)\n logits = end_points['Logits']\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue('AuxLogits' in end_points)\n aux_logits = end_points['AuxLogits']\n self.assertListEqual(aux_logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Conv2d_7b_1x1']\n self.assertListEqual(pre_pool.get_shape().as_list(),\n [batch_size, 8, 8, 1536])\n\n def testBuildBaseNetwork(self):\n batch_size = 5\n height, width = 299, 299\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n net, end_points = inception.inception_resnet_v2_base(inputs)\n self.assertTrue(net.op.name.startswith('InceptionResnetV2/Conv2d_7b_1x1'))\n self.assertListEqual(net.get_shape().as_list(),\n [batch_size, 8, 8, 1536])\n expected_endpoints = ['Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3',\n 'MaxPool_3a_3x3', 'Conv2d_3b_1x1', 'Conv2d_4a_3x3',\n 'MaxPool_5a_3x3', 'Mixed_5b', 'Mixed_6a',\n 'PreAuxLogits', 'Mixed_7a', 'Conv2d_7b_1x1']\n self.assertItemsEqual(end_points.keys(), expected_endpoints)\n\n def testBuildOnlyUptoFinalEndpoint(self):\n batch_size = 5\n height, width = 299, 299\n endpoints = ['Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3',\n 'MaxPool_3a_3x3', 'Conv2d_3b_1x1', 'Conv2d_4a_3x3',\n 'MaxPool_5a_3x3', 'Mixed_5b', 'Mixed_6a',\n 'PreAuxLogits', 'Mixed_7a', 'Conv2d_7b_1x1']\n for index, endpoint in enumerate(endpoints):\n with tf.Graph().as_default():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n out_tensor, end_points = inception.inception_resnet_v2_base(\n inputs, final_endpoint=endpoint)\n if endpoint != 'PreAuxLogits':\n self.assertTrue(out_tensor.op.name.startswith(\n 'InceptionResnetV2/' + endpoint))\n self.assertItemsEqual(endpoints[:index+1], end_points)\n\n def testBuildAndCheckAllEndPointsUptoPreAuxLogits(self):\n batch_size = 5\n height, width = 299, 299\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_resnet_v2_base(\n inputs, final_endpoint='PreAuxLogits')\n endpoints_shapes = {'Conv2d_1a_3x3': [5, 149, 149, 32],\n 'Conv2d_2a_3x3': [5, 147, 147, 32],\n 'Conv2d_2b_3x3': [5, 147, 147, 64],\n 'MaxPool_3a_3x3': [5, 73, 73, 64],\n 'Conv2d_3b_1x1': [5, 73, 73, 80],\n 'Conv2d_4a_3x3': [5, 71, 71, 192],\n 'MaxPool_5a_3x3': [5, 35, 35, 192],\n 'Mixed_5b': [5, 35, 35, 320],\n 'Mixed_6a': [5, 17, 17, 1088],\n 'PreAuxLogits': [5, 17, 17, 1088]\n }\n\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name in endpoints_shapes:\n expected_shape = endpoints_shapes[endpoint_name]\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testBuildAndCheckAllEndPointsUptoPreAuxLogitsWithAlignedFeatureMaps(self):\n batch_size = 5\n height, width = 299, 299\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_resnet_v2_base(\n inputs, final_endpoint='PreAuxLogits', align_feature_maps=True)\n endpoints_shapes = {'Conv2d_1a_3x3': [5, 150, 150, 32],\n 'Conv2d_2a_3x3': [5, 150, 150, 32],\n 'Conv2d_2b_3x3': [5, 150, 150, 64],\n 'MaxPool_3a_3x3': [5, 75, 75, 64],\n 'Conv2d_3b_1x1': [5, 75, 75, 80],\n 'Conv2d_4a_3x3': [5, 75, 75, 192],\n 'MaxPool_5a_3x3': [5, 38, 38, 192],\n 'Mixed_5b': [5, 38, 38, 320],\n 'Mixed_6a': [5, 19, 19, 1088],\n 'PreAuxLogits': [5, 19, 19, 1088]\n }\n\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name in endpoints_shapes:\n expected_shape = endpoints_shapes[endpoint_name]\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testBuildAndCheckAllEndPointsUptoPreAuxLogitsWithOutputStrideEight(self):\n batch_size = 5\n height, width = 299, 299\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_resnet_v2_base(\n inputs, final_endpoint='PreAuxLogits', output_stride=8)\n endpoints_shapes = {'Conv2d_1a_3x3': [5, 149, 149, 32],\n 'Conv2d_2a_3x3': [5, 147, 147, 32],\n 'Conv2d_2b_3x3': [5, 147, 147, 64],\n 'MaxPool_3a_3x3': [5, 73, 73, 64],\n 'Conv2d_3b_1x1': [5, 73, 73, 80],\n 'Conv2d_4a_3x3': [5, 71, 71, 192],\n 'MaxPool_5a_3x3': [5, 35, 35, 192],\n 'Mixed_5b': [5, 35, 35, 320],\n 'Mixed_6a': [5, 33, 33, 1088],\n 'PreAuxLogits': [5, 33, 33, 1088]\n }\n\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name in endpoints_shapes:\n expected_shape = endpoints_shapes[endpoint_name]\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testVariablesSetDevice(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n # Force all Variables to reside on the device.\n with tf.variable_scope('on_cpu'), tf.device('/cpu:0'):\n inception.inception_resnet_v2(inputs, num_classes)\n with tf.variable_scope('on_gpu'), tf.device('/gpu:0'):\n inception.inception_resnet_v2(inputs, num_classes)\n for v in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='on_cpu'):\n self.assertDeviceEqual(v.device, '/cpu:0')\n for v in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='on_gpu'):\n self.assertDeviceEqual(v.device, '/gpu:0')\n\n def testHalfSizeImages(self):\n batch_size = 5\n height, width = 150, 150\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_resnet_v2(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionResnetV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Conv2d_7b_1x1']\n self.assertListEqual(pre_pool.get_shape().as_list(),\n [batch_size, 3, 3, 1536])\n\n def testUnknownBatchSize(self):\n batch_size = 1\n height, width = 299, 299\n num_classes = 1000\n with self.test_session() as sess:\n inputs = tf.placeholder(tf.float32, (None, height, width, 3))\n logits, _ = inception.inception_resnet_v2(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionResnetV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, num_classes])\n images = tf.random_uniform((batch_size, height, width, 3))\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEquals(output.shape, (batch_size, num_classes))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 299, 299\n num_classes = 1000\n with self.test_session() as sess:\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = inception.inception_resnet_v2(eval_inputs,\n num_classes,\n is_training=False)\n predictions = tf.argmax(logits, 1)\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (batch_size,))\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 5\n eval_batch_size = 2\n height, width = 150, 150\n num_classes = 1000\n with self.test_session() as sess:\n train_inputs = tf.random_uniform((train_batch_size, height, width, 3))\n inception.inception_resnet_v2(train_inputs, num_classes)\n eval_inputs = tf.random_uniform((eval_batch_size, height, width, 3))\n logits, _ = inception.inception_resnet_v2(eval_inputs,\n num_classes,\n is_training=False,\n reuse=True)\n predictions = tf.argmax(logits, 1)\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (eval_batch_size,))\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/inception_utils.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains common code shared by all inception models.\n\nUsage of arg scope:\n with slim.arg_scope(inception_arg_scope()):\n logits, end_points = inception.inception_v3(images, num_classes,\n is_training=is_training)\n\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\n\n\ndef inception_arg_scope(weight_decay=0.00004,\n use_batch_norm=True,\n batch_norm_decay=0.9997,\n batch_norm_epsilon=0.001):\n \"\"\"Defines the default arg scope for inception models.\n\n Args:\n weight_decay: The weight decay to use for regularizing the model.\n use_batch_norm: \"If `True`, batch_norm is applied after each convolution.\n batch_norm_decay: Decay for batch norm moving average.\n batch_norm_epsilon: Small float added to variance to avoid dividing by zero\n in batch norm.\n\n Returns:\n An `arg_scope` to use for the inception models.\n \"\"\"\n batch_norm_params = {\n # Decay for the moving averages.\n 'decay': batch_norm_decay,\n # epsilon to prevent 0s in variance.\n 'epsilon': batch_norm_epsilon,\n # collection containing update_ops.\n 'updates_collections': tf.GraphKeys.UPDATE_OPS,\n }\n if use_batch_norm:\n normalizer_fn = slim.batch_norm\n normalizer_params = batch_norm_params\n else:\n normalizer_fn = None\n normalizer_params = {}\n # Set weight_decay for weights in Conv and FC layers.\n with slim.arg_scope([slim.conv2d, slim.fully_connected],\n weights_regularizer=slim.l2_regularizer(weight_decay)):\n with slim.arg_scope(\n [slim.conv2d],\n weights_initializer=slim.variance_scaling_initializer(),\n activation_fn=tf.nn.relu,\n normalizer_fn=normalizer_fn,\n normalizer_params=normalizer_params) as sc:\n return sc\n"
},
{
"path": "nets/inception_v1.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains the definition for inception v1 classification network.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import inception_utils\n\nslim = tf.contrib.slim\ntrunc_normal = lambda stddev: tf.truncated_normal_initializer(0.0, stddev)\n\n\ndef inception_v1_base(inputs,\n final_endpoint='Mixed_5c',\n scope='InceptionV1'):\n \"\"\"Defines the Inception V1 base architecture.\n\n This architecture is defined in:\n Going deeper with convolutions\n Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed,\n Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich.\n http://arxiv.org/pdf/1409.4842v1.pdf.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n final_endpoint: specifies the endpoint to construct the network up to. It\n can be one of ['Conv2d_1a_7x7', 'MaxPool_2a_3x3', 'Conv2d_2b_1x1',\n 'Conv2d_2c_3x3', 'MaxPool_3a_3x3', 'Mixed_3b', 'Mixed_3c',\n 'MaxPool_4a_3x3', 'Mixed_4b', 'Mixed_4c', 'Mixed_4d', 'Mixed_4e',\n 'Mixed_4f', 'MaxPool_5a_2x2', 'Mixed_5b', 'Mixed_5c']\n scope: Optional variable_scope.\n\n Returns:\n A dictionary from components of the network to the corresponding activation.\n\n Raises:\n ValueError: if final_endpoint is not set to one of the predefined values.\n \"\"\"\n end_points = {}\n with tf.variable_scope(scope, 'InceptionV1', [inputs]):\n with slim.arg_scope(\n [slim.conv2d, slim.fully_connected],\n weights_initializer=trunc_normal(0.01)):\n with slim.arg_scope([slim.conv2d, slim.max_pool2d],\n stride=1, padding='SAME'):\n end_point = 'Conv2d_1a_7x7'\n net = slim.conv2d(inputs, 64, [7, 7], stride=2, scope=end_point)\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n end_point = 'MaxPool_2a_3x3'\n net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n end_point = 'Conv2d_2b_1x1'\n net = slim.conv2d(net, 64, [1, 1], scope=end_point)\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n end_point = 'Conv2d_2c_3x3'\n net = slim.conv2d(net, 192, [3, 3], scope=end_point)\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n end_point = 'MaxPool_3a_3x3'\n net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_3b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 64, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 96, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 128, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 16, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 32, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 32, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_3c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 128, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 128, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 192, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'MaxPool_4a_3x3'\n net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_4b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 192, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 96, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 208, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 16, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 48, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_4c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 160, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 112, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 224, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 24, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 64, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_4d'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 128, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 128, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 256, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 24, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 64, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_4e'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 112, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 144, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 288, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 64, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 64, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_4f'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 256, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 160, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 320, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 128, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 128, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'MaxPool_5a_2x2'\n net = slim.max_pool2d(net, [2, 2], stride=2, scope=end_point)\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_5b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 256, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 160, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 320, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 32, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 128, [3, 3], scope='Conv2d_0a_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 128, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n\n end_point = 'Mixed_5c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 384, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 192, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 384, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, 48, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 128, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 128, [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if final_endpoint == end_point: return net, end_points\n raise ValueError('Unknown final endpoint %s' % final_endpoint)\n\n\ndef inception_v1(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.8,\n prediction_fn=slim.softmax,\n spatial_squeeze=True,\n reuse=None,\n scope='InceptionV1'):\n \"\"\"Defines the Inception V1 architecture.\n\n This architecture is defined in:\n\n Going deeper with convolutions\n Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed,\n Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich.\n http://arxiv.org/pdf/1409.4842v1.pdf.\n\n The default image size used to train this network is 224x224.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether is training or not.\n dropout_keep_prob: the percentage of activation values that are retained.\n prediction_fn: a function to get predictions out of logits.\n spatial_squeeze: if True, logits is of shape [B, C], if false logits is of\n shape [B, 1, 1, C], where B is batch_size and C is number of classes.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n scope: Optional variable_scope.\n\n Returns:\n logits: the pre-softmax activations, a tensor of size\n [batch_size, num_classes]\n end_points: a dictionary from components of the network to the corresponding\n activation.\n \"\"\"\n # Final pooling and prediction\n with tf.variable_scope(scope, 'InceptionV1', [inputs, num_classes],\n reuse=reuse) as scope:\n with slim.arg_scope([slim.batch_norm, slim.dropout],\n is_training=is_training):\n net, end_points = inception_v1_base(inputs, scope=scope)\n with tf.variable_scope('Logits'):\n net = slim.avg_pool2d(net, [7, 7], stride=1, scope='AvgPool_0a_7x7')\n net = slim.dropout(net,\n dropout_keep_prob, scope='Dropout_0b')\n logits = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,\n normalizer_fn=None, scope='Conv2d_0c_1x1')\n if spatial_squeeze:\n logits = tf.squeeze(logits, [1, 2], name='SpatialSqueeze')\n\n end_points['Logits'] = logits\n end_points['Predictions'] = prediction_fn(logits, scope='Predictions')\n return logits, end_points\ninception_v1.default_image_size = 224\n\ninception_v1_arg_scope = inception_utils.inception_arg_scope\n"
},
{
"path": "nets/inception_v1_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for nets.inception_v1.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport numpy as np\nimport tensorflow as tf\n\nfrom nets import inception\n\nslim = tf.contrib.slim\n\n\nclass InceptionV1Test(tf.test.TestCase):\n\n def testBuildClassificationNetwork(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_v1(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV1/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue('Predictions' in end_points)\n self.assertListEqual(end_points['Predictions'].get_shape().as_list(),\n [batch_size, num_classes])\n\n def testBuildBaseNetwork(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n mixed_6c, end_points = inception.inception_v1_base(inputs)\n self.assertTrue(mixed_6c.op.name.startswith('InceptionV1/Mixed_5c'))\n self.assertListEqual(mixed_6c.get_shape().as_list(),\n [batch_size, 7, 7, 1024])\n expected_endpoints = ['Conv2d_1a_7x7', 'MaxPool_2a_3x3', 'Conv2d_2b_1x1',\n 'Conv2d_2c_3x3', 'MaxPool_3a_3x3', 'Mixed_3b',\n 'Mixed_3c', 'MaxPool_4a_3x3', 'Mixed_4b', 'Mixed_4c',\n 'Mixed_4d', 'Mixed_4e', 'Mixed_4f', 'MaxPool_5a_2x2',\n 'Mixed_5b', 'Mixed_5c']\n self.assertItemsEqual(end_points.keys(), expected_endpoints)\n\n def testBuildOnlyUptoFinalEndpoint(self):\n batch_size = 5\n height, width = 224, 224\n endpoints = ['Conv2d_1a_7x7', 'MaxPool_2a_3x3', 'Conv2d_2b_1x1',\n 'Conv2d_2c_3x3', 'MaxPool_3a_3x3', 'Mixed_3b', 'Mixed_3c',\n 'MaxPool_4a_3x3', 'Mixed_4b', 'Mixed_4c', 'Mixed_4d',\n 'Mixed_4e', 'Mixed_4f', 'MaxPool_5a_2x2', 'Mixed_5b',\n 'Mixed_5c']\n for index, endpoint in enumerate(endpoints):\n with tf.Graph().as_default():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n out_tensor, end_points = inception.inception_v1_base(\n inputs, final_endpoint=endpoint)\n self.assertTrue(out_tensor.op.name.startswith(\n 'InceptionV1/' + endpoint))\n self.assertItemsEqual(endpoints[:index+1], end_points)\n\n def testBuildAndCheckAllEndPointsUptoMixed5c(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v1_base(inputs,\n final_endpoint='Mixed_5c')\n endpoints_shapes = {'Conv2d_1a_7x7': [5, 112, 112, 64],\n 'MaxPool_2a_3x3': [5, 56, 56, 64],\n 'Conv2d_2b_1x1': [5, 56, 56, 64],\n 'Conv2d_2c_3x3': [5, 56, 56, 192],\n 'MaxPool_3a_3x3': [5, 28, 28, 192],\n 'Mixed_3b': [5, 28, 28, 256],\n 'Mixed_3c': [5, 28, 28, 480],\n 'MaxPool_4a_3x3': [5, 14, 14, 480],\n 'Mixed_4b': [5, 14, 14, 512],\n 'Mixed_4c': [5, 14, 14, 512],\n 'Mixed_4d': [5, 14, 14, 512],\n 'Mixed_4e': [5, 14, 14, 528],\n 'Mixed_4f': [5, 14, 14, 832],\n 'MaxPool_5a_2x2': [5, 7, 7, 832],\n 'Mixed_5b': [5, 7, 7, 832],\n 'Mixed_5c': [5, 7, 7, 1024]}\n\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name in endpoints_shapes:\n expected_shape = endpoints_shapes[endpoint_name]\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testModelHasExpectedNumberOfParameters(self):\n batch_size = 5\n height, width = 224, 224\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope(inception.inception_v1_arg_scope()):\n inception.inception_v1_base(inputs)\n total_params, _ = slim.model_analyzer.analyze_vars(\n slim.get_model_variables())\n self.assertAlmostEqual(5607184, total_params)\n\n def testHalfSizeImages(self):\n batch_size = 5\n height, width = 112, 112\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n mixed_5c, _ = inception.inception_v1_base(inputs)\n self.assertTrue(mixed_5c.op.name.startswith('InceptionV1/Mixed_5c'))\n self.assertListEqual(mixed_5c.get_shape().as_list(),\n [batch_size, 4, 4, 1024])\n\n def testUnknownImageShape(self):\n tf.reset_default_graph()\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n input_np = np.random.uniform(0, 1, (batch_size, height, width, 3))\n with self.test_session() as sess:\n inputs = tf.placeholder(tf.float32, shape=(batch_size, None, None, 3))\n logits, end_points = inception.inception_v1(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV1/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Mixed_5c']\n feed_dict = {inputs: input_np}\n tf.global_variables_initializer().run()\n pre_pool_out = sess.run(pre_pool, feed_dict=feed_dict)\n self.assertListEqual(list(pre_pool_out.shape), [batch_size, 7, 7, 1024])\n\n def testUnknowBatchSize(self):\n batch_size = 1\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.placeholder(tf.float32, (None, height, width, 3))\n logits, _ = inception.inception_v1(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV1/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, num_classes])\n images = tf.random_uniform((batch_size, height, width, 3))\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEquals(output.shape, (batch_size, num_classes))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = inception.inception_v1(eval_inputs, num_classes,\n is_training=False)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (batch_size,))\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 5\n eval_batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n\n train_inputs = tf.random_uniform((train_batch_size, height, width, 3))\n inception.inception_v1(train_inputs, num_classes)\n eval_inputs = tf.random_uniform((eval_batch_size, height, width, 3))\n logits, _ = inception.inception_v1(eval_inputs, num_classes, reuse=True)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (eval_batch_size,))\n\n def testLogitsNotSqueezed(self):\n num_classes = 25\n images = tf.random_uniform([1, 224, 224, 3])\n logits, _ = inception.inception_v1(images,\n num_classes=num_classes,\n spatial_squeeze=False)\n\n with self.test_session() as sess:\n tf.global_variables_initializer().run()\n logits_out = sess.run(logits)\n self.assertListEqual(list(logits_out.shape), [1, 1, 1, num_classes])\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/inception_v2.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains the definition for inception v2 classification network.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import inception_utils\n\nslim = tf.contrib.slim\ntrunc_normal = lambda stddev: tf.truncated_normal_initializer(0.0, stddev)\n\n\ndef inception_v2_base(inputs,\n final_endpoint='Mixed_5c',\n min_depth=16,\n depth_multiplier=1.0,\n use_separable_conv=True,\n data_format='NHWC',\n scope=None):\n \"\"\"Inception v2 (6a2).\n\n Constructs an Inception v2 network from inputs to the given final endpoint.\n This method can construct the network up to the layer inception(5b) as\n described in http://arxiv.org/abs/1502.03167.\n\n Args:\n inputs: a tensor of shape [batch_size, height, width, channels].\n final_endpoint: specifies the endpoint to construct the network up to. It\n can be one of ['Conv2d_1a_7x7', 'MaxPool_2a_3x3', 'Conv2d_2b_1x1',\n 'Conv2d_2c_3x3', 'MaxPool_3a_3x3', 'Mixed_3b', 'Mixed_3c', 'Mixed_4a',\n 'Mixed_4b', 'Mixed_4c', 'Mixed_4d', 'Mixed_4e', 'Mixed_5a', 'Mixed_5b',\n 'Mixed_5c'].\n min_depth: Minimum depth value (number of channels) for all convolution ops.\n Enforced when depth_multiplier < 1, and not an active constraint when\n depth_multiplier >= 1.\n depth_multiplier: Float multiplier for the depth (number of channels)\n for all convolution ops. The value must be greater than zero. Typical\n usage will be to set this value in (0, 1) to reduce the number of\n parameters or computation cost of the model.\n use_separable_conv: Use a separable convolution for the first layer\n Conv2d_1a_7x7. If this is False, use a normal convolution instead.\n data_format: Data format of the activations ('NHWC' or 'NCHW').\n scope: Optional variable_scope.\n\n Returns:\n tensor_out: output tensor corresponding to the final_endpoint.\n end_points: a set of activations for external use, for example summaries or\n losses.\n\n Raises:\n ValueError: if final_endpoint is not set to one of the predefined values,\n or depth_multiplier <= 0\n \"\"\"\n\n # end_points will collect relevant activations for external use, for example\n # summaries or losses.\n end_points = {}\n\n # Used to find thinned depths for each layer.\n if depth_multiplier <= 0:\n raise ValueError('depth_multiplier is not greater than zero.')\n depth = lambda d: max(int(d * depth_multiplier), min_depth)\n\n if data_format != 'NHWC' and data_format != 'NCHW':\n raise ValueError('data_format must be either NHWC or NCHW.')\n if data_format == 'NCHW' and use_separable_conv:\n raise ValueError(\n 'separable convolution only supports NHWC layout. NCHW data format can'\n ' only be used when use_separable_conv is False.'\n )\n\n concat_dim = 3 if data_format == 'NHWC' else 1\n with tf.variable_scope(scope, 'InceptionV2', [inputs]):\n with slim.arg_scope(\n [slim.conv2d, slim.max_pool2d, slim.avg_pool2d],\n stride=1,\n padding='SAME',\n data_format=data_format):\n\n # Note that sizes in the comments below assume an input spatial size of\n # 224x224, however, the inputs can be of any size greater 32x32.\n\n # 224 x 224 x 3\n end_point = 'Conv2d_1a_7x7'\n\n if use_separable_conv:\n # depthwise_multiplier here is different from depth_multiplier.\n # depthwise_multiplier determines the output channels of the initial\n # depthwise conv (see docs for tf.nn.separable_conv2d), while\n # depth_multiplier controls the # channels of the subsequent 1x1\n # convolution. Must have\n # in_channels * depthwise_multipler <= out_channels\n # so that the separable convolution is not overparameterized.\n depthwise_multiplier = min(int(depth(64) / 3), 8)\n net = slim.separable_conv2d(\n inputs, depth(64), [7, 7],\n depth_multiplier=depthwise_multiplier,\n stride=2,\n padding='SAME',\n weights_initializer=trunc_normal(1.0),\n scope=end_point)\n else:\n # Use a normal convolution instead of a separable convolution.\n net = slim.conv2d(\n inputs,\n depth(64), [7, 7],\n stride=2,\n weights_initializer=trunc_normal(1.0),\n scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 112 x 112 x 64\n end_point = 'MaxPool_2a_3x3'\n net = slim.max_pool2d(net, [3, 3], scope=end_point, stride=2)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 56 x 56 x 64\n end_point = 'Conv2d_2b_1x1'\n net = slim.conv2d(net, depth(64), [1, 1], scope=end_point,\n weights_initializer=trunc_normal(0.1))\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 56 x 56 x 64\n end_point = 'Conv2d_2c_3x3'\n net = slim.conv2d(net, depth(192), [3, 3], scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 56 x 56 x 192\n end_point = 'MaxPool_3a_3x3'\n net = slim.max_pool2d(net, [3, 3], scope=end_point, stride=2)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 28 x 28 x 192\n # Inception module.\n end_point = 'Mixed_3b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(64), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(64), [3, 3],\n scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(64), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(32), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 28 x 28 x 256\n end_point = 'Mixed_3c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(64), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(96), [3, 3],\n scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(64), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(64), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 28 x 28 x 320\n end_point = 'Mixed_4a'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(\n net, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_0 = slim.conv2d(branch_0, depth(160), [3, 3], stride=2,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(64), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(\n branch_1, depth(96), [3, 3], scope='Conv2d_0b_3x3')\n branch_1 = slim.conv2d(\n branch_1, depth(96), [3, 3], stride=2, scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.max_pool2d(\n net, [3, 3], stride=2, scope='MaxPool_1a_3x3')\n net = tf.concat(axis=concat_dim, values=[branch_0, branch_1, branch_2])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 14 x 14 x 576\n end_point = 'Mixed_4b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(224), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(64), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(\n branch_1, depth(96), [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(96), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 14 x 14 x 576\n end_point = 'Mixed_4c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(96), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(128), [3, 3],\n scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(96), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(128), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 14 x 14 x 576\n end_point = 'Mixed_4d'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(160), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(160), [3, 3],\n scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(160), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(160), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(96), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 14 x 14 x 576\n end_point = 'Mixed_4e'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(96), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(192), [3, 3],\n scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(160), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(192), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(192), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(96), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 14 x 14 x 576\n end_point = 'Mixed_5a'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(\n net, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_0 = slim.conv2d(branch_0, depth(192), [3, 3], stride=2,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(192), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(256), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_1 = slim.conv2d(branch_1, depth(256), [3, 3], stride=2,\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.max_pool2d(net, [3, 3], stride=2,\n scope='MaxPool_1a_3x3')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 7 x 7 x 1024\n end_point = 'Mixed_5b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(352), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(192), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(320), [3, 3],\n scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(160), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 7 x 7 x 1024\n end_point = 'Mixed_5c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(352), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(\n net, depth(192), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(320), [3, 3],\n scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(\n net, depth(192), [1, 1],\n weights_initializer=trunc_normal(0.09),\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(224), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.max_pool2d(net, [3, 3], scope='MaxPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(128), [1, 1],\n weights_initializer=trunc_normal(0.1),\n scope='Conv2d_0b_1x1')\n net = tf.concat(\n axis=concat_dim, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n raise ValueError('Unknown final endpoint %s' % final_endpoint)\n\n\ndef inception_v2(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.8,\n min_depth=16,\n depth_multiplier=1.0,\n prediction_fn=slim.softmax,\n spatial_squeeze=True,\n reuse=None,\n scope='InceptionV2'):\n \"\"\"Inception v2 model for classification.\n\n Constructs an Inception v2 network for classification as described in\n http://arxiv.org/abs/1502.03167.\n\n The default image size used to train this network is 224x224.\n\n Args:\n inputs: a tensor of shape [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether is training or not.\n dropout_keep_prob: the percentage of activation values that are retained.\n min_depth: Minimum depth value (number of channels) for all convolution ops.\n Enforced when depth_multiplier < 1, and not an active constraint when\n depth_multiplier >= 1.\n depth_multiplier: Float multiplier for the depth (number of channels)\n for all convolution ops. The value must be greater than zero. Typical\n usage will be to set this value in (0, 1) to reduce the number of\n parameters or computation cost of the model.\n prediction_fn: a function to get predictions out of logits.\n spatial_squeeze: if True, logits is of shape [B, C], if false logits is of\n shape [B, 1, 1, C], where B is batch_size and C is number of classes.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n scope: Optional variable_scope.\n\n Returns:\n logits: the pre-softmax activations, a tensor of size\n [batch_size, num_classes]\n end_points: a dictionary from components of the network to the corresponding\n activation.\n\n Raises:\n ValueError: if final_endpoint is not set to one of the predefined values,\n or depth_multiplier <= 0\n \"\"\"\n if depth_multiplier <= 0:\n raise ValueError('depth_multiplier is not greater than zero.')\n\n # Final pooling and prediction\n with tf.variable_scope(scope, 'InceptionV2', [inputs, num_classes],\n reuse=reuse) as scope:\n with slim.arg_scope([slim.batch_norm, slim.dropout],\n is_training=is_training):\n net, end_points = inception_v2_base(\n inputs, scope=scope, min_depth=min_depth,\n depth_multiplier=depth_multiplier)\n with tf.variable_scope('Logits'):\n kernel_size = _reduced_kernel_size_for_small_input(net, [7, 7])\n net = slim.avg_pool2d(net, kernel_size, padding='VALID',\n scope='AvgPool_1a_{}x{}'.format(*kernel_size))\n # 1 x 1 x 1024\n net = slim.dropout(net, keep_prob=dropout_keep_prob, scope='Dropout_1b')\n logits = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,\n normalizer_fn=None, scope='Conv2d_1c_1x1')\n if spatial_squeeze:\n logits = tf.squeeze(logits, [1, 2], name='SpatialSqueeze')\n end_points['Logits'] = logits\n end_points['Predictions'] = prediction_fn(logits, scope='Predictions')\n return logits, end_points\ninception_v2.default_image_size = 224\n\n\ndef _reduced_kernel_size_for_small_input(input_tensor, kernel_size):\n \"\"\"Define kernel size which is automatically reduced for small input.\n\n If the shape of the input images is unknown at graph construction time this\n function assumes that the input images are is large enough.\n\n Args:\n input_tensor: input tensor of size [batch_size, height, width, channels].\n kernel_size: desired kernel size of length 2: [kernel_height, kernel_width]\n\n Returns:\n a tensor with the kernel size.\n\n TODO(jrru): Make this function work with unknown shapes. Theoretically, this\n can be done with the code below. Problems are two-fold: (1) If the shape was\n known, it will be lost. (2) inception.slim.ops._two_element_tuple cannot\n handle tensors that define the kernel size.\n shape = tf.shape(input_tensor)\n return = tf.stack([tf.minimum(shape[1], kernel_size[0]),\n tf.minimum(shape[2], kernel_size[1])])\n\n \"\"\"\n shape = input_tensor.get_shape().as_list()\n if shape[1] is None or shape[2] is None:\n kernel_size_out = kernel_size\n else:\n kernel_size_out = [min(shape[1], kernel_size[0]),\n min(shape[2], kernel_size[1])]\n return kernel_size_out\n\n\ninception_v2_arg_scope = inception_utils.inception_arg_scope\n"
},
{
"path": "nets/inception_v2_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for nets.inception_v2.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport numpy as np\nimport tensorflow as tf\n\nfrom nets import inception\n\nslim = tf.contrib.slim\n\n\nclass InceptionV2Test(tf.test.TestCase):\n\n def testBuildClassificationNetwork(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_v2(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue('Predictions' in end_points)\n self.assertListEqual(end_points['Predictions'].get_shape().as_list(),\n [batch_size, num_classes])\n\n def testBuildBaseNetwork(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n mixed_5c, end_points = inception.inception_v2_base(inputs)\n self.assertTrue(mixed_5c.op.name.startswith('InceptionV2/Mixed_5c'))\n self.assertListEqual(mixed_5c.get_shape().as_list(),\n [batch_size, 7, 7, 1024])\n expected_endpoints = ['Mixed_3b', 'Mixed_3c', 'Mixed_4a', 'Mixed_4b',\n 'Mixed_4c', 'Mixed_4d', 'Mixed_4e', 'Mixed_5a',\n 'Mixed_5b', 'Mixed_5c', 'Conv2d_1a_7x7',\n 'MaxPool_2a_3x3', 'Conv2d_2b_1x1', 'Conv2d_2c_3x3',\n 'MaxPool_3a_3x3']\n self.assertItemsEqual(end_points.keys(), expected_endpoints)\n\n def testBuildOnlyUptoFinalEndpoint(self):\n batch_size = 5\n height, width = 224, 224\n endpoints = ['Conv2d_1a_7x7', 'MaxPool_2a_3x3', 'Conv2d_2b_1x1',\n 'Conv2d_2c_3x3', 'MaxPool_3a_3x3', 'Mixed_3b', 'Mixed_3c',\n 'Mixed_4a', 'Mixed_4b', 'Mixed_4c', 'Mixed_4d', 'Mixed_4e',\n 'Mixed_5a', 'Mixed_5b', 'Mixed_5c']\n for index, endpoint in enumerate(endpoints):\n with tf.Graph().as_default():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n out_tensor, end_points = inception.inception_v2_base(\n inputs, final_endpoint=endpoint)\n self.assertTrue(out_tensor.op.name.startswith(\n 'InceptionV2/' + endpoint))\n self.assertItemsEqual(endpoints[:index+1], end_points)\n\n def testBuildAndCheckAllEndPointsUptoMixed5c(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v2_base(inputs,\n final_endpoint='Mixed_5c')\n endpoints_shapes = {'Mixed_3b': [batch_size, 28, 28, 256],\n 'Mixed_3c': [batch_size, 28, 28, 320],\n 'Mixed_4a': [batch_size, 14, 14, 576],\n 'Mixed_4b': [batch_size, 14, 14, 576],\n 'Mixed_4c': [batch_size, 14, 14, 576],\n 'Mixed_4d': [batch_size, 14, 14, 576],\n 'Mixed_4e': [batch_size, 14, 14, 576],\n 'Mixed_5a': [batch_size, 7, 7, 1024],\n 'Mixed_5b': [batch_size, 7, 7, 1024],\n 'Mixed_5c': [batch_size, 7, 7, 1024],\n 'Conv2d_1a_7x7': [batch_size, 112, 112, 64],\n 'MaxPool_2a_3x3': [batch_size, 56, 56, 64],\n 'Conv2d_2b_1x1': [batch_size, 56, 56, 64],\n 'Conv2d_2c_3x3': [batch_size, 56, 56, 192],\n 'MaxPool_3a_3x3': [batch_size, 28, 28, 192]}\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name in endpoints_shapes:\n expected_shape = endpoints_shapes[endpoint_name]\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testModelHasExpectedNumberOfParameters(self):\n batch_size = 5\n height, width = 224, 224\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope(inception.inception_v2_arg_scope()):\n inception.inception_v2_base(inputs)\n total_params, _ = slim.model_analyzer.analyze_vars(\n slim.get_model_variables())\n self.assertAlmostEqual(10173112, total_params)\n\n def testBuildEndPointsWithDepthMultiplierLessThanOne(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v2(inputs, num_classes)\n\n endpoint_keys = [key for key in end_points.keys()\n if key.startswith('Mixed') or key.startswith('Conv')]\n\n _, end_points_with_multiplier = inception.inception_v2(\n inputs, num_classes, scope='depth_multiplied_net',\n depth_multiplier=0.5)\n\n for key in endpoint_keys:\n original_depth = end_points[key].get_shape().as_list()[3]\n new_depth = end_points_with_multiplier[key].get_shape().as_list()[3]\n self.assertEqual(0.5 * original_depth, new_depth)\n\n def testBuildEndPointsWithDepthMultiplierGreaterThanOne(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v2(inputs, num_classes)\n\n endpoint_keys = [key for key in end_points.keys()\n if key.startswith('Mixed') or key.startswith('Conv')]\n\n _, end_points_with_multiplier = inception.inception_v2(\n inputs, num_classes, scope='depth_multiplied_net',\n depth_multiplier=2.0)\n\n for key in endpoint_keys:\n original_depth = end_points[key].get_shape().as_list()[3]\n new_depth = end_points_with_multiplier[key].get_shape().as_list()[3]\n self.assertEqual(2.0 * original_depth, new_depth)\n\n def testRaiseValueErrorWithInvalidDepthMultiplier(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with self.assertRaises(ValueError):\n _ = inception.inception_v2(inputs, num_classes, depth_multiplier=-0.1)\n with self.assertRaises(ValueError):\n _ = inception.inception_v2(inputs, num_classes, depth_multiplier=0.0)\n\n def testBuildEndPointsWithUseSeparableConvolutionFalse(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v2_base(inputs)\n\n endpoint_keys = [\n key for key in end_points.keys()\n if key.startswith('Mixed') or key.startswith('Conv')\n ]\n\n _, end_points_with_replacement = inception.inception_v2_base(\n inputs, use_separable_conv=False)\n\n # The endpoint shapes must be equal to the original shape even when the\n # separable convolution is replaced with a normal convolution.\n for key in endpoint_keys:\n original_shape = end_points[key].get_shape().as_list()\n self.assertTrue(key in end_points_with_replacement)\n new_shape = end_points_with_replacement[key].get_shape().as_list()\n self.assertListEqual(original_shape, new_shape)\n\n def testBuildEndPointsNCHWDataFormat(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v2_base(inputs)\n\n endpoint_keys = [\n key for key in end_points.keys()\n if key.startswith('Mixed') or key.startswith('Conv')\n ]\n\n inputs_in_nchw = tf.random_uniform((batch_size, 3, height, width))\n _, end_points_with_replacement = inception.inception_v2_base(\n inputs_in_nchw, use_separable_conv=False, data_format='NCHW')\n\n # With the 'NCHW' data format, all endpoint activations have a transposed\n # shape from the original shape with the 'NHWC' layout.\n for key in endpoint_keys:\n transposed_original_shape = tf.transpose(\n end_points[key], [0, 3, 1, 2]).get_shape().as_list()\n self.assertTrue(key in end_points_with_replacement)\n new_shape = end_points_with_replacement[key].get_shape().as_list()\n self.assertListEqual(transposed_original_shape, new_shape)\n\n def testBuildErrorsForDataFormats(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n\n # 'NCWH' data format is not supported.\n with self.assertRaises(ValueError):\n _ = inception.inception_v2_base(inputs, data_format='NCWH')\n\n # 'NCHW' data format is not supported for separable convolution.\n with self.assertRaises(ValueError):\n _ = inception.inception_v2_base(inputs, data_format='NCHW')\n\n def testHalfSizeImages(self):\n batch_size = 5\n height, width = 112, 112\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_v2(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Mixed_5c']\n self.assertListEqual(pre_pool.get_shape().as_list(),\n [batch_size, 4, 4, 1024])\n\n def testUnknownImageShape(self):\n tf.reset_default_graph()\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n input_np = np.random.uniform(0, 1, (batch_size, height, width, 3))\n with self.test_session() as sess:\n inputs = tf.placeholder(tf.float32, shape=(batch_size, None, None, 3))\n logits, end_points = inception.inception_v2(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Mixed_5c']\n feed_dict = {inputs: input_np}\n tf.global_variables_initializer().run()\n pre_pool_out = sess.run(pre_pool, feed_dict=feed_dict)\n self.assertListEqual(list(pre_pool_out.shape), [batch_size, 7, 7, 1024])\n\n def testUnknowBatchSize(self):\n batch_size = 1\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.placeholder(tf.float32, (None, height, width, 3))\n logits, _ = inception.inception_v2(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV2/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, num_classes])\n images = tf.random_uniform((batch_size, height, width, 3))\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEquals(output.shape, (batch_size, num_classes))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = inception.inception_v2(eval_inputs, num_classes,\n is_training=False)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (batch_size,))\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 5\n eval_batch_size = 2\n height, width = 150, 150\n num_classes = 1000\n\n train_inputs = tf.random_uniform((train_batch_size, height, width, 3))\n inception.inception_v2(train_inputs, num_classes)\n eval_inputs = tf.random_uniform((eval_batch_size, height, width, 3))\n logits, _ = inception.inception_v2(eval_inputs, num_classes, reuse=True)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (eval_batch_size,))\n\n def testLogitsNotSqueezed(self):\n num_classes = 25\n images = tf.random_uniform([1, 224, 224, 3])\n logits, _ = inception.inception_v2(images,\n num_classes=num_classes,\n spatial_squeeze=False)\n\n with self.test_session() as sess:\n tf.global_variables_initializer().run()\n logits_out = sess.run(logits)\n self.assertListEqual(list(logits_out.shape), [1, 1, 1, num_classes])\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/inception_v3.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains the definition for inception v3 classification network.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import inception_utils\n\nslim = tf.contrib.slim\ntrunc_normal = lambda stddev: tf.truncated_normal_initializer(0.0, stddev)\n\n\ndef inception_v3_base(inputs,\n final_endpoint='Mixed_7c',\n min_depth=16,\n depth_multiplier=1.0,\n scope=None):\n \"\"\"Inception model from http://arxiv.org/abs/1512.00567.\n\n Constructs an Inception v3 network from inputs to the given final endpoint.\n This method can construct the network up to the final inception block\n Mixed_7c.\n\n Note that the names of the layers in the paper do not correspond to the names\n of the endpoints registered by this function although they build the same\n network.\n\n Here is a mapping from the old_names to the new names:\n Old name | New name\n =======================================\n conv0 | Conv2d_1a_3x3\n conv1 | Conv2d_2a_3x3\n conv2 | Conv2d_2b_3x3\n pool1 | MaxPool_3a_3x3\n conv3 | Conv2d_3b_1x1\n conv4 | Conv2d_4a_3x3\n pool2 | MaxPool_5a_3x3\n mixed_35x35x256a | Mixed_5b\n mixed_35x35x288a | Mixed_5c\n mixed_35x35x288b | Mixed_5d\n mixed_17x17x768a | Mixed_6a\n mixed_17x17x768b | Mixed_6b\n mixed_17x17x768c | Mixed_6c\n mixed_17x17x768d | Mixed_6d\n mixed_17x17x768e | Mixed_6e\n mixed_8x8x1280a | Mixed_7a\n mixed_8x8x2048a | Mixed_7b\n mixed_8x8x2048b | Mixed_7c\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n final_endpoint: specifies the endpoint to construct the network up to. It\n can be one of ['Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3',\n 'MaxPool_3a_3x3', 'Conv2d_3b_1x1', 'Conv2d_4a_3x3', 'MaxPool_5a_3x3',\n 'Mixed_5b', 'Mixed_5c', 'Mixed_5d', 'Mixed_6a', 'Mixed_6b', 'Mixed_6c',\n 'Mixed_6d', 'Mixed_6e', 'Mixed_7a', 'Mixed_7b', 'Mixed_7c'].\n min_depth: Minimum depth value (number of channels) for all convolution ops.\n Enforced when depth_multiplier < 1, and not an active constraint when\n depth_multiplier >= 1.\n depth_multiplier: Float multiplier for the depth (number of channels)\n for all convolution ops. The value must be greater than zero. Typical\n usage will be to set this value in (0, 1) to reduce the number of\n parameters or computation cost of the model.\n scope: Optional variable_scope.\n\n Returns:\n tensor_out: output tensor corresponding to the final_endpoint.\n end_points: a set of activations for external use, for example summaries or\n losses.\n\n Raises:\n ValueError: if final_endpoint is not set to one of the predefined values,\n or depth_multiplier <= 0\n \"\"\"\n # end_points will collect relevant activations for external use, for example\n # summaries or losses.\n end_points = {}\n\n if depth_multiplier <= 0:\n raise ValueError('depth_multiplier is not greater than zero.')\n depth = lambda d: max(int(d * depth_multiplier), min_depth)\n\n with tf.variable_scope(scope, 'InceptionV3', [inputs]):\n with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d],\n stride=1, padding='VALID'):\n # 299 x 299 x 3\n end_point = 'Conv2d_1a_3x3'\n net = slim.conv2d(inputs, depth(32), [3, 3], stride=2, scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 149 x 149 x 32\n end_point = 'Conv2d_2a_3x3'\n net = slim.conv2d(net, depth(32), [3, 3], scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 147 x 147 x 32\n end_point = 'Conv2d_2b_3x3'\n net = slim.conv2d(net, depth(64), [3, 3], padding='SAME', scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 147 x 147 x 64\n end_point = 'MaxPool_3a_3x3'\n net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 73 x 73 x 64\n end_point = 'Conv2d_3b_1x1'\n net = slim.conv2d(net, depth(80), [1, 1], scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 73 x 73 x 80.\n end_point = 'Conv2d_4a_3x3'\n net = slim.conv2d(net, depth(192), [3, 3], scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 71 x 71 x 192.\n end_point = 'MaxPool_5a_3x3'\n net = slim.max_pool2d(net, [3, 3], stride=2, scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # 35 x 35 x 192.\n\n # Inception blocks\n with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d],\n stride=1, padding='SAME'):\n # mixed: 35 x 35 x 256.\n end_point = 'Mixed_5b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(48), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(64), [5, 5],\n scope='Conv2d_0b_5x5')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, depth(32), [1, 1],\n scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed_1: 35 x 35 x 288.\n end_point = 'Mixed_5c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(48), [1, 1], scope='Conv2d_0b_1x1')\n branch_1 = slim.conv2d(branch_1, depth(64), [5, 5],\n scope='Conv_1_0c_5x5')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(64), [1, 1],\n scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, depth(64), [1, 1],\n scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed_2: 35 x 35 x 288.\n end_point = 'Mixed_5d'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(48), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(64), [5, 5],\n scope='Conv2d_0b_5x5')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, depth(96), [3, 3],\n scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, depth(64), [1, 1],\n scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed_3: 17 x 17 x 768.\n end_point = 'Mixed_6a'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(384), [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(64), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(96), [3, 3],\n scope='Conv2d_0b_3x3')\n branch_1 = slim.conv2d(branch_1, depth(96), [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_1x1')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.max_pool2d(net, [3, 3], stride=2, padding='VALID',\n scope='MaxPool_1a_3x3')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed4: 17 x 17 x 768.\n end_point = 'Mixed_6b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(128), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(128), [1, 7],\n scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, depth(192), [7, 1],\n scope='Conv2d_0c_7x1')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(128), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(128), [7, 1],\n scope='Conv2d_0b_7x1')\n branch_2 = slim.conv2d(branch_2, depth(128), [1, 7],\n scope='Conv2d_0c_1x7')\n branch_2 = slim.conv2d(branch_2, depth(128), [7, 1],\n scope='Conv2d_0d_7x1')\n branch_2 = slim.conv2d(branch_2, depth(192), [1, 7],\n scope='Conv2d_0e_1x7')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, depth(192), [1, 1],\n scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed_5: 17 x 17 x 768.\n end_point = 'Mixed_6c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(160), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(160), [1, 7],\n scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, depth(192), [7, 1],\n scope='Conv2d_0c_7x1')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(160), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(160), [7, 1],\n scope='Conv2d_0b_7x1')\n branch_2 = slim.conv2d(branch_2, depth(160), [1, 7],\n scope='Conv2d_0c_1x7')\n branch_2 = slim.conv2d(branch_2, depth(160), [7, 1],\n scope='Conv2d_0d_7x1')\n branch_2 = slim.conv2d(branch_2, depth(192), [1, 7],\n scope='Conv2d_0e_1x7')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, depth(192), [1, 1],\n scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # mixed_6: 17 x 17 x 768.\n end_point = 'Mixed_6d'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(160), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(160), [1, 7],\n scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, depth(192), [7, 1],\n scope='Conv2d_0c_7x1')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(160), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(160), [7, 1],\n scope='Conv2d_0b_7x1')\n branch_2 = slim.conv2d(branch_2, depth(160), [1, 7],\n scope='Conv2d_0c_1x7')\n branch_2 = slim.conv2d(branch_2, depth(160), [7, 1],\n scope='Conv2d_0d_7x1')\n branch_2 = slim.conv2d(branch_2, depth(192), [1, 7],\n scope='Conv2d_0e_1x7')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, depth(192), [1, 1],\n scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed_7: 17 x 17 x 768.\n end_point = 'Mixed_6e'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(192), [1, 7],\n scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, depth(192), [7, 1],\n scope='Conv2d_0c_7x1')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, depth(192), [7, 1],\n scope='Conv2d_0b_7x1')\n branch_2 = slim.conv2d(branch_2, depth(192), [1, 7],\n scope='Conv2d_0c_1x7')\n branch_2 = slim.conv2d(branch_2, depth(192), [7, 1],\n scope='Conv2d_0d_7x1')\n branch_2 = slim.conv2d(branch_2, depth(192), [1, 7],\n scope='Conv2d_0e_1x7')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, depth(192), [1, 1],\n scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed_8: 8 x 8 x 1280.\n end_point = 'Mixed_7a'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n branch_0 = slim.conv2d(branch_0, depth(320), [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(192), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, depth(192), [1, 7],\n scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, depth(192), [7, 1],\n scope='Conv2d_0c_7x1')\n branch_1 = slim.conv2d(branch_1, depth(192), [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.max_pool2d(net, [3, 3], stride=2, padding='VALID',\n scope='MaxPool_1a_3x3')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n # mixed_9: 8 x 8 x 2048.\n end_point = 'Mixed_7b'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(320), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(384), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = tf.concat(axis=3, values=[\n slim.conv2d(branch_1, depth(384), [1, 3], scope='Conv2d_0b_1x3'),\n slim.conv2d(branch_1, depth(384), [3, 1], scope='Conv2d_0b_3x1')])\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(448), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(\n branch_2, depth(384), [3, 3], scope='Conv2d_0b_3x3')\n branch_2 = tf.concat(axis=3, values=[\n slim.conv2d(branch_2, depth(384), [1, 3], scope='Conv2d_0c_1x3'),\n slim.conv2d(branch_2, depth(384), [3, 1], scope='Conv2d_0d_3x1')])\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(192), [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n\n # mixed_10: 8 x 8 x 2048.\n end_point = 'Mixed_7c'\n with tf.variable_scope(end_point):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, depth(320), [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, depth(384), [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = tf.concat(axis=3, values=[\n slim.conv2d(branch_1, depth(384), [1, 3], scope='Conv2d_0b_1x3'),\n slim.conv2d(branch_1, depth(384), [3, 1], scope='Conv2d_0c_3x1')])\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(net, depth(448), [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(\n branch_2, depth(384), [3, 3], scope='Conv2d_0b_3x3')\n branch_2 = tf.concat(axis=3, values=[\n slim.conv2d(branch_2, depth(384), [1, 3], scope='Conv2d_0c_1x3'),\n slim.conv2d(branch_2, depth(384), [3, 1], scope='Conv2d_0d_3x1')])\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(net, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(\n branch_3, depth(192), [1, 1], scope='Conv2d_0b_1x1')\n net = tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n end_points[end_point] = net\n if end_point == final_endpoint: return net, end_points\n raise ValueError('Unknown final endpoint %s' % final_endpoint)\n\n\ndef inception_v3(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.8,\n min_depth=16,\n depth_multiplier=1.0,\n prediction_fn=slim.softmax,\n spatial_squeeze=True,\n reuse=None,\n create_aux_logits=True,\n scope='InceptionV3'):\n \"\"\"Inception model from http://arxiv.org/abs/1512.00567.\n\n \"Rethinking the Inception Architecture for Computer Vision\"\n\n Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens,\n Zbigniew Wojna.\n\n With the default arguments this method constructs the exact model defined in\n the paper. However, one can experiment with variations of the inception_v3\n network by changing arguments dropout_keep_prob, min_depth and\n depth_multiplier.\n\n The default image size used to train this network is 299x299.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether is training or not.\n dropout_keep_prob: the percentage of activation values that are retained.\n min_depth: Minimum depth value (number of channels) for all convolution ops.\n Enforced when depth_multiplier < 1, and not an active constraint when\n depth_multiplier >= 1.\n depth_multiplier: Float multiplier for the depth (number of channels)\n for all convolution ops. The value must be greater than zero. Typical\n usage will be to set this value in (0, 1) to reduce the number of\n parameters or computation cost of the model.\n prediction_fn: a function to get predictions out of logits.\n spatial_squeeze: if True, logits is of shape [B, C], if false logits is of\n shape [B, 1, 1, C], where B is batch_size and C is number of classes.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n create_aux_logits: Whether to create the auxiliary logits.\n scope: Optional variable_scope.\n\n Returns:\n logits: the pre-softmax activations, a tensor of size\n [batch_size, num_classes]\n end_points: a dictionary from components of the network to the corresponding\n activation.\n\n Raises:\n ValueError: if 'depth_multiplier' is less than or equal to zero.\n \"\"\"\n if depth_multiplier <= 0:\n raise ValueError('depth_multiplier is not greater than zero.')\n depth = lambda d: max(int(d * depth_multiplier), min_depth)\n\n with tf.variable_scope(scope, 'InceptionV3', [inputs, num_classes],\n reuse=reuse) as scope:\n with slim.arg_scope([slim.batch_norm, slim.dropout],\n is_training=is_training):\n net, end_points = inception_v3_base(\n inputs, scope=scope, min_depth=min_depth,\n depth_multiplier=depth_multiplier)\n\n # Auxiliary Head logits\n if create_aux_logits:\n with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d],\n stride=1, padding='SAME'):\n aux_logits = end_points['Mixed_6e']\n with tf.variable_scope('AuxLogits'):\n aux_logits = slim.avg_pool2d(\n aux_logits, [5, 5], stride=3, padding='VALID',\n scope='AvgPool_1a_5x5')\n aux_logits = slim.conv2d(aux_logits, depth(128), [1, 1],\n scope='Conv2d_1b_1x1')\n\n # Shape of feature map before the final layer.\n kernel_size = _reduced_kernel_size_for_small_input(\n aux_logits, [5, 5])\n aux_logits = slim.conv2d(\n aux_logits, depth(768), kernel_size,\n weights_initializer=trunc_normal(0.01),\n padding='VALID', scope='Conv2d_2a_{}x{}'.format(*kernel_size))\n aux_logits = slim.conv2d(\n aux_logits, num_classes, [1, 1], activation_fn=None,\n normalizer_fn=None, weights_initializer=trunc_normal(0.001),\n scope='Conv2d_2b_1x1')\n if spatial_squeeze:\n aux_logits = tf.squeeze(aux_logits, [1, 2], name='SpatialSqueeze')\n end_points['AuxLogits'] = aux_logits\n\n # Final pooling and prediction\n with tf.variable_scope('Logits'):\n kernel_size = _reduced_kernel_size_for_small_input(net, [8, 8])\n net = slim.avg_pool2d(net, kernel_size, padding='VALID',\n scope='AvgPool_1a_{}x{}'.format(*kernel_size))\n # 1 x 1 x 2048\n net = slim.dropout(net, keep_prob=dropout_keep_prob, scope='Dropout_1b')\n end_points['PreLogits'] = net\n # 2048\n logits = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,\n normalizer_fn=None, scope='Conv2d_1c_1x1')\n if spatial_squeeze:\n logits = tf.squeeze(logits, [1, 2], name='SpatialSqueeze')\n # 1000\n end_points['Logits'] = logits\n end_points['Predictions'] = prediction_fn(logits, scope='Predictions')\n return logits, end_points\ninception_v3.default_image_size = 299\n\n\ndef _reduced_kernel_size_for_small_input(input_tensor, kernel_size):\n \"\"\"Define kernel size which is automatically reduced for small input.\n\n If the shape of the input images is unknown at graph construction time this\n function assumes that the input images are is large enough.\n\n Args:\n input_tensor: input tensor of size [batch_size, height, width, channels].\n kernel_size: desired kernel size of length 2: [kernel_height, kernel_width]\n\n Returns:\n a tensor with the kernel size.\n\n TODO(jrru): Make this function work with unknown shapes. Theoretically, this\n can be done with the code below. Problems are two-fold: (1) If the shape was\n known, it will be lost. (2) inception.slim.ops._two_element_tuple cannot\n handle tensors that define the kernel size.\n shape = tf.shape(input_tensor)\n return = tf.stack([tf.minimum(shape[1], kernel_size[0]),\n tf.minimum(shape[2], kernel_size[1])])\n\n \"\"\"\n shape = input_tensor.get_shape().as_list()\n if shape[1] is None or shape[2] is None:\n kernel_size_out = kernel_size\n else:\n kernel_size_out = [min(shape[1], kernel_size[0]),\n min(shape[2], kernel_size[1])]\n return kernel_size_out\n\n\ninception_v3_arg_scope = inception_utils.inception_arg_scope\n"
},
{
"path": "nets/inception_v3_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for nets.inception_v1.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport numpy as np\nimport tensorflow as tf\n\nfrom nets import inception\n\nslim = tf.contrib.slim\n\n\nclass InceptionV3Test(tf.test.TestCase):\n\n def testBuildClassificationNetwork(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_v3(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV3/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue('Predictions' in end_points)\n self.assertListEqual(end_points['Predictions'].get_shape().as_list(),\n [batch_size, num_classes])\n\n def testBuildBaseNetwork(self):\n batch_size = 5\n height, width = 299, 299\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n final_endpoint, end_points = inception.inception_v3_base(inputs)\n self.assertTrue(final_endpoint.op.name.startswith(\n 'InceptionV3/Mixed_7c'))\n self.assertListEqual(final_endpoint.get_shape().as_list(),\n [batch_size, 8, 8, 2048])\n expected_endpoints = ['Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3',\n 'MaxPool_3a_3x3', 'Conv2d_3b_1x1', 'Conv2d_4a_3x3',\n 'MaxPool_5a_3x3', 'Mixed_5b', 'Mixed_5c', 'Mixed_5d',\n 'Mixed_6a', 'Mixed_6b', 'Mixed_6c', 'Mixed_6d',\n 'Mixed_6e', 'Mixed_7a', 'Mixed_7b', 'Mixed_7c']\n self.assertItemsEqual(end_points.keys(), expected_endpoints)\n\n def testBuildOnlyUptoFinalEndpoint(self):\n batch_size = 5\n height, width = 299, 299\n endpoints = ['Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3',\n 'MaxPool_3a_3x3', 'Conv2d_3b_1x1', 'Conv2d_4a_3x3',\n 'MaxPool_5a_3x3', 'Mixed_5b', 'Mixed_5c', 'Mixed_5d',\n 'Mixed_6a', 'Mixed_6b', 'Mixed_6c', 'Mixed_6d',\n 'Mixed_6e', 'Mixed_7a', 'Mixed_7b', 'Mixed_7c']\n\n for index, endpoint in enumerate(endpoints):\n with tf.Graph().as_default():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n out_tensor, end_points = inception.inception_v3_base(\n inputs, final_endpoint=endpoint)\n self.assertTrue(out_tensor.op.name.startswith(\n 'InceptionV3/' + endpoint))\n self.assertItemsEqual(endpoints[:index+1], end_points)\n\n def testBuildAndCheckAllEndPointsUptoMixed7c(self):\n batch_size = 5\n height, width = 299, 299\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v3_base(\n inputs, final_endpoint='Mixed_7c')\n endpoints_shapes = {'Conv2d_1a_3x3': [batch_size, 149, 149, 32],\n 'Conv2d_2a_3x3': [batch_size, 147, 147, 32],\n 'Conv2d_2b_3x3': [batch_size, 147, 147, 64],\n 'MaxPool_3a_3x3': [batch_size, 73, 73, 64],\n 'Conv2d_3b_1x1': [batch_size, 73, 73, 80],\n 'Conv2d_4a_3x3': [batch_size, 71, 71, 192],\n 'MaxPool_5a_3x3': [batch_size, 35, 35, 192],\n 'Mixed_5b': [batch_size, 35, 35, 256],\n 'Mixed_5c': [batch_size, 35, 35, 288],\n 'Mixed_5d': [batch_size, 35, 35, 288],\n 'Mixed_6a': [batch_size, 17, 17, 768],\n 'Mixed_6b': [batch_size, 17, 17, 768],\n 'Mixed_6c': [batch_size, 17, 17, 768],\n 'Mixed_6d': [batch_size, 17, 17, 768],\n 'Mixed_6e': [batch_size, 17, 17, 768],\n 'Mixed_7a': [batch_size, 8, 8, 1280],\n 'Mixed_7b': [batch_size, 8, 8, 2048],\n 'Mixed_7c': [batch_size, 8, 8, 2048]}\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name in endpoints_shapes:\n expected_shape = endpoints_shapes[endpoint_name]\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testModelHasExpectedNumberOfParameters(self):\n batch_size = 5\n height, width = 299, 299\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope(inception.inception_v3_arg_scope()):\n inception.inception_v3_base(inputs)\n total_params, _ = slim.model_analyzer.analyze_vars(\n slim.get_model_variables())\n self.assertAlmostEqual(21802784, total_params)\n\n def testBuildEndPoints(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v3(inputs, num_classes)\n self.assertTrue('Logits' in end_points)\n logits = end_points['Logits']\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue('AuxLogits' in end_points)\n aux_logits = end_points['AuxLogits']\n self.assertListEqual(aux_logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue('Mixed_7c' in end_points)\n pre_pool = end_points['Mixed_7c']\n self.assertListEqual(pre_pool.get_shape().as_list(),\n [batch_size, 8, 8, 2048])\n self.assertTrue('PreLogits' in end_points)\n pre_logits = end_points['PreLogits']\n self.assertListEqual(pre_logits.get_shape().as_list(),\n [batch_size, 1, 1, 2048])\n\n def testBuildEndPointsWithDepthMultiplierLessThanOne(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v3(inputs, num_classes)\n\n endpoint_keys = [key for key in end_points.keys()\n if key.startswith('Mixed') or key.startswith('Conv')]\n\n _, end_points_with_multiplier = inception.inception_v3(\n inputs, num_classes, scope='depth_multiplied_net',\n depth_multiplier=0.5)\n\n for key in endpoint_keys:\n original_depth = end_points[key].get_shape().as_list()[3]\n new_depth = end_points_with_multiplier[key].get_shape().as_list()[3]\n self.assertEqual(0.5 * original_depth, new_depth)\n\n def testBuildEndPointsWithDepthMultiplierGreaterThanOne(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v3(inputs, num_classes)\n\n endpoint_keys = [key for key in end_points.keys()\n if key.startswith('Mixed') or key.startswith('Conv')]\n\n _, end_points_with_multiplier = inception.inception_v3(\n inputs, num_classes, scope='depth_multiplied_net',\n depth_multiplier=2.0)\n\n for key in endpoint_keys:\n original_depth = end_points[key].get_shape().as_list()[3]\n new_depth = end_points_with_multiplier[key].get_shape().as_list()[3]\n self.assertEqual(2.0 * original_depth, new_depth)\n\n def testRaiseValueErrorWithInvalidDepthMultiplier(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with self.assertRaises(ValueError):\n _ = inception.inception_v3(inputs, num_classes, depth_multiplier=-0.1)\n with self.assertRaises(ValueError):\n _ = inception.inception_v3(inputs, num_classes, depth_multiplier=0.0)\n\n def testHalfSizeImages(self):\n batch_size = 5\n height, width = 150, 150\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_v3(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV3/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Mixed_7c']\n self.assertListEqual(pre_pool.get_shape().as_list(),\n [batch_size, 3, 3, 2048])\n\n def testUnknownImageShape(self):\n tf.reset_default_graph()\n batch_size = 2\n height, width = 299, 299\n num_classes = 1000\n input_np = np.random.uniform(0, 1, (batch_size, height, width, 3))\n with self.test_session() as sess:\n inputs = tf.placeholder(tf.float32, shape=(batch_size, None, None, 3))\n logits, end_points = inception.inception_v3(inputs, num_classes)\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Mixed_7c']\n feed_dict = {inputs: input_np}\n tf.global_variables_initializer().run()\n pre_pool_out = sess.run(pre_pool, feed_dict=feed_dict)\n self.assertListEqual(list(pre_pool_out.shape), [batch_size, 8, 8, 2048])\n\n def testUnknowBatchSize(self):\n batch_size = 1\n height, width = 299, 299\n num_classes = 1000\n\n inputs = tf.placeholder(tf.float32, (None, height, width, 3))\n logits, _ = inception.inception_v3(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV3/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, num_classes])\n images = tf.random_uniform((batch_size, height, width, 3))\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEquals(output.shape, (batch_size, num_classes))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 299, 299\n num_classes = 1000\n\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = inception.inception_v3(eval_inputs, num_classes,\n is_training=False)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (batch_size,))\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 5\n eval_batch_size = 2\n height, width = 150, 150\n num_classes = 1000\n\n train_inputs = tf.random_uniform((train_batch_size, height, width, 3))\n inception.inception_v3(train_inputs, num_classes)\n eval_inputs = tf.random_uniform((eval_batch_size, height, width, 3))\n logits, _ = inception.inception_v3(eval_inputs, num_classes,\n is_training=False, reuse=True)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (eval_batch_size,))\n\n def testLogitsNotSqueezed(self):\n num_classes = 25\n images = tf.random_uniform([1, 299, 299, 3])\n logits, _ = inception.inception_v3(images,\n num_classes=num_classes,\n spatial_squeeze=False)\n\n with self.test_session() as sess:\n tf.global_variables_initializer().run()\n logits_out = sess.run(logits)\n self.assertListEqual(list(logits_out.shape), [1, 1, 1, num_classes])\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/inception_v4.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains the definition of the Inception V4 architecture.\n\nAs described in http://arxiv.org/abs/1602.07261.\n\n Inception-v4, Inception-ResNet and the Impact of Residual Connections\n on Learning\n Christian Szegedy, Sergey Ioffe, Vincent Vanhoucke, Alex Alemi\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import inception_utils\n\nslim = tf.contrib.slim\n\n\ndef block_inception_a(inputs, scope=None, reuse=None):\n \"\"\"Builds Inception-A block for Inception v4 network.\"\"\"\n # By default use stride=1 and SAME padding\n with slim.arg_scope([slim.conv2d, slim.avg_pool2d, slim.max_pool2d],\n stride=1, padding='SAME'):\n with tf.variable_scope(scope, 'BlockInceptionA', [inputs], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(inputs, 96, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(inputs, 64, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 96, [3, 3], scope='Conv2d_0b_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(inputs, 64, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope='Conv2d_0b_3x3')\n branch_2 = slim.conv2d(branch_2, 96, [3, 3], scope='Conv2d_0c_3x3')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(inputs, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 96, [1, 1], scope='Conv2d_0b_1x1')\n return tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n\n\ndef block_reduction_a(inputs, scope=None, reuse=None):\n \"\"\"Builds Reduction-A block for Inception v4 network.\"\"\"\n # By default use stride=1 and SAME padding\n with slim.arg_scope([slim.conv2d, slim.avg_pool2d, slim.max_pool2d],\n stride=1, padding='SAME'):\n with tf.variable_scope(scope, 'BlockReductionA', [inputs], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(inputs, 384, [3, 3], stride=2, padding='VALID',\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(inputs, 192, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 224, [3, 3], scope='Conv2d_0b_3x3')\n branch_1 = slim.conv2d(branch_1, 256, [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.max_pool2d(inputs, [3, 3], stride=2, padding='VALID',\n scope='MaxPool_1a_3x3')\n return tf.concat(axis=3, values=[branch_0, branch_1, branch_2])\n\n\ndef block_inception_b(inputs, scope=None, reuse=None):\n \"\"\"Builds Inception-B block for Inception v4 network.\"\"\"\n # By default use stride=1 and SAME padding\n with slim.arg_scope([slim.conv2d, slim.avg_pool2d, slim.max_pool2d],\n stride=1, padding='SAME'):\n with tf.variable_scope(scope, 'BlockInceptionB', [inputs], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(inputs, 384, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(inputs, 192, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 224, [1, 7], scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, 256, [7, 1], scope='Conv2d_0c_7x1')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(inputs, 192, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 192, [7, 1], scope='Conv2d_0b_7x1')\n branch_2 = slim.conv2d(branch_2, 224, [1, 7], scope='Conv2d_0c_1x7')\n branch_2 = slim.conv2d(branch_2, 224, [7, 1], scope='Conv2d_0d_7x1')\n branch_2 = slim.conv2d(branch_2, 256, [1, 7], scope='Conv2d_0e_1x7')\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(inputs, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 128, [1, 1], scope='Conv2d_0b_1x1')\n return tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n\n\ndef block_reduction_b(inputs, scope=None, reuse=None):\n \"\"\"Builds Reduction-B block for Inception v4 network.\"\"\"\n # By default use stride=1 and SAME padding\n with slim.arg_scope([slim.conv2d, slim.avg_pool2d, slim.max_pool2d],\n stride=1, padding='SAME'):\n with tf.variable_scope(scope, 'BlockReductionB', [inputs], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(inputs, 192, [1, 1], scope='Conv2d_0a_1x1')\n branch_0 = slim.conv2d(branch_0, 192, [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(inputs, 256, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 256, [1, 7], scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, 320, [7, 1], scope='Conv2d_0c_7x1')\n branch_1 = slim.conv2d(branch_1, 320, [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.max_pool2d(inputs, [3, 3], stride=2, padding='VALID',\n scope='MaxPool_1a_3x3')\n return tf.concat(axis=3, values=[branch_0, branch_1, branch_2])\n\n\ndef block_inception_c(inputs, scope=None, reuse=None):\n \"\"\"Builds Inception-C block for Inception v4 network.\"\"\"\n # By default use stride=1 and SAME padding\n with slim.arg_scope([slim.conv2d, slim.avg_pool2d, slim.max_pool2d],\n stride=1, padding='SAME'):\n with tf.variable_scope(scope, 'BlockInceptionC', [inputs], reuse=reuse):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(inputs, 256, [1, 1], scope='Conv2d_0a_1x1')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(inputs, 384, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = tf.concat(axis=3, values=[\n slim.conv2d(branch_1, 256, [1, 3], scope='Conv2d_0b_1x3'),\n slim.conv2d(branch_1, 256, [3, 1], scope='Conv2d_0c_3x1')])\n with tf.variable_scope('Branch_2'):\n branch_2 = slim.conv2d(inputs, 384, [1, 1], scope='Conv2d_0a_1x1')\n branch_2 = slim.conv2d(branch_2, 448, [3, 1], scope='Conv2d_0b_3x1')\n branch_2 = slim.conv2d(branch_2, 512, [1, 3], scope='Conv2d_0c_1x3')\n branch_2 = tf.concat(axis=3, values=[\n slim.conv2d(branch_2, 256, [1, 3], scope='Conv2d_0d_1x3'),\n slim.conv2d(branch_2, 256, [3, 1], scope='Conv2d_0e_3x1')])\n with tf.variable_scope('Branch_3'):\n branch_3 = slim.avg_pool2d(inputs, [3, 3], scope='AvgPool_0a_3x3')\n branch_3 = slim.conv2d(branch_3, 256, [1, 1], scope='Conv2d_0b_1x1')\n return tf.concat(axis=3, values=[branch_0, branch_1, branch_2, branch_3])\n\n\ndef inception_v4_base(inputs, final_endpoint='Mixed_7d', scope=None):\n \"\"\"Creates the Inception V4 network up to the given final endpoint.\n\n Args:\n inputs: a 4-D tensor of size [batch_size, height, width, 3].\n final_endpoint: specifies the endpoint to construct the network up to.\n It can be one of [ 'Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3',\n 'Mixed_3a', 'Mixed_4a', 'Mixed_5a', 'Mixed_5b', 'Mixed_5c', 'Mixed_5d',\n 'Mixed_5e', 'Mixed_6a', 'Mixed_6b', 'Mixed_6c', 'Mixed_6d', 'Mixed_6e',\n 'Mixed_6f', 'Mixed_6g', 'Mixed_6h', 'Mixed_7a', 'Mixed_7b', 'Mixed_7c',\n 'Mixed_7d']\n scope: Optional variable_scope.\n\n Returns:\n logits: the logits outputs of the model.\n end_points: the set of end_points from the inception model.\n\n Raises:\n ValueError: if final_endpoint is not set to one of the predefined values,\n \"\"\"\n end_points = {}\n\n def add_and_check_final(name, net):\n end_points[name] = net\n return name == final_endpoint\n\n with tf.variable_scope(scope, 'InceptionV4', [inputs]):\n with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d],\n stride=1, padding='SAME'):\n # 299 x 299 x 3\n net = slim.conv2d(inputs, 32, [3, 3], stride=2,\n padding='VALID', scope='Conv2d_1a_3x3')\n if add_and_check_final('Conv2d_1a_3x3', net): return net, end_points\n # 149 x 149 x 32\n net = slim.conv2d(net, 32, [3, 3], padding='VALID',\n scope='Conv2d_2a_3x3')\n if add_and_check_final('Conv2d_2a_3x3', net): return net, end_points\n # 147 x 147 x 32\n net = slim.conv2d(net, 64, [3, 3], scope='Conv2d_2b_3x3')\n if add_and_check_final('Conv2d_2b_3x3', net): return net, end_points\n # 147 x 147 x 64\n with tf.variable_scope('Mixed_3a'):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.max_pool2d(net, [3, 3], stride=2, padding='VALID',\n scope='MaxPool_0a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 96, [3, 3], stride=2, padding='VALID',\n scope='Conv2d_0a_3x3')\n net = tf.concat(axis=3, values=[branch_0, branch_1])\n if add_and_check_final('Mixed_3a', net): return net, end_points\n\n # 73 x 73 x 160\n with tf.variable_scope('Mixed_4a'):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 64, [1, 1], scope='Conv2d_0a_1x1')\n branch_0 = slim.conv2d(branch_0, 96, [3, 3], padding='VALID',\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.conv2d(net, 64, [1, 1], scope='Conv2d_0a_1x1')\n branch_1 = slim.conv2d(branch_1, 64, [1, 7], scope='Conv2d_0b_1x7')\n branch_1 = slim.conv2d(branch_1, 64, [7, 1], scope='Conv2d_0c_7x1')\n branch_1 = slim.conv2d(branch_1, 96, [3, 3], padding='VALID',\n scope='Conv2d_1a_3x3')\n net = tf.concat(axis=3, values=[branch_0, branch_1])\n if add_and_check_final('Mixed_4a', net): return net, end_points\n\n # 71 x 71 x 192\n with tf.variable_scope('Mixed_5a'):\n with tf.variable_scope('Branch_0'):\n branch_0 = slim.conv2d(net, 192, [3, 3], stride=2, padding='VALID',\n scope='Conv2d_1a_3x3')\n with tf.variable_scope('Branch_1'):\n branch_1 = slim.max_pool2d(net, [3, 3], stride=2, padding='VALID',\n scope='MaxPool_1a_3x3')\n net = tf.concat(axis=3, values=[branch_0, branch_1])\n if add_and_check_final('Mixed_5a', net): return net, end_points\n\n # 35 x 35 x 384\n # 4 x Inception-A blocks\n for idx in range(4):\n block_scope = 'Mixed_5' + chr(ord('b') + idx)\n net = block_inception_a(net, block_scope)\n if add_and_check_final(block_scope, net): return net, end_points\n\n # 35 x 35 x 384\n # Reduction-A block\n net = block_reduction_a(net, 'Mixed_6a')\n if add_and_check_final('Mixed_6a', net): return net, end_points\n\n # 17 x 17 x 1024\n # 7 x Inception-B blocks\n for idx in range(7):\n block_scope = 'Mixed_6' + chr(ord('b') + idx)\n net = block_inception_b(net, block_scope)\n if add_and_check_final(block_scope, net): return net, end_points\n\n # 17 x 17 x 1024\n # Reduction-B block\n net = block_reduction_b(net, 'Mixed_7a')\n if add_and_check_final('Mixed_7a', net): return net, end_points\n\n # 8 x 8 x 1536\n # 3 x Inception-C blocks\n for idx in range(3):\n block_scope = 'Mixed_7' + chr(ord('b') + idx)\n net = block_inception_c(net, block_scope)\n if add_and_check_final(block_scope, net): return net, end_points\n raise ValueError('Unknown final endpoint %s' % final_endpoint)\n\n\ndef inception_v4(inputs, num_classes=1001, is_training=True,\n dropout_keep_prob=0.8,\n reuse=None,\n scope='InceptionV4',\n create_aux_logits=True):\n \"\"\"Creates the Inception V4 model.\n\n Args:\n inputs: a 4-D tensor of size [batch_size, height, width, 3].\n num_classes: number of predicted classes.\n is_training: whether is training or not.\n dropout_keep_prob: float, the fraction to keep before final layer.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n scope: Optional variable_scope.\n create_aux_logits: Whether to include the auxiliary logits.\n\n Returns:\n logits: the logits outputs of the model.\n end_points: the set of end_points from the inception model.\n \"\"\"\n end_points = {}\n with tf.variable_scope(scope, 'InceptionV4', [inputs], reuse=reuse) as scope:\n with slim.arg_scope([slim.batch_norm, slim.dropout],\n is_training=is_training):\n net, end_points = inception_v4_base(inputs, scope=scope)\n\n with slim.arg_scope([slim.conv2d, slim.max_pool2d, slim.avg_pool2d],\n stride=1, padding='SAME'):\n # Auxiliary Head logits\n if create_aux_logits:\n with tf.variable_scope('AuxLogits'):\n # 17 x 17 x 1024\n aux_logits = end_points['Mixed_6h']\n aux_logits = slim.avg_pool2d(aux_logits, [5, 5], stride=3,\n padding='VALID',\n scope='AvgPool_1a_5x5')\n aux_logits = slim.conv2d(aux_logits, 128, [1, 1],\n scope='Conv2d_1b_1x1')\n aux_logits = slim.conv2d(aux_logits, 768,\n [5, 5],\n padding='VALID', scope='Conv2d_2a')\n aux_logits = tf.reshape(aux_logits, [aux_logits.get_shape()[0], aux_logits.get_shape()[-1]])\n aux_logits = slim.fully_connected(aux_logits, num_classes,\n activation_fn=None,\n scope='Aux_logits')\n end_points['AuxLogits'] = aux_logits\n\n # Final pooling and prediction\n with tf.variable_scope('Logits'):\n # 8 x 8 x 1536\n net = slim.avg_pool2d(net, [8, 8], padding='VALID',\n scope='AvgPool_1a')\n # 1 x 1 x 1536\n net = slim.dropout(net, dropout_keep_prob, scope='Dropout_1b')\n net = tf.reshape(net, [net.get_shape()[0], net.get_shape()[-1]])\n end_points['PreLogitsFlatten'] = net\n # 1536\n logits = slim.fully_connected(net, num_classes, activation_fn=None,\n scope='Logits')\n end_points['Logits'] = logits\n end_points['Predictions'] = tf.nn.softmax(logits, name='Predictions')\n return logits, end_points\ninception_v4.default_image_size = 299\n\n\ninception_v4_arg_scope = inception_utils.inception_arg_scope\n"
},
{
"path": "nets/inception_v4_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for slim.inception_v4.\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import inception\n\n\nclass InceptionTest(tf.test.TestCase):\n\n def testBuildLogits(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_v4(inputs, num_classes)\n auxlogits = end_points['AuxLogits']\n predictions = end_points['Predictions']\n self.assertTrue(auxlogits.op.name.startswith('InceptionV4/AuxLogits'))\n self.assertListEqual(auxlogits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue(logits.op.name.startswith('InceptionV4/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue(predictions.op.name.startswith(\n 'InceptionV4/Logits/Predictions'))\n self.assertListEqual(predictions.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testBuildWithoutAuxLogits(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, endpoints = inception.inception_v4(inputs, num_classes,\n create_aux_logits=False)\n self.assertFalse('AuxLogits' in endpoints)\n self.assertTrue(logits.op.name.startswith('InceptionV4/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testAllEndPointsShapes(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = inception.inception_v4(inputs, num_classes)\n endpoints_shapes = {'Conv2d_1a_3x3': [batch_size, 149, 149, 32],\n 'Conv2d_2a_3x3': [batch_size, 147, 147, 32],\n 'Conv2d_2b_3x3': [batch_size, 147, 147, 64],\n 'Mixed_3a': [batch_size, 73, 73, 160],\n 'Mixed_4a': [batch_size, 71, 71, 192],\n 'Mixed_5a': [batch_size, 35, 35, 384],\n # 4 x Inception-A blocks\n 'Mixed_5b': [batch_size, 35, 35, 384],\n 'Mixed_5c': [batch_size, 35, 35, 384],\n 'Mixed_5d': [batch_size, 35, 35, 384],\n 'Mixed_5e': [batch_size, 35, 35, 384],\n # Reduction-A block\n 'Mixed_6a': [batch_size, 17, 17, 1024],\n # 7 x Inception-B blocks\n 'Mixed_6b': [batch_size, 17, 17, 1024],\n 'Mixed_6c': [batch_size, 17, 17, 1024],\n 'Mixed_6d': [batch_size, 17, 17, 1024],\n 'Mixed_6e': [batch_size, 17, 17, 1024],\n 'Mixed_6f': [batch_size, 17, 17, 1024],\n 'Mixed_6g': [batch_size, 17, 17, 1024],\n 'Mixed_6h': [batch_size, 17, 17, 1024],\n # Reduction-A block\n 'Mixed_7a': [batch_size, 8, 8, 1536],\n # 3 x Inception-C blocks\n 'Mixed_7b': [batch_size, 8, 8, 1536],\n 'Mixed_7c': [batch_size, 8, 8, 1536],\n 'Mixed_7d': [batch_size, 8, 8, 1536],\n # Logits and predictions\n 'AuxLogits': [batch_size, num_classes],\n 'PreLogitsFlatten': [batch_size, 1536],\n 'Logits': [batch_size, num_classes],\n 'Predictions': [batch_size, num_classes]}\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name in endpoints_shapes:\n expected_shape = endpoints_shapes[endpoint_name]\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testBuildBaseNetwork(self):\n batch_size = 5\n height, width = 299, 299\n inputs = tf.random_uniform((batch_size, height, width, 3))\n net, end_points = inception.inception_v4_base(inputs)\n self.assertTrue(net.op.name.startswith(\n 'InceptionV4/Mixed_7d'))\n self.assertListEqual(net.get_shape().as_list(), [batch_size, 8, 8, 1536])\n expected_endpoints = [\n 'Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3', 'Mixed_3a',\n 'Mixed_4a', 'Mixed_5a', 'Mixed_5b', 'Mixed_5c', 'Mixed_5d',\n 'Mixed_5e', 'Mixed_6a', 'Mixed_6b', 'Mixed_6c', 'Mixed_6d',\n 'Mixed_6e', 'Mixed_6f', 'Mixed_6g', 'Mixed_6h', 'Mixed_7a',\n 'Mixed_7b', 'Mixed_7c', 'Mixed_7d']\n self.assertItemsEqual(end_points.keys(), expected_endpoints)\n for name, op in end_points.iteritems():\n self.assertTrue(op.name.startswith('InceptionV4/' + name))\n\n def testBuildOnlyUpToFinalEndpoint(self):\n batch_size = 5\n height, width = 299, 299\n all_endpoints = [\n 'Conv2d_1a_3x3', 'Conv2d_2a_3x3', 'Conv2d_2b_3x3', 'Mixed_3a',\n 'Mixed_4a', 'Mixed_5a', 'Mixed_5b', 'Mixed_5c', 'Mixed_5d',\n 'Mixed_5e', 'Mixed_6a', 'Mixed_6b', 'Mixed_6c', 'Mixed_6d',\n 'Mixed_6e', 'Mixed_6f', 'Mixed_6g', 'Mixed_6h', 'Mixed_7a',\n 'Mixed_7b', 'Mixed_7c', 'Mixed_7d']\n for index, endpoint in enumerate(all_endpoints):\n with tf.Graph().as_default():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n out_tensor, end_points = inception.inception_v4_base(\n inputs, final_endpoint=endpoint)\n self.assertTrue(out_tensor.op.name.startswith(\n 'InceptionV4/' + endpoint))\n self.assertItemsEqual(all_endpoints[:index+1], end_points)\n\n def testVariablesSetDevice(self):\n batch_size = 5\n height, width = 299, 299\n num_classes = 1000\n inputs = tf.random_uniform((batch_size, height, width, 3))\n # Force all Variables to reside on the device.\n with tf.variable_scope('on_cpu'), tf.device('/cpu:0'):\n inception.inception_v4(inputs, num_classes)\n with tf.variable_scope('on_gpu'), tf.device('/gpu:0'):\n inception.inception_v4(inputs, num_classes)\n for v in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='on_cpu'):\n self.assertDeviceEqual(v.device, '/cpu:0')\n for v in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='on_gpu'):\n self.assertDeviceEqual(v.device, '/gpu:0')\n\n def testHalfSizeImages(self):\n batch_size = 5\n height, width = 150, 150\n num_classes = 1000\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = inception.inception_v4(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV4/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Mixed_7d']\n self.assertListEqual(pre_pool.get_shape().as_list(),\n [batch_size, 3, 3, 1536])\n\n def testUnknownBatchSize(self):\n batch_size = 1\n height, width = 299, 299\n num_classes = 1000\n with self.test_session() as sess:\n inputs = tf.placeholder(tf.float32, (None, height, width, 3))\n logits, _ = inception.inception_v4(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('InceptionV4/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, num_classes])\n images = tf.random_uniform((batch_size, height, width, 3))\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEquals(output.shape, (batch_size, num_classes))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 299, 299\n num_classes = 1000\n with self.test_session() as sess:\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = inception.inception_v4(eval_inputs,\n num_classes,\n is_training=False)\n predictions = tf.argmax(logits, 1)\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (batch_size,))\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 5\n eval_batch_size = 2\n height, width = 150, 150\n num_classes = 1000\n with self.test_session() as sess:\n train_inputs = tf.random_uniform((train_batch_size, height, width, 3))\n inception.inception_v4(train_inputs, num_classes)\n eval_inputs = tf.random_uniform((eval_batch_size, height, width, 3))\n logits, _ = inception.inception_v4(eval_inputs,\n num_classes,\n is_training=False,\n reuse=True)\n predictions = tf.argmax(logits, 1)\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (eval_batch_size,))\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/lenet.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains a variant of the LeNet model definition.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\n\n\ndef lenet(images, num_classes=10, is_training=False,\n dropout_keep_prob=0.5,\n prediction_fn=slim.softmax,\n scope='LeNet'):\n \"\"\"Creates a variant of the LeNet model.\n\n Note that since the output is a set of 'logits', the values fall in the\n interval of (-infinity, infinity). Consequently, to convert the outputs to a\n probability distribution over the characters, one will need to convert them\n using the softmax function:\n\n logits = lenet.lenet(images, is_training=False)\n probabilities = tf.nn.softmax(logits)\n predictions = tf.argmax(logits, 1)\n\n Args:\n images: A batch of `Tensors` of size [batch_size, height, width, channels].\n num_classes: the number of classes in the dataset.\n is_training: specifies whether or not we're currently training the model.\n This variable will determine the behaviour of the dropout layer.\n dropout_keep_prob: the percentage of activation values that are retained.\n prediction_fn: a function to get predictions out of logits.\n scope: Optional variable_scope.\n\n Returns:\n logits: the pre-softmax activations, a tensor of size\n [batch_size, `num_classes`]\n end_points: a dictionary from components of the network to the corresponding\n activation.\n \"\"\"\n end_points = {}\n\n with tf.variable_scope(scope, 'LeNet', [images, num_classes]):\n net = slim.conv2d(images, 32, [5, 5], scope='conv1')\n net = slim.max_pool2d(net, [2, 2], 2, scope='pool1')\n net = slim.conv2d(net, 64, [5, 5], scope='conv2')\n net = slim.max_pool2d(net, [2, 2], 2, scope='pool2')\n net = slim.flatten(net)\n end_points['Flatten'] = net\n\n net = slim.fully_connected(net, 1024, scope='fc3')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout3')\n logits = slim.fully_connected(net, num_classes, activation_fn=None,\n scope='fc4')\n\n end_points['Logits'] = logits\n end_points['Predictions'] = prediction_fn(logits, scope='Predictions')\n\n return logits, end_points\nlenet.default_image_size = 28\n\n\ndef lenet_arg_scope(weight_decay=0.0):\n \"\"\"Defines the default lenet argument scope.\n\n Args:\n weight_decay: The weight decay to use for regularizing the model.\n\n Returns:\n An `arg_scope` to use for the inception v3 model.\n \"\"\"\n with slim.arg_scope(\n [slim.conv2d, slim.fully_connected],\n weights_regularizer=slim.l2_regularizer(weight_decay),\n weights_initializer=tf.truncated_normal_initializer(stddev=0.1),\n activation_fn=tf.nn.relu) as sc:\n return sc\n"
},
{
"path": "nets/mobilenet_v1.md",
"content": "# MobileNet_v1\n\n[MobileNets](https://arxiv.org/abs/1704.04861) are small, low-latency, low-power models parameterized to meet the resource constraints of a variety of use cases. They can be built upon for classification, detection, embeddings and segmentation similar to how other popular large scale models, such as Inception, are used. MobileNets can be run efficiently on mobile devices with [TensorFlow Mobile](https://www.tensorflow.org/mobile/).\n\nMobileNets trade off between latency, size and accuracy while comparing favorably with popular models from the literature.\n\n![alt text](mobilenet_v1.png \"MobileNet Graph\")\n\n# Pre-trained Models\n\nChoose the right MobileNet model to fit your latency and size budget. The size of the network in memory and on disk is proportional to the number of parameters. The latency and power usage of the network scales with the number of Multiply-Accumulates (MACs) which measures the number of fused Multiplication and Addition operations. These MobileNet models have been trained on the\n[ILSVRC-2012-CLS](http://www.image-net.org/challenges/LSVRC/2012/)\nimage classification dataset. Accuracies were computed by evaluating using a single image crop.\n\nModel Checkpoint | Million MACs | Million Parameters | Top-1 Accuracy| Top-5 Accuracy |\n:----:|:------------:|:----------:|:-------:|:-------:|\n[MobileNet_v1_1.0_224](http://download.tensorflow.org/models/mobilenet_v1_1.0_224_2017_06_14.tar.gz)|569|4.24|70.7|89.5|\n[MobileNet_v1_1.0_192](http://download.tensorflow.org/models/mobilenet_v1_1.0_192_2017_06_14.tar.gz)|418|4.24|69.3|88.9|\n[MobileNet_v1_1.0_160](http://download.tensorflow.org/models/mobilenet_v1_1.0_160_2017_06_14.tar.gz)|291|4.24|67.2|87.5|\n[MobileNet_v1_1.0_128](http://download.tensorflow.org/models/mobilenet_v1_1.0_128_2017_06_14.tar.gz)|186|4.24|64.1|85.3|\n[MobileNet_v1_0.75_224](http://download.tensorflow.org/models/mobilenet_v1_0.75_224_2017_06_14.tar.gz)|317|2.59|68.4|88.2|\n[MobileNet_v1_0.75_192](http://download.tensorflow.org/models/mobilenet_v1_0.75_192_2017_06_14.tar.gz)|233|2.59|67.4|87.3|\n[MobileNet_v1_0.75_160](http://download.tensorflow.org/models/mobilenet_v1_0.75_160_2017_06_14.tar.gz)|162|2.59|65.2|86.1|\n[MobileNet_v1_0.75_128](http://download.tensorflow.org/models/mobilenet_v1_0.75_128_2017_06_14.tar.gz)|104|2.59|61.8|83.6|\n[MobileNet_v1_0.50_224](http://download.tensorflow.org/models/mobilenet_v1_0.50_224_2017_06_14.tar.gz)|150|1.34|64.0|85.4|\n[MobileNet_v1_0.50_192](http://download.tensorflow.org/models/mobilenet_v1_0.50_192_2017_06_14.tar.gz)|110|1.34|62.1|84.0|\n[MobileNet_v1_0.50_160](http://download.tensorflow.org/models/mobilenet_v1_0.50_160_2017_06_14.tar.gz)|77|1.34|59.9|82.5|\n[MobileNet_v1_0.50_128](http://download.tensorflow.org/models/mobilenet_v1_0.50_128_2017_06_14.tar.gz)|49|1.34|56.2|79.6|\n[MobileNet_v1_0.25_224](http://download.tensorflow.org/models/mobilenet_v1_0.25_224_2017_06_14.tar.gz)|41|0.47|50.6|75.0|\n[MobileNet_v1_0.25_192](http://download.tensorflow.org/models/mobilenet_v1_0.25_192_2017_06_14.tar.gz)|34|0.47|49.0|73.6|\n[MobileNet_v1_0.25_160](http://download.tensorflow.org/models/mobilenet_v1_0.25_160_2017_06_14.tar.gz)|21|0.47|46.0|70.7|\n[MobileNet_v1_0.25_128](http://download.tensorflow.org/models/mobilenet_v1_0.25_128_2017_06_14.tar.gz)|14|0.47|41.3|66.2|\n\n\nHere is an example of how to download the MobileNet_v1_1.0_224 checkpoint:\n\n```shell\n$ CHECKPOINT_DIR=/tmp/checkpoints\n$ mkdir ${CHECKPOINT_DIR}\n$ wget http://download.tensorflow.org/models/mobilenet_v1_1.0_224_2017_06_14.tar.gz\n$ tar -xvf mobilenet_v1_1.0_224_2017_06_14.tar.gz\n$ mv mobilenet_v1_1.0_224.ckpt.* ${CHECKPOINT_DIR}\n$ rm mobilenet_v1_1.0_224_2017_06_14.tar.gz\n```\nMore information on integrating MobileNets into your project can be found at the [TF-Slim Image Classification Library](https://github.com/tensorflow/models/blob/master/slim/README.md).\n\nTo get started running models on-device go to [TensorFlow Mobile](https://www.tensorflow.org/mobile/).\n"
},
{
"path": "nets/mobilenet_v1.py",
"content": "# Copyright 2017 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# =============================================================================\n\"\"\"MobileNet v1.\n\nMobileNet is a general architecture and can be used for multiple use cases.\nDepending on the use case, it can use different input layer size and different\nhead (for example: embeddings, localization and classification).\n\nAs described in https://arxiv.org/abs/1704.04861.\n\n MobileNets: Efficient Convolutional Neural Networks for\n Mobile Vision Applications\n Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang,\n Tobias Weyand, Marco Andreetto, Hartwig Adam\n\n100% Mobilenet V1 (base) with input size 224x224:\n\nSee mobilenet_v1()\n\nLayer params macs\n--------------------------------------------------------------------------------\nMobilenetV1/Conv2d_0/Conv2D: 864 10,838,016\nMobilenetV1/Conv2d_1_depthwise/depthwise: 288 3,612,672\nMobilenetV1/Conv2d_1_pointwise/Conv2D: 2,048 25,690,112\nMobilenetV1/Conv2d_2_depthwise/depthwise: 576 1,806,336\nMobilenetV1/Conv2d_2_pointwise/Conv2D: 8,192 25,690,112\nMobilenetV1/Conv2d_3_depthwise/depthwise: 1,152 3,612,672\nMobilenetV1/Conv2d_3_pointwise/Conv2D: 16,384 51,380,224\nMobilenetV1/Conv2d_4_depthwise/depthwise: 1,152 903,168\nMobilenetV1/Conv2d_4_pointwise/Conv2D: 32,768 25,690,112\nMobilenetV1/Conv2d_5_depthwise/depthwise: 2,304 1,806,336\nMobilenetV1/Conv2d_5_pointwise/Conv2D: 65,536 51,380,224\nMobilenetV1/Conv2d_6_depthwise/depthwise: 2,304 451,584\nMobilenetV1/Conv2d_6_pointwise/Conv2D: 131,072 25,690,112\nMobilenetV1/Conv2d_7_depthwise/depthwise: 4,608 903,168\nMobilenetV1/Conv2d_7_pointwise/Conv2D: 262,144 51,380,224\nMobilenetV1/Conv2d_8_depthwise/depthwise: 4,608 903,168\nMobilenetV1/Conv2d_8_pointwise/Conv2D: 262,144 51,380,224\nMobilenetV1/Conv2d_9_depthwise/depthwise: 4,608 903,168\nMobilenetV1/Conv2d_9_pointwise/Conv2D: 262,144 51,380,224\nMobilenetV1/Conv2d_10_depthwise/depthwise: 4,608 903,168\nMobilenetV1/Conv2d_10_pointwise/Conv2D: 262,144 51,380,224\nMobilenetV1/Conv2d_11_depthwise/depthwise: 4,608 903,168\nMobilenetV1/Conv2d_11_pointwise/Conv2D: 262,144 51,380,224\nMobilenetV1/Conv2d_12_depthwise/depthwise: 4,608 225,792\nMobilenetV1/Conv2d_12_pointwise/Conv2D: 524,288 25,690,112\nMobilenetV1/Conv2d_13_depthwise/depthwise: 9,216 451,584\nMobilenetV1/Conv2d_13_pointwise/Conv2D: 1,048,576 51,380,224\n--------------------------------------------------------------------------------\nTotal: 3,185,088 567,716,352\n\n\n75% Mobilenet V1 (base) with input size 128x128:\n\nSee mobilenet_v1_075()\n\nLayer params macs\n--------------------------------------------------------------------------------\nMobilenetV1/Conv2d_0/Conv2D: 648 2,654,208\nMobilenetV1/Conv2d_1_depthwise/depthwise: 216 884,736\nMobilenetV1/Conv2d_1_pointwise/Conv2D: 1,152 4,718,592\nMobilenetV1/Conv2d_2_depthwise/depthwise: 432 442,368\nMobilenetV1/Conv2d_2_pointwise/Conv2D: 4,608 4,718,592\nMobilenetV1/Conv2d_3_depthwise/depthwise: 864 884,736\nMobilenetV1/Conv2d_3_pointwise/Conv2D: 9,216 9,437,184\nMobilenetV1/Conv2d_4_depthwise/depthwise: 864 221,184\nMobilenetV1/Conv2d_4_pointwise/Conv2D: 18,432 4,718,592\nMobilenetV1/Conv2d_5_depthwise/depthwise: 1,728 442,368\nMobilenetV1/Conv2d_5_pointwise/Conv2D: 36,864 9,437,184\nMobilenetV1/Conv2d_6_depthwise/depthwise: 1,728 110,592\nMobilenetV1/Conv2d_6_pointwise/Conv2D: 73,728 4,718,592\nMobilenetV1/Conv2d_7_depthwise/depthwise: 3,456 221,184\nMobilenetV1/Conv2d_7_pointwise/Conv2D: 147,456 9,437,184\nMobilenetV1/Conv2d_8_depthwise/depthwise: 3,456 221,184\nMobilenetV1/Conv2d_8_pointwise/Conv2D: 147,456 9,437,184\nMobilenetV1/Conv2d_9_depthwise/depthwise: 3,456 221,184\nMobilenetV1/Conv2d_9_pointwise/Conv2D: 147,456 9,437,184\nMobilenetV1/Conv2d_10_depthwise/depthwise: 3,456 221,184\nMobilenetV1/Conv2d_10_pointwise/Conv2D: 147,456 9,437,184\nMobilenetV1/Conv2d_11_depthwise/depthwise: 3,456 221,184\nMobilenetV1/Conv2d_11_pointwise/Conv2D: 147,456 9,437,184\nMobilenetV1/Conv2d_12_depthwise/depthwise: 3,456 55,296\nMobilenetV1/Conv2d_12_pointwise/Conv2D: 294,912 4,718,592\nMobilenetV1/Conv2d_13_depthwise/depthwise: 6,912 110,592\nMobilenetV1/Conv2d_13_pointwise/Conv2D: 589,824 9,437,184\n--------------------------------------------------------------------------------\nTotal: 1,800,144 106,002,432\n\n\"\"\"\n\n# Tensorflow mandates these.\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nfrom collections import namedtuple\nimport functools\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\n\n# Conv and DepthSepConv namedtuple define layers of the MobileNet architecture\n# Conv defines 3x3 convolution layers\n# DepthSepConv defines 3x3 depthwise convolution followed by 1x1 convolution.\n# stride is the stride of the convolution\n# depth is the number of channels or filters in a layer\nConv = namedtuple('Conv', ['kernel', 'stride', 'depth'])\nDepthSepConv = namedtuple('DepthSepConv', ['kernel', 'stride', 'depth'])\n\n# _CONV_DEFS specifies the MobileNet body\n_CONV_DEFS = [\n Conv(kernel=[3, 3], stride=2, depth=32),\n DepthSepConv(kernel=[3, 3], stride=1, depth=64),\n DepthSepConv(kernel=[3, 3], stride=2, depth=128),\n DepthSepConv(kernel=[3, 3], stride=1, depth=128),\n DepthSepConv(kernel=[3, 3], stride=2, depth=256),\n DepthSepConv(kernel=[3, 3], stride=1, depth=256),\n DepthSepConv(kernel=[3, 3], stride=2, depth=512),\n DepthSepConv(kernel=[3, 3], stride=1, depth=512),\n DepthSepConv(kernel=[3, 3], stride=1, depth=512),\n DepthSepConv(kernel=[3, 3], stride=1, depth=512),\n DepthSepConv(kernel=[3, 3], stride=1, depth=512),\n DepthSepConv(kernel=[3, 3], stride=1, depth=512),\n DepthSepConv(kernel=[3, 3], stride=2, depth=1024),\n DepthSepConv(kernel=[3, 3], stride=1, depth=1024)\n]\n\n\ndef mobilenet_v1_base(inputs,\n final_endpoint='Conv2d_13_pointwise',\n min_depth=8,\n depth_multiplier=1.0,\n conv_defs=None,\n output_stride=None,\n scope=None):\n \"\"\"Mobilenet v1.\n\n Constructs a Mobilenet v1 network from inputs to the given final endpoint.\n\n Args:\n inputs: a tensor of shape [batch_size, height, width, channels].\n final_endpoint: specifies the endpoint to construct the network up to. It\n can be one of ['Conv2d_0', 'Conv2d_1_pointwise', 'Conv2d_2_pointwise',\n 'Conv2d_3_pointwise', 'Conv2d_4_pointwise', 'Conv2d_5'_pointwise,\n 'Conv2d_6_pointwise', 'Conv2d_7_pointwise', 'Conv2d_8_pointwise',\n 'Conv2d_9_pointwise', 'Conv2d_10_pointwise', 'Conv2d_11_pointwise',\n 'Conv2d_12_pointwise', 'Conv2d_13_pointwise'].\n min_depth: Minimum depth value (number of channels) for all convolution ops.\n Enforced when depth_multiplier < 1, and not an active constraint when\n depth_multiplier >= 1.\n depth_multiplier: Float multiplier for the depth (number of channels)\n for all convolution ops. The value must be greater than zero. Typical\n usage will be to set this value in (0, 1) to reduce the number of\n parameters or computation cost of the model.\n conv_defs: A list of ConvDef namedtuples specifying the net architecture.\n output_stride: An integer that specifies the requested ratio of input to\n output spatial resolution. If not None, then we invoke atrous convolution\n if necessary to prevent the network from reducing the spatial resolution\n of the activation maps. Allowed values are 8 (accurate fully convolutional\n mode), 16 (fast fully convolutional mode), 32 (classification mode).\n scope: Optional variable_scope.\n\n Returns:\n tensor_out: output tensor corresponding to the final_endpoint.\n end_points: a set of activations for external use, for example summaries or\n losses.\n\n Raises:\n ValueError: if final_endpoint is not set to one of the predefined values,\n or depth_multiplier <= 0, or the target output_stride is not\n allowed.\n \"\"\"\n depth = lambda d: max(int(d * depth_multiplier), min_depth)\n end_points = {}\n\n # Used to find thinned depths for each layer.\n if depth_multiplier <= 0:\n raise ValueError('depth_multiplier is not greater than zero.')\n\n if conv_defs is None:\n conv_defs = _CONV_DEFS\n\n if output_stride is not None and output_stride not in [8, 16, 32]:\n raise ValueError('Only allowed output_stride values are 8, 16, 32.')\n\n with tf.variable_scope(scope, 'MobilenetV1', [inputs]):\n with slim.arg_scope([slim.conv2d, slim.separable_conv2d], padding='SAME'):\n # The current_stride variable keeps track of the output stride of the\n # activations, i.e., the running product of convolution strides up to the\n # current network layer. This allows us to invoke atrous convolution\n # whenever applying the next convolution would result in the activations\n # having output stride larger than the target output_stride.\n current_stride = 1\n\n # The atrous convolution rate parameter.\n rate = 1\n\n net = inputs\n for i, conv_def in enumerate(conv_defs):\n end_point_base = 'Conv2d_%d' % i\n\n if output_stride is not None and current_stride == output_stride:\n # If we have reached the target output_stride, then we need to employ\n # atrous convolution with stride=1 and multiply the atrous rate by the\n # current unit's stride for use in subsequent layers.\n layer_stride = 1\n layer_rate = rate\n rate *= conv_def.stride\n else:\n layer_stride = conv_def.stride\n layer_rate = 1\n current_stride *= conv_def.stride\n\n if isinstance(conv_def, Conv):\n end_point = end_point_base\n net = slim.conv2d(net, depth(conv_def.depth), conv_def.kernel,\n stride=conv_def.stride,\n normalizer_fn=slim.batch_norm,\n scope=end_point)\n end_points[end_point] = net\n if end_point == final_endpoint:\n return net, end_points\n\n elif isinstance(conv_def, DepthSepConv):\n end_point = end_point_base + '_depthwise'\n\n # By passing filters=None\n # separable_conv2d produces only a depthwise convolution layer\n net = slim.separable_conv2d(net, None, conv_def.kernel,\n depth_multiplier=1,\n stride=layer_stride,\n rate=layer_rate,\n normalizer_fn=slim.batch_norm,\n scope=end_point)\n\n end_points[end_point] = net\n if end_point == final_endpoint:\n return net, end_points\n\n end_point = end_point_base + '_pointwise'\n\n net = slim.conv2d(net, depth(conv_def.depth), [1, 1],\n stride=1,\n normalizer_fn=slim.batch_norm,\n scope=end_point)\n\n end_points[end_point] = net\n if end_point == final_endpoint:\n return net, end_points\n else:\n raise ValueError('Unknown convolution type %s for layer %d'\n % (conv_def.ltype, i))\n raise ValueError('Unknown final endpoint %s' % final_endpoint)\n\n\ndef mobilenet_v1(inputs,\n num_classes=1000,\n dropout_keep_prob=0.999,\n is_training=True,\n min_depth=8,\n depth_multiplier=1.0,\n conv_defs=None,\n prediction_fn=tf.contrib.layers.softmax,\n spatial_squeeze=True,\n reuse=None,\n scope='MobilenetV1'):\n \"\"\"Mobilenet v1 model for classification.\n\n Args:\n inputs: a tensor of shape [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n dropout_keep_prob: the percentage of activation values that are retained.\n is_training: whether is training or not.\n min_depth: Minimum depth value (number of channels) for all convolution ops.\n Enforced when depth_multiplier < 1, and not an active constraint when\n depth_multiplier >= 1.\n depth_multiplier: Float multiplier for the depth (number of channels)\n for all convolution ops. The value must be greater than zero. Typical\n usage will be to set this value in (0, 1) to reduce the number of\n parameters or computation cost of the model.\n conv_defs: A list of ConvDef namedtuples specifying the net architecture.\n prediction_fn: a function to get predictions out of logits.\n spatial_squeeze: if True, logits is of shape is [B, C], if false logits is\n of shape [B, 1, 1, C], where B is batch_size and C is number of classes.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n scope: Optional variable_scope.\n\n Returns:\n logits: the pre-softmax activations, a tensor of size\n [batch_size, num_classes]\n end_points: a dictionary from components of the network to the corresponding\n activation.\n\n Raises:\n ValueError: Input rank is invalid.\n \"\"\"\n input_shape = inputs.get_shape().as_list()\n if len(input_shape) != 4:\n raise ValueError('Invalid input tensor rank, expected 4, was: %d' %\n len(input_shape))\n\n with tf.variable_scope(scope, 'MobilenetV1', [inputs, num_classes],\n reuse=reuse) as scope:\n with slim.arg_scope([slim.batch_norm, slim.dropout],\n is_training=is_training):\n net, end_points = mobilenet_v1_base(inputs, scope=scope,\n min_depth=min_depth,\n depth_multiplier=depth_multiplier,\n conv_defs=conv_defs)\n with tf.variable_scope('Logits'):\n kernel_size = _reduced_kernel_size_for_small_input(net, [7, 7])\n net = slim.avg_pool2d(net, kernel_size, padding='VALID',\n scope='AvgPool_1a')\n end_points['AvgPool_1a'] = net\n # 1 x 1 x 1024\n net = slim.dropout(net, keep_prob=dropout_keep_prob, scope='Dropout_1b')\n logits = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,\n normalizer_fn=None, scope='Conv2d_1c_1x1')\n if spatial_squeeze:\n logits = tf.squeeze(logits, [1, 2], name='SpatialSqueeze')\n end_points['Logits'] = logits\n if prediction_fn:\n end_points['Predictions'] = prediction_fn(logits, scope='Predictions')\n return logits, end_points\n\nmobilenet_v1.default_image_size = 224\n\n\ndef wrapped_partial(func, *args, **kwargs):\n partial_func = functools.partial(func, *args, **kwargs)\n functools.update_wrapper(partial_func, func)\n return partial_func\n\n\nmobilenet_v1_075 = wrapped_partial(mobilenet_v1, depth_multiplier=0.75)\nmobilenet_v1_050 = wrapped_partial(mobilenet_v1, depth_multiplier=0.50)\nmobilenet_v1_025 = wrapped_partial(mobilenet_v1, depth_multiplier=0.25)\n\n\ndef _reduced_kernel_size_for_small_input(input_tensor, kernel_size):\n \"\"\"Define kernel size which is automatically reduced for small input.\n\n If the shape of the input images is unknown at graph construction time this\n function assumes that the input images are large enough.\n\n Args:\n input_tensor: input tensor of size [batch_size, height, width, channels].\n kernel_size: desired kernel size of length 2: [kernel_height, kernel_width]\n\n Returns:\n a tensor with the kernel size.\n \"\"\"\n shape = input_tensor.get_shape().as_list()\n if shape[1] is None or shape[2] is None:\n kernel_size_out = kernel_size\n else:\n kernel_size_out = [min(shape[1], kernel_size[0]),\n min(shape[2], kernel_size[1])]\n return kernel_size_out\n\n\ndef mobilenet_v1_arg_scope(is_training=True,\n weight_decay=0.00004,\n stddev=0.09,\n regularize_depthwise=False):\n \"\"\"Defines the default MobilenetV1 arg scope.\n\n Args:\n is_training: Whether or not we're training the model.\n weight_decay: The weight decay to use for regularizing the model.\n stddev: The standard deviation of the trunctated normal weight initializer.\n regularize_depthwise: Whether or not apply regularization on depthwise.\n\n Returns:\n An `arg_scope` to use for the mobilenet v1 model.\n \"\"\"\n batch_norm_params = {\n 'is_training': is_training,\n 'center': True,\n 'scale': True,\n 'decay': 0.9997,\n 'epsilon': 0.001,\n }\n\n # Set weight_decay for weights in Conv and DepthSepConv layers.\n weights_init = tf.truncated_normal_initializer(stddev=stddev)\n regularizer = tf.contrib.layers.l2_regularizer(weight_decay)\n if regularize_depthwise:\n depthwise_regularizer = regularizer\n else:\n depthwise_regularizer = None\n with slim.arg_scope([slim.conv2d, slim.separable_conv2d],\n weights_initializer=weights_init,\n activation_fn=tf.nn.relu6, normalizer_fn=slim.batch_norm):\n with slim.arg_scope([slim.batch_norm], **batch_norm_params):\n with slim.arg_scope([slim.conv2d], weights_regularizer=regularizer):\n with slim.arg_scope([slim.separable_conv2d],\n weights_regularizer=depthwise_regularizer) as sc:\n return sc\n"
},
{
"path": "nets/mobilenet_v1_test.py",
"content": "# Copyright 2017 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# =============================================================================\n\"\"\"Tests for MobileNet v1.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport numpy as np\nimport tensorflow as tf\n\nfrom nets import mobilenet_v1\n\nslim = tf.contrib.slim\n\n\nclass MobilenetV1Test(tf.test.TestCase):\n\n def testBuildClassificationNetwork(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n self.assertTrue('Predictions' in end_points)\n self.assertListEqual(end_points['Predictions'].get_shape().as_list(),\n [batch_size, num_classes])\n\n def testBuildBaseNetwork(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n net, end_points = mobilenet_v1.mobilenet_v1_base(inputs)\n self.assertTrue(net.op.name.startswith('MobilenetV1/Conv2d_13'))\n self.assertListEqual(net.get_shape().as_list(),\n [batch_size, 7, 7, 1024])\n expected_endpoints = ['Conv2d_0',\n 'Conv2d_1_depthwise', 'Conv2d_1_pointwise',\n 'Conv2d_2_depthwise', 'Conv2d_2_pointwise',\n 'Conv2d_3_depthwise', 'Conv2d_3_pointwise',\n 'Conv2d_4_depthwise', 'Conv2d_4_pointwise',\n 'Conv2d_5_depthwise', 'Conv2d_5_pointwise',\n 'Conv2d_6_depthwise', 'Conv2d_6_pointwise',\n 'Conv2d_7_depthwise', 'Conv2d_7_pointwise',\n 'Conv2d_8_depthwise', 'Conv2d_8_pointwise',\n 'Conv2d_9_depthwise', 'Conv2d_9_pointwise',\n 'Conv2d_10_depthwise', 'Conv2d_10_pointwise',\n 'Conv2d_11_depthwise', 'Conv2d_11_pointwise',\n 'Conv2d_12_depthwise', 'Conv2d_12_pointwise',\n 'Conv2d_13_depthwise', 'Conv2d_13_pointwise']\n self.assertItemsEqual(end_points.keys(), expected_endpoints)\n\n def testBuildOnlyUptoFinalEndpoint(self):\n batch_size = 5\n height, width = 224, 224\n endpoints = ['Conv2d_0',\n 'Conv2d_1_depthwise', 'Conv2d_1_pointwise',\n 'Conv2d_2_depthwise', 'Conv2d_2_pointwise',\n 'Conv2d_3_depthwise', 'Conv2d_3_pointwise',\n 'Conv2d_4_depthwise', 'Conv2d_4_pointwise',\n 'Conv2d_5_depthwise', 'Conv2d_5_pointwise',\n 'Conv2d_6_depthwise', 'Conv2d_6_pointwise',\n 'Conv2d_7_depthwise', 'Conv2d_7_pointwise',\n 'Conv2d_8_depthwise', 'Conv2d_8_pointwise',\n 'Conv2d_9_depthwise', 'Conv2d_9_pointwise',\n 'Conv2d_10_depthwise', 'Conv2d_10_pointwise',\n 'Conv2d_11_depthwise', 'Conv2d_11_pointwise',\n 'Conv2d_12_depthwise', 'Conv2d_12_pointwise',\n 'Conv2d_13_depthwise', 'Conv2d_13_pointwise']\n for index, endpoint in enumerate(endpoints):\n with tf.Graph().as_default():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n out_tensor, end_points = mobilenet_v1.mobilenet_v1_base(\n inputs, final_endpoint=endpoint)\n self.assertTrue(out_tensor.op.name.startswith(\n 'MobilenetV1/' + endpoint))\n self.assertItemsEqual(endpoints[:index+1], end_points)\n\n def testBuildCustomNetworkUsingConvDefs(self):\n batch_size = 5\n height, width = 224, 224\n conv_defs = [\n mobilenet_v1.Conv(kernel=[3, 3], stride=2, depth=32),\n mobilenet_v1.DepthSepConv(kernel=[3, 3], stride=1, depth=64),\n mobilenet_v1.DepthSepConv(kernel=[3, 3], stride=2, depth=128),\n mobilenet_v1.DepthSepConv(kernel=[3, 3], stride=1, depth=512)\n ]\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n net, end_points = mobilenet_v1.mobilenet_v1_base(\n inputs, final_endpoint='Conv2d_3_pointwise', conv_defs=conv_defs)\n self.assertTrue(net.op.name.startswith('MobilenetV1/Conv2d_3'))\n self.assertListEqual(net.get_shape().as_list(),\n [batch_size, 56, 56, 512])\n expected_endpoints = ['Conv2d_0',\n 'Conv2d_1_depthwise', 'Conv2d_1_pointwise',\n 'Conv2d_2_depthwise', 'Conv2d_2_pointwise',\n 'Conv2d_3_depthwise', 'Conv2d_3_pointwise']\n self.assertItemsEqual(end_points.keys(), expected_endpoints)\n\n def testBuildAndCheckAllEndPointsUptoConv2d_13(self):\n batch_size = 5\n height, width = 224, 224\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope([slim.conv2d, slim.separable_conv2d],\n normalizer_fn=slim.batch_norm):\n _, end_points = mobilenet_v1.mobilenet_v1_base(\n inputs, final_endpoint='Conv2d_13_pointwise')\n endpoints_shapes = {'Conv2d_0': [batch_size, 112, 112, 32],\n 'Conv2d_1_depthwise': [batch_size, 112, 112, 32],\n 'Conv2d_1_pointwise': [batch_size, 112, 112, 64],\n 'Conv2d_2_depthwise': [batch_size, 56, 56, 64],\n 'Conv2d_2_pointwise': [batch_size, 56, 56, 128],\n 'Conv2d_3_depthwise': [batch_size, 56, 56, 128],\n 'Conv2d_3_pointwise': [batch_size, 56, 56, 128],\n 'Conv2d_4_depthwise': [batch_size, 28, 28, 128],\n 'Conv2d_4_pointwise': [batch_size, 28, 28, 256],\n 'Conv2d_5_depthwise': [batch_size, 28, 28, 256],\n 'Conv2d_5_pointwise': [batch_size, 28, 28, 256],\n 'Conv2d_6_depthwise': [batch_size, 14, 14, 256],\n 'Conv2d_6_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_7_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_7_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_8_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_8_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_9_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_9_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_10_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_10_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_11_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_11_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_12_depthwise': [batch_size, 7, 7, 512],\n 'Conv2d_12_pointwise': [batch_size, 7, 7, 1024],\n 'Conv2d_13_depthwise': [batch_size, 7, 7, 1024],\n 'Conv2d_13_pointwise': [batch_size, 7, 7, 1024]}\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name, expected_shape in endpoints_shapes.iteritems():\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testOutputStride16BuildAndCheckAllEndPointsUptoConv2d_13(self):\n batch_size = 5\n height, width = 224, 224\n output_stride = 16\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope([slim.conv2d, slim.separable_conv2d],\n normalizer_fn=slim.batch_norm):\n _, end_points = mobilenet_v1.mobilenet_v1_base(\n inputs, output_stride=output_stride,\n final_endpoint='Conv2d_13_pointwise')\n endpoints_shapes = {'Conv2d_0': [batch_size, 112, 112, 32],\n 'Conv2d_1_depthwise': [batch_size, 112, 112, 32],\n 'Conv2d_1_pointwise': [batch_size, 112, 112, 64],\n 'Conv2d_2_depthwise': [batch_size, 56, 56, 64],\n 'Conv2d_2_pointwise': [batch_size, 56, 56, 128],\n 'Conv2d_3_depthwise': [batch_size, 56, 56, 128],\n 'Conv2d_3_pointwise': [batch_size, 56, 56, 128],\n 'Conv2d_4_depthwise': [batch_size, 28, 28, 128],\n 'Conv2d_4_pointwise': [batch_size, 28, 28, 256],\n 'Conv2d_5_depthwise': [batch_size, 28, 28, 256],\n 'Conv2d_5_pointwise': [batch_size, 28, 28, 256],\n 'Conv2d_6_depthwise': [batch_size, 14, 14, 256],\n 'Conv2d_6_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_7_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_7_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_8_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_8_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_9_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_9_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_10_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_10_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_11_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_11_pointwise': [batch_size, 14, 14, 512],\n 'Conv2d_12_depthwise': [batch_size, 14, 14, 512],\n 'Conv2d_12_pointwise': [batch_size, 14, 14, 1024],\n 'Conv2d_13_depthwise': [batch_size, 14, 14, 1024],\n 'Conv2d_13_pointwise': [batch_size, 14, 14, 1024]}\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name, expected_shape in endpoints_shapes.iteritems():\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testOutputStride8BuildAndCheckAllEndPointsUptoConv2d_13(self):\n batch_size = 5\n height, width = 224, 224\n output_stride = 8\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope([slim.conv2d, slim.separable_conv2d],\n normalizer_fn=slim.batch_norm):\n _, end_points = mobilenet_v1.mobilenet_v1_base(\n inputs, output_stride=output_stride,\n final_endpoint='Conv2d_13_pointwise')\n endpoints_shapes = {'Conv2d_0': [batch_size, 112, 112, 32],\n 'Conv2d_1_depthwise': [batch_size, 112, 112, 32],\n 'Conv2d_1_pointwise': [batch_size, 112, 112, 64],\n 'Conv2d_2_depthwise': [batch_size, 56, 56, 64],\n 'Conv2d_2_pointwise': [batch_size, 56, 56, 128],\n 'Conv2d_3_depthwise': [batch_size, 56, 56, 128],\n 'Conv2d_3_pointwise': [batch_size, 56, 56, 128],\n 'Conv2d_4_depthwise': [batch_size, 28, 28, 128],\n 'Conv2d_4_pointwise': [batch_size, 28, 28, 256],\n 'Conv2d_5_depthwise': [batch_size, 28, 28, 256],\n 'Conv2d_5_pointwise': [batch_size, 28, 28, 256],\n 'Conv2d_6_depthwise': [batch_size, 28, 28, 256],\n 'Conv2d_6_pointwise': [batch_size, 28, 28, 512],\n 'Conv2d_7_depthwise': [batch_size, 28, 28, 512],\n 'Conv2d_7_pointwise': [batch_size, 28, 28, 512],\n 'Conv2d_8_depthwise': [batch_size, 28, 28, 512],\n 'Conv2d_8_pointwise': [batch_size, 28, 28, 512],\n 'Conv2d_9_depthwise': [batch_size, 28, 28, 512],\n 'Conv2d_9_pointwise': [batch_size, 28, 28, 512],\n 'Conv2d_10_depthwise': [batch_size, 28, 28, 512],\n 'Conv2d_10_pointwise': [batch_size, 28, 28, 512],\n 'Conv2d_11_depthwise': [batch_size, 28, 28, 512],\n 'Conv2d_11_pointwise': [batch_size, 28, 28, 512],\n 'Conv2d_12_depthwise': [batch_size, 28, 28, 512],\n 'Conv2d_12_pointwise': [batch_size, 28, 28, 1024],\n 'Conv2d_13_depthwise': [batch_size, 28, 28, 1024],\n 'Conv2d_13_pointwise': [batch_size, 28, 28, 1024]}\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name, expected_shape in endpoints_shapes.iteritems():\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testBuildAndCheckAllEndPointsApproximateFaceNet(self):\n batch_size = 5\n height, width = 128, 128\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope([slim.conv2d, slim.separable_conv2d],\n normalizer_fn=slim.batch_norm):\n _, end_points = mobilenet_v1.mobilenet_v1_base(\n inputs, final_endpoint='Conv2d_13_pointwise', depth_multiplier=0.75)\n # For the Conv2d_0 layer FaceNet has depth=16\n endpoints_shapes = {'Conv2d_0': [batch_size, 64, 64, 24],\n 'Conv2d_1_depthwise': [batch_size, 64, 64, 24],\n 'Conv2d_1_pointwise': [batch_size, 64, 64, 48],\n 'Conv2d_2_depthwise': [batch_size, 32, 32, 48],\n 'Conv2d_2_pointwise': [batch_size, 32, 32, 96],\n 'Conv2d_3_depthwise': [batch_size, 32, 32, 96],\n 'Conv2d_3_pointwise': [batch_size, 32, 32, 96],\n 'Conv2d_4_depthwise': [batch_size, 16, 16, 96],\n 'Conv2d_4_pointwise': [batch_size, 16, 16, 192],\n 'Conv2d_5_depthwise': [batch_size, 16, 16, 192],\n 'Conv2d_5_pointwise': [batch_size, 16, 16, 192],\n 'Conv2d_6_depthwise': [batch_size, 8, 8, 192],\n 'Conv2d_6_pointwise': [batch_size, 8, 8, 384],\n 'Conv2d_7_depthwise': [batch_size, 8, 8, 384],\n 'Conv2d_7_pointwise': [batch_size, 8, 8, 384],\n 'Conv2d_8_depthwise': [batch_size, 8, 8, 384],\n 'Conv2d_8_pointwise': [batch_size, 8, 8, 384],\n 'Conv2d_9_depthwise': [batch_size, 8, 8, 384],\n 'Conv2d_9_pointwise': [batch_size, 8, 8, 384],\n 'Conv2d_10_depthwise': [batch_size, 8, 8, 384],\n 'Conv2d_10_pointwise': [batch_size, 8, 8, 384],\n 'Conv2d_11_depthwise': [batch_size, 8, 8, 384],\n 'Conv2d_11_pointwise': [batch_size, 8, 8, 384],\n 'Conv2d_12_depthwise': [batch_size, 4, 4, 384],\n 'Conv2d_12_pointwise': [batch_size, 4, 4, 768],\n 'Conv2d_13_depthwise': [batch_size, 4, 4, 768],\n 'Conv2d_13_pointwise': [batch_size, 4, 4, 768]}\n self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())\n for endpoint_name, expected_shape in endpoints_shapes.iteritems():\n self.assertTrue(endpoint_name in end_points)\n self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),\n expected_shape)\n\n def testModelHasExpectedNumberOfParameters(self):\n batch_size = 5\n height, width = 224, 224\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with slim.arg_scope([slim.conv2d, slim.separable_conv2d],\n normalizer_fn=slim.batch_norm):\n mobilenet_v1.mobilenet_v1_base(inputs)\n total_params, _ = slim.model_analyzer.analyze_vars(\n slim.get_model_variables())\n self.assertAlmostEqual(3217920L, total_params)\n\n def testBuildEndPointsWithDepthMultiplierLessThanOne(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes)\n\n endpoint_keys = [key for key in end_points.keys() if key.startswith('Conv')]\n\n _, end_points_with_multiplier = mobilenet_v1.mobilenet_v1(\n inputs, num_classes, scope='depth_multiplied_net',\n depth_multiplier=0.5)\n\n for key in endpoint_keys:\n original_depth = end_points[key].get_shape().as_list()[3]\n new_depth = end_points_with_multiplier[key].get_shape().as_list()[3]\n self.assertEqual(0.5 * original_depth, new_depth)\n\n def testBuildEndPointsWithDepthMultiplierGreaterThanOne(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes)\n\n endpoint_keys = [key for key in end_points.keys()\n if key.startswith('Mixed') or key.startswith('Conv')]\n\n _, end_points_with_multiplier = mobilenet_v1.mobilenet_v1(\n inputs, num_classes, scope='depth_multiplied_net',\n depth_multiplier=2.0)\n\n for key in endpoint_keys:\n original_depth = end_points[key].get_shape().as_list()[3]\n new_depth = end_points_with_multiplier[key].get_shape().as_list()[3]\n self.assertEqual(2.0 * original_depth, new_depth)\n\n def testRaiseValueErrorWithInvalidDepthMultiplier(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n with self.assertRaises(ValueError):\n _ = mobilenet_v1.mobilenet_v1(\n inputs, num_classes, depth_multiplier=-0.1)\n with self.assertRaises(ValueError):\n _ = mobilenet_v1.mobilenet_v1(\n inputs, num_classes, depth_multiplier=0.0)\n\n def testHalfSizeImages(self):\n batch_size = 5\n height, width = 112, 112\n num_classes = 1000\n\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Conv2d_13_pointwise']\n self.assertListEqual(pre_pool.get_shape().as_list(),\n [batch_size, 4, 4, 1024])\n\n def testUnknownImageShape(self):\n tf.reset_default_graph()\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n input_np = np.random.uniform(0, 1, (batch_size, height, width, 3))\n with self.test_session() as sess:\n inputs = tf.placeholder(tf.float32, shape=(batch_size, None, None, 3))\n logits, end_points = mobilenet_v1.mobilenet_v1(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n pre_pool = end_points['Conv2d_13_pointwise']\n feed_dict = {inputs: input_np}\n tf.global_variables_initializer().run()\n pre_pool_out = sess.run(pre_pool, feed_dict=feed_dict)\n self.assertListEqual(list(pre_pool_out.shape), [batch_size, 7, 7, 1024])\n\n def testUnknowBatchSize(self):\n batch_size = 1\n height, width = 224, 224\n num_classes = 1000\n\n inputs = tf.placeholder(tf.float32, (None, height, width, 3))\n logits, _ = mobilenet_v1.mobilenet_v1(inputs, num_classes)\n self.assertTrue(logits.op.name.startswith('MobilenetV1/Logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, num_classes])\n images = tf.random_uniform((batch_size, height, width, 3))\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEquals(output.shape, (batch_size, num_classes))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = mobilenet_v1.mobilenet_v1(eval_inputs, num_classes,\n is_training=False)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (batch_size,))\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 5\n eval_batch_size = 2\n height, width = 150, 150\n num_classes = 1000\n\n train_inputs = tf.random_uniform((train_batch_size, height, width, 3))\n mobilenet_v1.mobilenet_v1(train_inputs, num_classes)\n eval_inputs = tf.random_uniform((eval_batch_size, height, width, 3))\n logits, _ = mobilenet_v1.mobilenet_v1(eval_inputs, num_classes,\n reuse=True)\n predictions = tf.argmax(logits, 1)\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(predictions)\n self.assertEquals(output.shape, (eval_batch_size,))\n\n def testLogitsNotSqueezed(self):\n num_classes = 25\n images = tf.random_uniform([1, 224, 224, 3])\n logits, _ = mobilenet_v1.mobilenet_v1(images,\n num_classes=num_classes,\n spatial_squeeze=False)\n\n with self.test_session() as sess:\n tf.global_variables_initializer().run()\n logits_out = sess.run(logits)\n self.assertListEqual(list(logits_out.shape), [1, 1, 1, num_classes])\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/nets_factory.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains a factory for building various models.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\nimport functools\n\nimport tensorflow as tf\n\nfrom nets import alexnet\nfrom nets import cifarnet\nfrom nets import inception\nfrom nets import lenet\nfrom nets import mobilenet_v1\nfrom nets import overfeat\nfrom nets import resnet_v1\nfrom nets import resnet_v2\nfrom nets import vgg\n\nslim = tf.contrib.slim\n\nnetworks_map = {'alexnet_v2': alexnet.alexnet_v2,\n 'cifarnet': cifarnet.cifarnet,\n 'overfeat': overfeat.overfeat,\n 'vgg_a': vgg.vgg_a,\n 'vgg_16': vgg.vgg_16,\n 'vgg_19': vgg.vgg_19,\n 'inception_v1': inception.inception_v1,\n 'inception_v2': inception.inception_v2,\n 'inception_v3': inception.inception_v3,\n 'inception_v4': inception.inception_v4,\n 'inception_resnet_v2': inception.inception_resnet_v2,\n 'lenet': lenet.lenet,\n 'resnet_v1_50': resnet_v1.resnet_v1_50,\n 'resnet_v1_101': resnet_v1.resnet_v1_101,\n 'resnet_v1_152': resnet_v1.resnet_v1_152,\n 'resnet_v1_200': resnet_v1.resnet_v1_200,\n 'resnet_v2_50': resnet_v2.resnet_v2_50,\n 'resnet_v2_101': resnet_v2.resnet_v2_101,\n 'resnet_v2_152': resnet_v2.resnet_v2_152,\n 'resnet_v2_200': resnet_v2.resnet_v2_200,\n 'mobilenet_v1': mobilenet_v1.mobilenet_v1,\n 'mobilenet_v1_075': mobilenet_v1.mobilenet_v1_075,\n 'mobilenet_v1_050': mobilenet_v1.mobilenet_v1_050,\n 'mobilenet_v1_025': mobilenet_v1.mobilenet_v1_025,\n }\n\narg_scopes_map = {'alexnet_v2': alexnet.alexnet_v2_arg_scope,\n 'cifarnet': cifarnet.cifarnet_arg_scope,\n 'overfeat': overfeat.overfeat_arg_scope,\n 'vgg_a': vgg.vgg_arg_scope,\n 'vgg_16': vgg.vgg_arg_scope,\n 'vgg_19': vgg.vgg_arg_scope,\n 'inception_v1': inception.inception_v3_arg_scope,\n 'inception_v2': inception.inception_v3_arg_scope,\n 'inception_v3': inception.inception_v3_arg_scope,\n 'inception_v4': inception.inception_v4_arg_scope,\n 'inception_resnet_v2':\n inception.inception_resnet_v2_arg_scope,\n 'lenet': lenet.lenet_arg_scope,\n 'resnet_v1_50': resnet_v1.resnet_arg_scope,\n 'resnet_v1_101': resnet_v1.resnet_arg_scope,\n 'resnet_v1_152': resnet_v1.resnet_arg_scope,\n 'resnet_v1_200': resnet_v1.resnet_arg_scope,\n 'resnet_v2_50': resnet_v2.resnet_arg_scope,\n 'resnet_v2_101': resnet_v2.resnet_arg_scope,\n 'resnet_v2_152': resnet_v2.resnet_arg_scope,\n 'resnet_v2_200': resnet_v2.resnet_arg_scope,\n 'mobilenet_v1': mobilenet_v1.mobilenet_v1_arg_scope,\n 'mobilenet_v1_075': mobilenet_v1.mobilenet_v1_arg_scope,\n 'mobilenet_v1_050': mobilenet_v1.mobilenet_v1_arg_scope,\n 'mobilenet_v1_025': mobilenet_v1.mobilenet_v1_arg_scope,\n }\n\n\ndef get_network_fn(name, num_classes, weight_decay=0.0, is_training=False):\n \"\"\"Returns a network_fn such as `logits, end_points = network_fn(images)`.\n\n Args:\n name: The name of the network.\n num_classes: The number of classes to use for classification.\n weight_decay: The l2 coefficient for the model weights.\n is_training: `True` if the model is being used for training and `False`\n otherwise.\n\n Returns:\n network_fn: A function that applies the model to a batch of images. It has\n the following signature:\n logits, end_points = network_fn(images)\n Raises:\n ValueError: If network `name` is not recognized.\n \"\"\"\n if name not in networks_map:\n raise ValueError('Name of network unknown %s' % name)\n func = networks_map[name]\n @functools.wraps(func)\n def network_fn(images):\n arg_scope = arg_scopes_map[name](weight_decay=weight_decay)\n with slim.arg_scope(arg_scope):\n return func(images, num_classes, is_training=is_training)\n if hasattr(func, 'default_image_size'):\n network_fn.default_image_size = func.default_image_size\n\n return network_fn\n"
},
{
"path": "nets/nets_factory_test.py",
"content": "# Copyright 2016 Google Inc. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\n\"\"\"Tests for slim.inception.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import nets_factory\n\nslim = tf.contrib.slim\n\n\nclass NetworksTest(tf.test.TestCase):\n\n def testGetNetworkFn(self):\n batch_size = 5\n num_classes = 1000\n for net in nets_factory.networks_map:\n with self.test_session():\n net_fn = nets_factory.get_network_fn(net, num_classes)\n # Most networks use 224 as their default_image_size\n image_size = getattr(net_fn, 'default_image_size', 224)\n inputs = tf.random_uniform((batch_size, image_size, image_size, 3))\n logits, end_points = net_fn(inputs)\n self.assertTrue(isinstance(logits, tf.Tensor))\n self.assertTrue(isinstance(end_points, dict))\n self.assertEqual(logits.get_shape().as_list()[0], batch_size)\n self.assertEqual(logits.get_shape().as_list()[-1], num_classes)\n\n def testGetNetworkFnArgScope(self):\n batch_size = 5\n num_classes = 10\n net = 'cifarnet'\n with self.test_session(use_gpu=True):\n net_fn = nets_factory.get_network_fn(net, num_classes)\n image_size = getattr(net_fn, 'default_image_size', 224)\n with slim.arg_scope([slim.model_variable, slim.variable],\n device='/CPU:0'):\n inputs = tf.random_uniform((batch_size, image_size, image_size, 3))\n net_fn(inputs)\n weights = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, 'CifarNet/conv1')[0]\n self.assertDeviceEqual('/CPU:0', weights.device)\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/overfeat.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains the model definition for the OverFeat network.\n\nThe definition for the network was obtained from:\n OverFeat: Integrated Recognition, Localization and Detection using\n Convolutional Networks\n Pierre Sermanet, David Eigen, Xiang Zhang, Michael Mathieu, Rob Fergus and\n Yann LeCun, 2014\n http://arxiv.org/abs/1312.6229\n\nUsage:\n with slim.arg_scope(overfeat.overfeat_arg_scope()):\n outputs, end_points = overfeat.overfeat(inputs)\n\n@@overfeat\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\ntrunc_normal = lambda stddev: tf.truncated_normal_initializer(0.0, stddev)\n\n\ndef overfeat_arg_scope(weight_decay=0.0005):\n with slim.arg_scope([slim.conv2d, slim.fully_connected],\n activation_fn=tf.nn.relu,\n weights_regularizer=slim.l2_regularizer(weight_decay),\n biases_initializer=tf.zeros_initializer()):\n with slim.arg_scope([slim.conv2d], padding='SAME'):\n with slim.arg_scope([slim.max_pool2d], padding='VALID') as arg_sc:\n return arg_sc\n\n\ndef overfeat(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.5,\n spatial_squeeze=True,\n scope='overfeat'):\n \"\"\"Contains the model definition for the OverFeat network.\n\n The definition for the network was obtained from:\n OverFeat: Integrated Recognition, Localization and Detection using\n Convolutional Networks\n Pierre Sermanet, David Eigen, Xiang Zhang, Michael Mathieu, Rob Fergus and\n Yann LeCun, 2014\n http://arxiv.org/abs/1312.6229\n\n Note: All the fully_connected layers have been transformed to conv2d layers.\n To use in classification mode, resize input to 231x231. To use in fully\n convolutional mode, set spatial_squeeze to false.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether or not the model is being trained.\n dropout_keep_prob: the probability that activations are kept in the dropout\n layers during training.\n spatial_squeeze: whether or not should squeeze the spatial dimensions of the\n outputs. Useful to remove unnecessary dimensions for classification.\n scope: Optional scope for the variables.\n\n Returns:\n the last op containing the log predictions and end_points dict.\n\n \"\"\"\n with tf.variable_scope(scope, 'overfeat', [inputs]) as sc:\n end_points_collection = sc.name + '_end_points'\n # Collect outputs for conv2d, fully_connected and max_pool2d\n with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.max_pool2d],\n outputs_collections=end_points_collection):\n net = slim.conv2d(inputs, 64, [11, 11], 4, padding='VALID',\n scope='conv1')\n net = slim.max_pool2d(net, [2, 2], scope='pool1')\n net = slim.conv2d(net, 256, [5, 5], padding='VALID', scope='conv2')\n net = slim.max_pool2d(net, [2, 2], scope='pool2')\n net = slim.conv2d(net, 512, [3, 3], scope='conv3')\n net = slim.conv2d(net, 1024, [3, 3], scope='conv4')\n net = slim.conv2d(net, 1024, [3, 3], scope='conv5')\n net = slim.max_pool2d(net, [2, 2], scope='pool5')\n with slim.arg_scope([slim.conv2d],\n weights_initializer=trunc_normal(0.005),\n biases_initializer=tf.constant_initializer(0.1)):\n # Use conv2d instead of fully_connected layers.\n net = slim.conv2d(net, 3072, [6, 6], padding='VALID', scope='fc6')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout6')\n net = slim.conv2d(net, 4096, [1, 1], scope='fc7')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout7')\n net = slim.conv2d(net, num_classes, [1, 1],\n activation_fn=None,\n normalizer_fn=None,\n biases_initializer=tf.zeros_initializer(),\n scope='fc8')\n # Convert end_points_collection into a end_point dict.\n end_points = slim.utils.convert_collection_to_dict(end_points_collection)\n if spatial_squeeze:\n net = tf.squeeze(net, [1, 2], name='fc8/squeezed')\n end_points[sc.name + '/fc8'] = net\n return net, end_points\noverfeat.default_image_size = 231\n"
},
{
"path": "nets/overfeat_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for slim.nets.overfeat.\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import overfeat\n\nslim = tf.contrib.slim\n\n\nclass OverFeatTest(tf.test.TestCase):\n\n def testBuild(self):\n batch_size = 5\n height, width = 231, 231\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = overfeat.overfeat(inputs, num_classes)\n self.assertEquals(logits.op.name, 'overfeat/fc8/squeezed')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testFullyConvolutional(self):\n batch_size = 1\n height, width = 281, 281\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = overfeat.overfeat(inputs, num_classes, spatial_squeeze=False)\n self.assertEquals(logits.op.name, 'overfeat/fc8/BiasAdd')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, 2, 2, num_classes])\n\n def testEndPoints(self):\n batch_size = 5\n height, width = 231, 231\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = overfeat.overfeat(inputs, num_classes)\n expected_names = ['overfeat/conv1',\n 'overfeat/pool1',\n 'overfeat/conv2',\n 'overfeat/pool2',\n 'overfeat/conv3',\n 'overfeat/conv4',\n 'overfeat/conv5',\n 'overfeat/pool5',\n 'overfeat/fc6',\n 'overfeat/fc7',\n 'overfeat/fc8'\n ]\n self.assertSetEqual(set(end_points.keys()), set(expected_names))\n\n def testModelVariables(self):\n batch_size = 5\n height, width = 231, 231\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n overfeat.overfeat(inputs, num_classes)\n expected_names = ['overfeat/conv1/weights',\n 'overfeat/conv1/biases',\n 'overfeat/conv2/weights',\n 'overfeat/conv2/biases',\n 'overfeat/conv3/weights',\n 'overfeat/conv3/biases',\n 'overfeat/conv4/weights',\n 'overfeat/conv4/biases',\n 'overfeat/conv5/weights',\n 'overfeat/conv5/biases',\n 'overfeat/fc6/weights',\n 'overfeat/fc6/biases',\n 'overfeat/fc7/weights',\n 'overfeat/fc7/biases',\n 'overfeat/fc8/weights',\n 'overfeat/fc8/biases',\n ]\n model_variables = [v.op.name for v in slim.get_model_variables()]\n self.assertSetEqual(set(model_variables), set(expected_names))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 231, 231\n num_classes = 1000\n with self.test_session():\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = overfeat.overfeat(eval_inputs, is_training=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n predictions = tf.argmax(logits, 1)\n self.assertListEqual(predictions.get_shape().as_list(), [batch_size])\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 2\n eval_batch_size = 1\n train_height, train_width = 231, 231\n eval_height, eval_width = 281, 281\n num_classes = 1000\n with self.test_session():\n train_inputs = tf.random_uniform(\n (train_batch_size, train_height, train_width, 3))\n logits, _ = overfeat.overfeat(train_inputs)\n self.assertListEqual(logits.get_shape().as_list(),\n [train_batch_size, num_classes])\n tf.get_variable_scope().reuse_variables()\n eval_inputs = tf.random_uniform(\n (eval_batch_size, eval_height, eval_width, 3))\n logits, _ = overfeat.overfeat(eval_inputs, is_training=False,\n spatial_squeeze=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [eval_batch_size, 2, 2, num_classes])\n logits = tf.reduce_mean(logits, [1, 2])\n predictions = tf.argmax(logits, 1)\n self.assertEquals(predictions.get_shape().as_list(), [eval_batch_size])\n\n def testForward(self):\n batch_size = 1\n height, width = 231, 231\n with self.test_session() as sess:\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = overfeat.overfeat(inputs)\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits)\n self.assertTrue(output.any())\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/resnet_utils.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains building blocks for various versions of Residual Networks.\n\nResidual networks (ResNets) were proposed in:\n Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun\n Deep Residual Learning for Image Recognition. arXiv:1512.03385, 2015\n\nMore variants were introduced in:\n Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun\n Identity Mappings in Deep Residual Networks. arXiv: 1603.05027, 2016\n\nWe can obtain different ResNet variants by changing the network depth, width,\nand form of residual unit. This module implements the infrastructure for\nbuilding them. Concrete ResNet units and full ResNet networks are implemented in\nthe accompanying resnet_v1.py and resnet_v2.py modules.\n\nCompared to https://github.com/KaimingHe/deep-residual-networks, in the current\nimplementation we subsample the output activations in the last residual unit of\neach block, instead of subsampling the input activations in the first residual\nunit of each block. The two implementations give identical results but our\nimplementation is more memory efficient.\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport collections\nimport tensorflow as tf\n\nslim = tf.contrib.slim\n\n\nclass Block(collections.namedtuple('Block', ['scope', 'unit_fn', 'args'])):\n \"\"\"A named tuple describing a ResNet block.\n\n Its parts are:\n scope: The scope of the `Block`.\n unit_fn: The ResNet unit function which takes as input a `Tensor` and\n returns another `Tensor` with the output of the ResNet unit.\n args: A list of length equal to the number of units in the `Block`. The list\n contains one (depth, depth_bottleneck, stride) tuple for each unit in the\n block to serve as argument to unit_fn.\n \"\"\"\n\n\ndef subsample(inputs, factor, scope=None):\n \"\"\"Subsamples the input along the spatial dimensions.\n\n Args:\n inputs: A `Tensor` of size [batch, height_in, width_in, channels].\n factor: The subsampling factor.\n scope: Optional variable_scope.\n\n Returns:\n output: A `Tensor` of size [batch, height_out, width_out, channels] with the\n input, either intact (if factor == 1) or subsampled (if factor > 1).\n \"\"\"\n if factor == 1:\n return inputs\n else:\n return slim.max_pool2d(inputs, [1, 1], stride=factor, scope=scope)\n\n\ndef conv2d_same(inputs, num_outputs, kernel_size, stride, rate=1, scope=None):\n \"\"\"Strided 2-D convolution with 'SAME' padding.\n\n When stride > 1, then we do explicit zero-padding, followed by conv2d with\n 'VALID' padding.\n\n Note that\n\n net = conv2d_same(inputs, num_outputs, 3, stride=stride)\n\n is equivalent to\n\n net = slim.conv2d(inputs, num_outputs, 3, stride=1, padding='SAME')\n net = subsample(net, factor=stride)\n\n whereas\n\n net = slim.conv2d(inputs, num_outputs, 3, stride=stride, padding='SAME')\n\n is different when the input's height or width is even, which is why we add the\n current function. For more details, see ResnetUtilsTest.testConv2DSameEven().\n\n Args:\n inputs: A 4-D tensor of size [batch, height_in, width_in, channels].\n num_outputs: An integer, the number of output filters.\n kernel_size: An int with the kernel_size of the filters.\n stride: An integer, the output stride.\n rate: An integer, rate for atrous convolution.\n scope: Scope.\n\n Returns:\n output: A 4-D tensor of size [batch, height_out, width_out, channels] with\n the convolution output.\n \"\"\"\n if stride == 1:\n return slim.conv2d(inputs, num_outputs, kernel_size, stride=1, rate=rate,\n padding='SAME', scope=scope)\n else:\n kernel_size_effective = kernel_size + (kernel_size - 1) * (rate - 1)\n pad_total = kernel_size_effective - 1\n pad_beg = pad_total // 2\n pad_end = pad_total - pad_beg\n inputs = tf.pad(inputs,\n [[0, 0], [pad_beg, pad_end], [pad_beg, pad_end], [0, 0]])\n return slim.conv2d(inputs, num_outputs, kernel_size, stride=stride,\n rate=rate, padding='VALID', scope=scope)\n\n\n@slim.add_arg_scope\ndef stack_blocks_dense(net, blocks, output_stride=None,\n outputs_collections=None):\n \"\"\"Stacks ResNet `Blocks` and controls output feature density.\n\n First, this function creates scopes for the ResNet in the form of\n 'block_name/unit_1', 'block_name/unit_2', etc.\n\n Second, this function allows the user to explicitly control the ResNet\n output_stride, which is the ratio of the input to output spatial resolution.\n This is useful for dense prediction tasks such as semantic segmentation or\n object detection.\n\n Most ResNets consist of 4 ResNet blocks and subsample the activations by a\n factor of 2 when transitioning between consecutive ResNet blocks. This results\n to a nominal ResNet output_stride equal to 8. If we set the output_stride to\n half the nominal network stride (e.g., output_stride=4), then we compute\n responses twice.\n\n Control of the output feature density is implemented by atrous convolution.\n\n Args:\n net: A `Tensor` of size [batch, height, width, channels].\n blocks: A list of length equal to the number of ResNet `Blocks`. Each\n element is a ResNet `Block` object describing the units in the `Block`.\n output_stride: If `None`, then the output will be computed at the nominal\n network stride. If output_stride is not `None`, it specifies the requested\n ratio of input to output spatial resolution, which needs to be equal to\n the product of unit strides from the start up to some level of the ResNet.\n For example, if the ResNet employs units with strides 1, 2, 1, 3, 4, 1,\n then valid values for the output_stride are 1, 2, 6, 24 or None (which\n is equivalent to output_stride=24).\n outputs_collections: Collection to add the ResNet block outputs.\n\n Returns:\n net: Output tensor with stride equal to the specified output_stride.\n\n Raises:\n ValueError: If the target output_stride is not valid.\n \"\"\"\n # The current_stride variable keeps track of the effective stride of the\n # activations. This allows us to invoke atrous convolution whenever applying\n # the next residual unit would result in the activations having stride larger\n # than the target output_stride.\n current_stride = 1\n\n # The atrous convolution rate parameter.\n rate = 1\n\n for block in blocks:\n with tf.variable_scope(block.scope, 'block', [net]) as sc:\n for i, unit in enumerate(block.args):\n if output_stride is not None and current_stride > output_stride:\n raise ValueError('The target output_stride cannot be reached.')\n\n with tf.variable_scope('unit_%d' % (i + 1), values=[net]):\n # If we have reached the target output_stride, then we need to employ\n # atrous convolution with stride=1 and multiply the atrous rate by the\n # current unit's stride for use in subsequent layers.\n if output_stride is not None and current_stride == output_stride:\n net = block.unit_fn(net, rate=rate, **dict(unit, stride=1))\n rate *= unit.get('stride', 1)\n\n else:\n net = block.unit_fn(net, rate=1, **unit)\n current_stride *= unit.get('stride', 1)\n net = slim.utils.collect_named_outputs(outputs_collections, sc.name, net)\n\n if output_stride is not None and current_stride != output_stride:\n raise ValueError('The target output_stride cannot be reached.')\n\n return net\n\n\ndef resnet_arg_scope(weight_decay=0.0001,\n batch_norm_decay=0.997,\n batch_norm_epsilon=1e-5,\n batch_norm_scale=True,\n activation_fn=tf.nn.relu,\n use_batch_norm=True):\n \"\"\"Defines the default ResNet arg scope.\n\n TODO(gpapan): The batch-normalization related default values above are\n appropriate for use in conjunction with the reference ResNet models\n released at https://github.com/KaimingHe/deep-residual-networks. When\n training ResNets from scratch, they might need to be tuned.\n\n Args:\n weight_decay: The weight decay to use for regularizing the model.\n batch_norm_decay: The moving average decay when estimating layer activation\n statistics in batch normalization.\n batch_norm_epsilon: Small constant to prevent division by zero when\n normalizing activations by their variance in batch normalization.\n batch_norm_scale: If True, uses an explicit `gamma` multiplier to scale the\n activations in the batch normalization layer.\n activation_fn: The activation function which is used in ResNet.\n use_batch_norm: Whether or not to use batch normalization.\n\n Returns:\n An `arg_scope` to use for the resnet models.\n \"\"\"\n batch_norm_params = {\n 'decay': batch_norm_decay,\n 'epsilon': batch_norm_epsilon,\n 'scale': batch_norm_scale,\n 'updates_collections': tf.GraphKeys.UPDATE_OPS,\n }\n\n with slim.arg_scope(\n [slim.conv2d],\n weights_regularizer=slim.l2_regularizer(weight_decay),\n weights_initializer=slim.variance_scaling_initializer(),\n activation_fn=activation_fn,\n normalizer_fn=slim.batch_norm if use_batch_norm else None,\n normalizer_params=batch_norm_params):\n with slim.arg_scope([slim.batch_norm], **batch_norm_params):\n # The following implies padding='SAME' for pool1, which makes feature\n # alignment easier for dense prediction tasks. This is also used in\n # https://github.com/facebook/fb.resnet.torch. However the accompanying\n # code of 'Deep Residual Learning for Image Recognition' uses\n # padding='VALID' for pool1. You can switch to that choice by setting\n # slim.arg_scope([slim.max_pool2d], padding='VALID').\n with slim.arg_scope([slim.max_pool2d], padding='SAME') as arg_sc:\n return arg_sc\n"
},
{
"path": "nets/resnet_v1.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains definitions for the original form of Residual Networks.\n\nThe 'v1' residual networks (ResNets) implemented in this module were proposed\nby:\n[1] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun\n Deep Residual Learning for Image Recognition. arXiv:1512.03385\n\nOther variants were introduced in:\n[2] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun\n Identity Mappings in Deep Residual Networks. arXiv: 1603.05027\n\nThe networks defined in this module utilize the bottleneck building block of\n[1] with projection shortcuts only for increasing depths. They employ batch\nnormalization *after* every weight layer. This is the architecture used by\nMSRA in the Imagenet and MSCOCO 2016 competition models ResNet-101 and\nResNet-152. See [2; Fig. 1a] for a comparison between the current 'v1'\narchitecture and the alternative 'v2' architecture of [2] which uses batch\nnormalization *before* every weight layer in the so-called full pre-activation\nunits.\n\nTypical use:\n\n from tensorflow.contrib.slim.nets import resnet_v1\n\nResNet-101 for image classification into 1000 classes:\n\n # inputs has shape [batch, 224, 224, 3]\n with slim.arg_scope(resnet_v1.resnet_arg_scope()):\n net, end_points = resnet_v1.resnet_v1_101(inputs, 1000, is_training=False)\n\nResNet-101 for semantic segmentation into 21 classes:\n\n # inputs has shape [batch, 513, 513, 3]\n with slim.arg_scope(resnet_v1.resnet_arg_scope()):\n net, end_points = resnet_v1.resnet_v1_101(inputs,\n 21,\n is_training=False,\n global_pool=False,\n output_stride=16)\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import resnet_utils\n\n\nresnet_arg_scope = resnet_utils.resnet_arg_scope\nslim = tf.contrib.slim\n\n\n@slim.add_arg_scope\ndef bottleneck(inputs,\n depth,\n depth_bottleneck,\n stride,\n rate=1,\n outputs_collections=None,\n scope=None,\n use_bounded_activations=False):\n \"\"\"Bottleneck residual unit variant with BN after convolutions.\n\n This is the original residual unit proposed in [1]. See Fig. 1(a) of [2] for\n its definition. Note that we use here the bottleneck variant which has an\n extra bottleneck layer.\n\n When putting together two consecutive ResNet blocks that use this unit, one\n should use stride = 2 in the last unit of the first block.\n\n Args:\n inputs: A tensor of size [batch, height, width, channels].\n depth: The depth of the ResNet unit output.\n depth_bottleneck: The depth of the bottleneck layers.\n stride: The ResNet unit's stride. Determines the amount of downsampling of\n the units output compared to its input.\n rate: An integer, rate for atrous convolution.\n outputs_collections: Collection to add the ResNet unit output.\n scope: Optional variable_scope.\n use_bounded_activations: Whether or not to use bounded activations. Bounded\n activations better lend themselves to quantized inference.\n\n Returns:\n The ResNet unit's output.\n \"\"\"\n with tf.variable_scope(scope, 'bottleneck_v1', [inputs]) as sc:\n depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)\n if depth == depth_in:\n shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')\n else:\n shortcut = slim.conv2d(\n inputs,\n depth, [1, 1],\n stride=stride,\n activation_fn=tf.nn.relu6 if use_bounded_activations else None,\n scope='shortcut')\n\n residual = slim.conv2d(inputs, depth_bottleneck, [1, 1], stride=1,\n scope='conv1')\n residual = resnet_utils.conv2d_same(residual, depth_bottleneck, 3, stride,\n rate=rate, scope='conv2')\n residual = slim.conv2d(residual, depth, [1, 1], stride=1,\n activation_fn=None, scope='conv3')\n\n if use_bounded_activations:\n # Use clip_by_value to simulate bandpass activation.\n residual = tf.clip_by_value(residual, -6.0, 6.0)\n output = tf.nn.relu6(shortcut + residual)\n else:\n output = tf.nn.relu(shortcut + residual)\n\n return slim.utils.collect_named_outputs(outputs_collections,\n sc.original_name_scope,\n output)\n\n\ndef resnet_v1(inputs,\n blocks,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n include_root_block=True,\n spatial_squeeze=True,\n reuse=None,\n scope=None):\n \"\"\"Generator for v1 ResNet models.\n\n This function generates a family of ResNet v1 models. See the resnet_v1_*()\n methods for specific model instantiations, obtained by selecting different\n block instantiations that produce ResNets of various depths.\n\n Training for image classification on Imagenet is usually done with [224, 224]\n inputs, resulting in [7, 7] feature maps at the output of the last ResNet\n block for the ResNets defined in [1] that have nominal stride equal to 32.\n However, for dense prediction tasks we advise that one uses inputs with\n spatial dimensions that are multiples of 32 plus 1, e.g., [321, 321]. In\n this case the feature maps at the ResNet output will have spatial shape\n [(height - 1) / output_stride + 1, (width - 1) / output_stride + 1]\n and corners exactly aligned with the input image corners, which greatly\n facilitates alignment of the features to the image. Using as input [225, 225]\n images results in [8, 8] feature maps at the output of the last ResNet block.\n\n For dense prediction tasks, the ResNet needs to run in fully-convolutional\n (FCN) mode and global_pool needs to be set to False. The ResNets in [1, 2] all\n have nominal stride equal to 32 and a good choice in FCN mode is to use\n output_stride=16 in order to increase the density of the computed features at\n small computational and memory overhead, cf. http://arxiv.org/abs/1606.00915.\n\n Args:\n inputs: A tensor of size [batch, height_in, width_in, channels].\n blocks: A list of length equal to the number of ResNet blocks. Each element\n is a resnet_utils.Block object describing the units in the block.\n num_classes: Number of predicted classes for classification tasks. If None\n we return the features before the logit layer.\n is_training: whether is training or not.\n global_pool: If True, we perform global average pooling before computing the\n logits. Set to True for image classification, False for dense prediction.\n output_stride: If None, then the output will be computed at the nominal\n network stride. If output_stride is not None, it specifies the requested\n ratio of input to output spatial resolution.\n include_root_block: If True, include the initial convolution followed by\n max-pooling, if False excludes it.\n spatial_squeeze: if True, logits is of shape [B, C], if false logits is\n of shape [B, 1, 1, C], where B is batch_size and C is number of classes.\n To use this parameter, the input images must be smaller than 300x300\n pixels, in which case the output logit layer does not contain spatial\n information and can be removed.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n scope: Optional variable_scope.\n\n Returns:\n net: A rank-4 tensor of size [batch, height_out, width_out, channels_out].\n If global_pool is False, then height_out and width_out are reduced by a\n factor of output_stride compared to the respective height_in and width_in,\n else both height_out and width_out equal one. If num_classes is None, then\n net is the output of the last ResNet block, potentially after global\n average pooling. If num_classes is not None, net contains the pre-softmax\n activations.\n end_points: A dictionary from components of the network to the corresponding\n activation.\n\n Raises:\n ValueError: If the target output_stride is not valid.\n \"\"\"\n with tf.variable_scope(scope, 'resnet_v1', [inputs], reuse=reuse) as sc:\n end_points_collection = sc.name + '_end_points'\n with slim.arg_scope([slim.conv2d, bottleneck,\n resnet_utils.stack_blocks_dense],\n outputs_collections=end_points_collection):\n with slim.arg_scope([slim.batch_norm], is_training=is_training):\n net = inputs\n if include_root_block:\n if output_stride is not None:\n if output_stride % 4 != 0:\n raise ValueError('The output_stride needs to be a multiple of 4.')\n output_stride /= 4\n net = resnet_utils.conv2d_same(net, 64, 7, stride=2, scope='conv1')\n net = slim.max_pool2d(net, [3, 3], stride=2, scope='pool1')\n net = resnet_utils.stack_blocks_dense(net, blocks, output_stride)\n if global_pool:\n # Global average pooling.\n net = tf.reduce_mean(net, [1, 2], name='pool5', keep_dims=True)\n if num_classes is not None:\n net = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,\n normalizer_fn=None, scope='logits')\n if spatial_squeeze:\n net = tf.squeeze(net, [1, 2], name='SpatialSqueeze')\n # Convert end_points_collection into a dictionary of end_points.\n end_points = slim.utils.convert_collection_to_dict(\n end_points_collection)\n if num_classes is not None:\n end_points['predictions'] = slim.softmax(net, scope='predictions')\n return net, end_points\nresnet_v1.default_image_size = 224\n\n\ndef resnet_v1_block(scope, base_depth, num_units, stride):\n \"\"\"Helper function for creating a resnet_v1 bottleneck block.\n\n Args:\n scope: The scope of the block.\n base_depth: The depth of the bottleneck layer for each unit.\n num_units: The number of units in the block.\n stride: The stride of the block, implemented as a stride in the last unit.\n All other units have stride=1.\n\n Returns:\n A resnet_v1 bottleneck block.\n \"\"\"\n return resnet_utils.Block(scope, bottleneck, [{\n 'depth': base_depth * 4,\n 'depth_bottleneck': base_depth,\n 'stride': 1\n }] * (num_units - 1) + [{\n 'depth': base_depth * 4,\n 'depth_bottleneck': base_depth,\n 'stride': stride\n }])\n\n\ndef resnet_v1_50(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v1_50'):\n \"\"\"ResNet-50 model of [1]. See resnet_v1() for arg and return description.\"\"\"\n blocks = [\n resnet_v1_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v1_block('block2', base_depth=128, num_units=4, stride=2),\n resnet_v1_block('block3', base_depth=256, num_units=6, stride=2),\n resnet_v1_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v1(inputs, blocks, num_classes, is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v1_50.default_image_size = resnet_v1.default_image_size\n\n\ndef resnet_v1_101(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v1_101'):\n \"\"\"ResNet-101 model of [1]. See resnet_v1() for arg and return description.\"\"\"\n blocks = [\n resnet_v1_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v1_block('block2', base_depth=128, num_units=4, stride=2),\n resnet_v1_block('block3', base_depth=256, num_units=23, stride=2),\n resnet_v1_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v1(inputs, blocks, num_classes, is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v1_101.default_image_size = resnet_v1.default_image_size\n\n\ndef resnet_v1_152(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v1_152'):\n \"\"\"ResNet-152 model of [1]. See resnet_v1() for arg and return description.\"\"\"\n blocks = [\n resnet_v1_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v1_block('block2', base_depth=128, num_units=8, stride=2),\n resnet_v1_block('block3', base_depth=256, num_units=36, stride=2),\n resnet_v1_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v1(inputs, blocks, num_classes, is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v1_152.default_image_size = resnet_v1.default_image_size\n\n\ndef resnet_v1_200(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v1_200'):\n \"\"\"ResNet-200 model of [2]. See resnet_v1() for arg and return description.\"\"\"\n blocks = [\n resnet_v1_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v1_block('block2', base_depth=128, num_units=24, stride=2),\n resnet_v1_block('block3', base_depth=256, num_units=36, stride=2),\n resnet_v1_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v1(inputs, blocks, num_classes, is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v1_200.default_image_size = resnet_v1.default_image_size\n"
},
{
"path": "nets/resnet_v1_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for slim.nets.resnet_v1.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport numpy as np\nimport tensorflow as tf\n\nfrom nets import resnet_utils\nfrom nets import resnet_v1\n\nslim = tf.contrib.slim\n\n\ndef create_test_input(batch_size, height, width, channels):\n \"\"\"Create test input tensor.\n\n Args:\n batch_size: The number of images per batch or `None` if unknown.\n height: The height of each image or `None` if unknown.\n width: The width of each image or `None` if unknown.\n channels: The number of channels per image or `None` if unknown.\n\n Returns:\n Either a placeholder `Tensor` of dimension\n [batch_size, height, width, channels] if any of the inputs are `None` or a\n constant `Tensor` with the mesh grid values along the spatial dimensions.\n \"\"\"\n if None in [batch_size, height, width, channels]:\n return tf.placeholder(tf.float32, (batch_size, height, width, channels))\n else:\n return tf.to_float(\n np.tile(\n np.reshape(\n np.reshape(np.arange(height), [height, 1]) +\n np.reshape(np.arange(width), [1, width]),\n [1, height, width, 1]),\n [batch_size, 1, 1, channels]))\n\n\nclass ResnetUtilsTest(tf.test.TestCase):\n\n def testSubsampleThreeByThree(self):\n x = tf.reshape(tf.to_float(tf.range(9)), [1, 3, 3, 1])\n x = resnet_utils.subsample(x, 2)\n expected = tf.reshape(tf.constant([0, 2, 6, 8]), [1, 2, 2, 1])\n with self.test_session():\n self.assertAllClose(x.eval(), expected.eval())\n\n def testSubsampleFourByFour(self):\n x = tf.reshape(tf.to_float(tf.range(16)), [1, 4, 4, 1])\n x = resnet_utils.subsample(x, 2)\n expected = tf.reshape(tf.constant([0, 2, 8, 10]), [1, 2, 2, 1])\n with self.test_session():\n self.assertAllClose(x.eval(), expected.eval())\n\n def testConv2DSameEven(self):\n n, n2 = 4, 2\n\n # Input image.\n x = create_test_input(1, n, n, 1)\n\n # Convolution kernel.\n w = create_test_input(1, 3, 3, 1)\n w = tf.reshape(w, [3, 3, 1, 1])\n\n tf.get_variable('Conv/weights', initializer=w)\n tf.get_variable('Conv/biases', initializer=tf.zeros([1]))\n tf.get_variable_scope().reuse_variables()\n\n y1 = slim.conv2d(x, 1, [3, 3], stride=1, scope='Conv')\n y1_expected = tf.to_float([[14, 28, 43, 26],\n [28, 48, 66, 37],\n [43, 66, 84, 46],\n [26, 37, 46, 22]])\n y1_expected = tf.reshape(y1_expected, [1, n, n, 1])\n\n y2 = resnet_utils.subsample(y1, 2)\n y2_expected = tf.to_float([[14, 43],\n [43, 84]])\n y2_expected = tf.reshape(y2_expected, [1, n2, n2, 1])\n\n y3 = resnet_utils.conv2d_same(x, 1, 3, stride=2, scope='Conv')\n y3_expected = y2_expected\n\n y4 = slim.conv2d(x, 1, [3, 3], stride=2, scope='Conv')\n y4_expected = tf.to_float([[48, 37],\n [37, 22]])\n y4_expected = tf.reshape(y4_expected, [1, n2, n2, 1])\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n self.assertAllClose(y1.eval(), y1_expected.eval())\n self.assertAllClose(y2.eval(), y2_expected.eval())\n self.assertAllClose(y3.eval(), y3_expected.eval())\n self.assertAllClose(y4.eval(), y4_expected.eval())\n\n def testConv2DSameOdd(self):\n n, n2 = 5, 3\n\n # Input image.\n x = create_test_input(1, n, n, 1)\n\n # Convolution kernel.\n w = create_test_input(1, 3, 3, 1)\n w = tf.reshape(w, [3, 3, 1, 1])\n\n tf.get_variable('Conv/weights', initializer=w)\n tf.get_variable('Conv/biases', initializer=tf.zeros([1]))\n tf.get_variable_scope().reuse_variables()\n\n y1 = slim.conv2d(x, 1, [3, 3], stride=1, scope='Conv')\n y1_expected = tf.to_float([[14, 28, 43, 58, 34],\n [28, 48, 66, 84, 46],\n [43, 66, 84, 102, 55],\n [58, 84, 102, 120, 64],\n [34, 46, 55, 64, 30]])\n y1_expected = tf.reshape(y1_expected, [1, n, n, 1])\n\n y2 = resnet_utils.subsample(y1, 2)\n y2_expected = tf.to_float([[14, 43, 34],\n [43, 84, 55],\n [34, 55, 30]])\n y2_expected = tf.reshape(y2_expected, [1, n2, n2, 1])\n\n y3 = resnet_utils.conv2d_same(x, 1, 3, stride=2, scope='Conv')\n y3_expected = y2_expected\n\n y4 = slim.conv2d(x, 1, [3, 3], stride=2, scope='Conv')\n y4_expected = y2_expected\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n self.assertAllClose(y1.eval(), y1_expected.eval())\n self.assertAllClose(y2.eval(), y2_expected.eval())\n self.assertAllClose(y3.eval(), y3_expected.eval())\n self.assertAllClose(y4.eval(), y4_expected.eval())\n\n def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):\n \"\"\"A plain ResNet without extra layers before or after the ResNet blocks.\"\"\"\n with tf.variable_scope(scope, values=[inputs]):\n with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):\n net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)\n end_points = slim.utils.convert_collection_to_dict('end_points')\n return net, end_points\n\n def testEndPointsV1(self):\n \"\"\"Test the end points of a tiny v1 bottleneck network.\"\"\"\n blocks = [\n resnet_v1.resnet_v1_block(\n 'block1', base_depth=1, num_units=2, stride=2),\n resnet_v1.resnet_v1_block(\n 'block2', base_depth=2, num_units=2, stride=1),\n ]\n inputs = create_test_input(2, 32, 16, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_plain(inputs, blocks, scope='tiny')\n expected = [\n 'tiny/block1/unit_1/bottleneck_v1/shortcut',\n 'tiny/block1/unit_1/bottleneck_v1/conv1',\n 'tiny/block1/unit_1/bottleneck_v1/conv2',\n 'tiny/block1/unit_1/bottleneck_v1/conv3',\n 'tiny/block1/unit_2/bottleneck_v1/conv1',\n 'tiny/block1/unit_2/bottleneck_v1/conv2',\n 'tiny/block1/unit_2/bottleneck_v1/conv3',\n 'tiny/block2/unit_1/bottleneck_v1/shortcut',\n 'tiny/block2/unit_1/bottleneck_v1/conv1',\n 'tiny/block2/unit_1/bottleneck_v1/conv2',\n 'tiny/block2/unit_1/bottleneck_v1/conv3',\n 'tiny/block2/unit_2/bottleneck_v1/conv1',\n 'tiny/block2/unit_2/bottleneck_v1/conv2',\n 'tiny/block2/unit_2/bottleneck_v1/conv3']\n self.assertItemsEqual(expected, end_points)\n\n def _stack_blocks_nondense(self, net, blocks):\n \"\"\"A simplified ResNet Block stacker without output stride control.\"\"\"\n for block in blocks:\n with tf.variable_scope(block.scope, 'block', [net]):\n for i, unit in enumerate(block.args):\n with tf.variable_scope('unit_%d' % (i + 1), values=[net]):\n net = block.unit_fn(net, rate=1, **unit)\n return net\n\n def testAtrousValuesBottleneck(self):\n \"\"\"Verify the values of dense feature extraction by atrous convolution.\n\n Make sure that dense feature extraction by stack_blocks_dense() followed by\n subsampling gives identical results to feature extraction at the nominal\n network output stride using the simple self._stack_blocks_nondense() above.\n \"\"\"\n block = resnet_v1.resnet_v1_block\n blocks = [\n block('block1', base_depth=1, num_units=2, stride=2),\n block('block2', base_depth=2, num_units=2, stride=2),\n block('block3', base_depth=4, num_units=2, stride=2),\n block('block4', base_depth=8, num_units=2, stride=1),\n ]\n nominal_stride = 8\n\n # Test both odd and even input dimensions.\n height = 30\n width = 31\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n with slim.arg_scope([slim.batch_norm], is_training=False):\n for output_stride in [1, 2, 4, 8, None]:\n with tf.Graph().as_default():\n with self.test_session() as sess:\n tf.set_random_seed(0)\n inputs = create_test_input(1, height, width, 3)\n # Dense feature extraction followed by subsampling.\n output = resnet_utils.stack_blocks_dense(inputs,\n blocks,\n output_stride)\n if output_stride is None:\n factor = 1\n else:\n factor = nominal_stride // output_stride\n\n output = resnet_utils.subsample(output, factor)\n # Make the two networks use the same weights.\n tf.get_variable_scope().reuse_variables()\n # Feature extraction at the nominal network rate.\n expected = self._stack_blocks_nondense(inputs, blocks)\n sess.run(tf.global_variables_initializer())\n output, expected = sess.run([output, expected])\n self.assertAllClose(output, expected, atol=1e-4, rtol=1e-4)\n\n\nclass ResnetCompleteNetworkTest(tf.test.TestCase):\n \"\"\"Tests with complete small ResNet v1 networks.\"\"\"\n\n def _resnet_small(self,\n inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n include_root_block=True,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v1_small'):\n \"\"\"A shallow and thin ResNet v1 for faster tests.\"\"\"\n block = resnet_v1.resnet_v1_block\n blocks = [\n block('block1', base_depth=1, num_units=3, stride=2),\n block('block2', base_depth=2, num_units=3, stride=2),\n block('block3', base_depth=4, num_units=3, stride=2),\n block('block4', base_depth=8, num_units=2, stride=1),\n ]\n return resnet_v1.resnet_v1(inputs, blocks, num_classes,\n is_training=is_training,\n global_pool=global_pool,\n output_stride=output_stride,\n include_root_block=include_root_block,\n spatial_squeeze=spatial_squeeze,\n reuse=reuse,\n scope=scope)\n\n def testClassificationEndPoints(self):\n global_pool = True\n num_classes = 10\n inputs = create_test_input(2, 224, 224, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n logits, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n spatial_squeeze=False,\n scope='resnet')\n self.assertTrue(logits.op.name.startswith('resnet/logits'))\n self.assertListEqual(logits.get_shape().as_list(), [2, 1, 1, num_classes])\n self.assertTrue('predictions' in end_points)\n self.assertListEqual(end_points['predictions'].get_shape().as_list(),\n [2, 1, 1, num_classes])\n\n def testClassificationShapes(self):\n global_pool = True\n num_classes = 10\n inputs = create_test_input(2, 224, 224, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 28, 28, 4],\n 'resnet/block2': [2, 14, 14, 8],\n 'resnet/block3': [2, 7, 7, 16],\n 'resnet/block4': [2, 7, 7, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testFullyConvolutionalEndpointShapes(self):\n global_pool = False\n num_classes = 10\n inputs = create_test_input(2, 321, 321, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n spatial_squeeze=False,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 41, 41, 4],\n 'resnet/block2': [2, 21, 21, 8],\n 'resnet/block3': [2, 11, 11, 16],\n 'resnet/block4': [2, 11, 11, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testRootlessFullyConvolutionalEndpointShapes(self):\n global_pool = False\n num_classes = 10\n inputs = create_test_input(2, 128, 128, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n include_root_block=False,\n spatial_squeeze=False,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 64, 64, 4],\n 'resnet/block2': [2, 32, 32, 8],\n 'resnet/block3': [2, 16, 16, 16],\n 'resnet/block4': [2, 16, 16, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testAtrousFullyConvolutionalEndpointShapes(self):\n global_pool = False\n num_classes = 10\n output_stride = 8\n inputs = create_test_input(2, 321, 321, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs,\n num_classes,\n global_pool=global_pool,\n output_stride=output_stride,\n spatial_squeeze=False,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 41, 41, 4],\n 'resnet/block2': [2, 41, 41, 8],\n 'resnet/block3': [2, 41, 41, 16],\n 'resnet/block4': [2, 41, 41, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testAtrousFullyConvolutionalValues(self):\n \"\"\"Verify dense feature extraction with atrous convolution.\"\"\"\n nominal_stride = 32\n for output_stride in [4, 8, 16, 32, None]:\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n with tf.Graph().as_default():\n with self.test_session() as sess:\n tf.set_random_seed(0)\n inputs = create_test_input(2, 81, 81, 3)\n # Dense feature extraction followed by subsampling.\n output, _ = self._resnet_small(inputs, None, is_training=False,\n global_pool=False,\n output_stride=output_stride)\n if output_stride is None:\n factor = 1\n else:\n factor = nominal_stride // output_stride\n output = resnet_utils.subsample(output, factor)\n # Make the two networks use the same weights.\n tf.get_variable_scope().reuse_variables()\n # Feature extraction at the nominal network rate.\n expected, _ = self._resnet_small(inputs, None, is_training=False,\n global_pool=False)\n sess.run(tf.global_variables_initializer())\n self.assertAllClose(output.eval(), expected.eval(),\n atol=1e-4, rtol=1e-4)\n\n def testUnknownBatchSize(self):\n batch = 2\n height, width = 65, 65\n global_pool = True\n num_classes = 10\n inputs = create_test_input(None, height, width, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n logits, _ = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n spatial_squeeze=False,\n scope='resnet')\n self.assertTrue(logits.op.name.startswith('resnet/logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, 1, 1, num_classes])\n images = create_test_input(batch, height, width, 3)\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEqual(output.shape, (batch, 1, 1, num_classes))\n\n def testFullyConvolutionalUnknownHeightWidth(self):\n batch = 2\n height, width = 65, 65\n global_pool = False\n inputs = create_test_input(batch, None, None, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n output, _ = self._resnet_small(inputs, None, global_pool=global_pool)\n self.assertListEqual(output.get_shape().as_list(),\n [batch, None, None, 32])\n images = create_test_input(batch, height, width, 3)\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(output, {inputs: images.eval()})\n self.assertEqual(output.shape, (batch, 3, 3, 32))\n\n def testAtrousFullyConvolutionalUnknownHeightWidth(self):\n batch = 2\n height, width = 65, 65\n global_pool = False\n output_stride = 8\n inputs = create_test_input(batch, None, None, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n output, _ = self._resnet_small(inputs,\n None,\n global_pool=global_pool,\n output_stride=output_stride)\n self.assertListEqual(output.get_shape().as_list(),\n [batch, None, None, 32])\n images = create_test_input(batch, height, width, 3)\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(output, {inputs: images.eval()})\n self.assertEqual(output.shape, (batch, 9, 9, 32))\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/resnet_v2.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains definitions for the preactivation form of Residual Networks.\n\nResidual networks (ResNets) were originally proposed in:\n[1] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun\n Deep Residual Learning for Image Recognition. arXiv:1512.03385\n\nThe full preactivation 'v2' ResNet variant implemented in this module was\nintroduced by:\n[2] Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun\n Identity Mappings in Deep Residual Networks. arXiv: 1603.05027\n\nThe key difference of the full preactivation 'v2' variant compared to the\n'v1' variant in [1] is the use of batch normalization before every weight layer.\n\nTypical use:\n\n from tensorflow.contrib.slim.nets import resnet_v2\n\nResNet-101 for image classification into 1000 classes:\n\n # inputs has shape [batch, 224, 224, 3]\n with slim.arg_scope(resnet_v2.resnet_arg_scope()):\n net, end_points = resnet_v2.resnet_v2_101(inputs, 1000, is_training=False)\n\nResNet-101 for semantic segmentation into 21 classes:\n\n # inputs has shape [batch, 513, 513, 3]\n with slim.arg_scope(resnet_v2.resnet_arg_scope(is_training)):\n net, end_points = resnet_v2.resnet_v2_101(inputs,\n 21,\n is_training=False,\n global_pool=False,\n output_stride=16)\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import resnet_utils\n\nslim = tf.contrib.slim\nresnet_arg_scope = resnet_utils.resnet_arg_scope\n\n\n@slim.add_arg_scope\ndef bottleneck(inputs, depth, depth_bottleneck, stride, rate=1,\n outputs_collections=None, scope=None):\n \"\"\"Bottleneck residual unit variant with BN before convolutions.\n\n This is the full preactivation residual unit variant proposed in [2]. See\n Fig. 1(b) of [2] for its definition. Note that we use here the bottleneck\n variant which has an extra bottleneck layer.\n\n When putting together two consecutive ResNet blocks that use this unit, one\n should use stride = 2 in the last unit of the first block.\n\n Args:\n inputs: A tensor of size [batch, height, width, channels].\n depth: The depth of the ResNet unit output.\n depth_bottleneck: The depth of the bottleneck layers.\n stride: The ResNet unit's stride. Determines the amount of downsampling of\n the units output compared to its input.\n rate: An integer, rate for atrous convolution.\n outputs_collections: Collection to add the ResNet unit output.\n scope: Optional variable_scope.\n\n Returns:\n The ResNet unit's output.\n \"\"\"\n with tf.variable_scope(scope, 'bottleneck_v2', [inputs]) as sc:\n depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)\n preact = slim.batch_norm(inputs, activation_fn=tf.nn.relu, scope='preact')\n if depth == depth_in:\n shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')\n else:\n shortcut = slim.conv2d(preact, depth, [1, 1], stride=stride,\n normalizer_fn=None, activation_fn=None,\n scope='shortcut')\n\n residual = slim.conv2d(preact, depth_bottleneck, [1, 1], stride=1,\n scope='conv1')\n residual = resnet_utils.conv2d_same(residual, depth_bottleneck, 3, stride,\n rate=rate, scope='conv2')\n residual = slim.conv2d(residual, depth, [1, 1], stride=1,\n normalizer_fn=None, activation_fn=None,\n scope='conv3')\n\n output = shortcut + residual\n\n return slim.utils.collect_named_outputs(outputs_collections,\n sc.original_name_scope,\n output)\n\n\ndef resnet_v2(inputs,\n blocks,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n include_root_block=True,\n spatial_squeeze=True,\n reuse=None,\n scope=None):\n \"\"\"Generator for v2 (preactivation) ResNet models.\n\n This function generates a family of ResNet v2 models. See the resnet_v2_*()\n methods for specific model instantiations, obtained by selecting different\n block instantiations that produce ResNets of various depths.\n\n Training for image classification on Imagenet is usually done with [224, 224]\n inputs, resulting in [7, 7] feature maps at the output of the last ResNet\n block for the ResNets defined in [1] that have nominal stride equal to 32.\n However, for dense prediction tasks we advise that one uses inputs with\n spatial dimensions that are multiples of 32 plus 1, e.g., [321, 321]. In\n this case the feature maps at the ResNet output will have spatial shape\n [(height - 1) / output_stride + 1, (width - 1) / output_stride + 1]\n and corners exactly aligned with the input image corners, which greatly\n facilitates alignment of the features to the image. Using as input [225, 225]\n images results in [8, 8] feature maps at the output of the last ResNet block.\n\n For dense prediction tasks, the ResNet needs to run in fully-convolutional\n (FCN) mode and global_pool needs to be set to False. The ResNets in [1, 2] all\n have nominal stride equal to 32 and a good choice in FCN mode is to use\n output_stride=16 in order to increase the density of the computed features at\n small computational and memory overhead, cf. http://arxiv.org/abs/1606.00915.\n\n Args:\n inputs: A tensor of size [batch, height_in, width_in, channels].\n blocks: A list of length equal to the number of ResNet blocks. Each element\n is a resnet_utils.Block object describing the units in the block.\n num_classes: Number of predicted classes for classification tasks. If None\n we return the features before the logit layer.\n is_training: whether is training or not.\n global_pool: If True, we perform global average pooling before computing the\n logits. Set to True for image classification, False for dense prediction.\n output_stride: If None, then the output will be computed at the nominal\n network stride. If output_stride is not None, it specifies the requested\n ratio of input to output spatial resolution.\n include_root_block: If True, include the initial convolution followed by\n max-pooling, if False excludes it. If excluded, `inputs` should be the\n results of an activation-less convolution.\n spatial_squeeze: if True, logits is of shape [B, C], if false logits is\n of shape [B, 1, 1, C], where B is batch_size and C is number of classes.\n To use this parameter, the input images must be smaller than 300x300\n pixels, in which case the output logit layer does not contain spatial\n information and can be removed.\n reuse: whether or not the network and its variables should be reused. To be\n able to reuse 'scope' must be given.\n scope: Optional variable_scope.\n\n\n Returns:\n net: A rank-4 tensor of size [batch, height_out, width_out, channels_out].\n If global_pool is False, then height_out and width_out are reduced by a\n factor of output_stride compared to the respective height_in and width_in,\n else both height_out and width_out equal one. If num_classes is None, then\n net is the output of the last ResNet block, potentially after global\n average pooling. If num_classes is not None, net contains the pre-softmax\n activations.\n end_points: A dictionary from components of the network to the corresponding\n activation.\n\n Raises:\n ValueError: If the target output_stride is not valid.\n \"\"\"\n with tf.variable_scope(scope, 'resnet_v2', [inputs], reuse=reuse) as sc:\n end_points_collection = sc.name + '_end_points'\n with slim.arg_scope([slim.conv2d, bottleneck,\n resnet_utils.stack_blocks_dense],\n outputs_collections=end_points_collection):\n with slim.arg_scope([slim.batch_norm], is_training=is_training):\n net = inputs\n if include_root_block:\n if output_stride is not None:\n if output_stride % 4 != 0:\n raise ValueError('The output_stride needs to be a multiple of 4.')\n output_stride /= 4\n # We do not include batch normalization or activation functions in\n # conv1 because the first ResNet unit will perform these. Cf.\n # Appendix of [2].\n with slim.arg_scope([slim.conv2d],\n activation_fn=None, normalizer_fn=None):\n net = resnet_utils.conv2d_same(net, 64, 7, stride=2, scope='conv1')\n net = slim.max_pool2d(net, [3, 3], stride=2, scope='pool1')\n net = resnet_utils.stack_blocks_dense(net, blocks, output_stride)\n # This is needed because the pre-activation variant does not have batch\n # normalization or activation functions in the residual unit output. See\n # Appendix of [2].\n net = slim.batch_norm(net, activation_fn=tf.nn.relu, scope='postnorm')\n if global_pool:\n # Global average pooling.\n net = tf.reduce_mean(net, [1, 2], name='pool5', keep_dims=True)\n if num_classes is not None:\n net = slim.conv2d(net, num_classes, [1, 1], activation_fn=None,\n normalizer_fn=None, scope='logits')\n if spatial_squeeze:\n net = tf.squeeze(net, [1, 2], name='SpatialSqueeze')\n # Convert end_points_collection into a dictionary of end_points.\n end_points = slim.utils.convert_collection_to_dict(\n end_points_collection)\n if num_classes is not None:\n end_points['predictions'] = slim.softmax(net, scope='predictions')\n return net, end_points\nresnet_v2.default_image_size = 224\n\n\ndef resnet_v2_block(scope, base_depth, num_units, stride):\n \"\"\"Helper function for creating a resnet_v2 bottleneck block.\n\n Args:\n scope: The scope of the block.\n base_depth: The depth of the bottleneck layer for each unit.\n num_units: The number of units in the block.\n stride: The stride of the block, implemented as a stride in the last unit.\n All other units have stride=1.\n\n Returns:\n A resnet_v2 bottleneck block.\n \"\"\"\n return resnet_utils.Block(scope, bottleneck, [{\n 'depth': base_depth * 4,\n 'depth_bottleneck': base_depth,\n 'stride': 1\n }] * (num_units - 1) + [{\n 'depth': base_depth * 4,\n 'depth_bottleneck': base_depth,\n 'stride': stride\n }])\nresnet_v2.default_image_size = 224\n\n\ndef resnet_v2_50(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v2_50'):\n \"\"\"ResNet-50 model of [1]. See resnet_v2() for arg and return description.\"\"\"\n blocks = [\n resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v2_block('block2', base_depth=128, num_units=4, stride=2),\n resnet_v2_block('block3', base_depth=256, num_units=6, stride=2),\n resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v2(inputs, blocks, num_classes, is_training=is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v2_50.default_image_size = resnet_v2.default_image_size\n\n\ndef resnet_v2_101(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v2_101'):\n \"\"\"ResNet-101 model of [1]. See resnet_v2() for arg and return description.\"\"\"\n blocks = [\n resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v2_block('block2', base_depth=128, num_units=4, stride=2),\n resnet_v2_block('block3', base_depth=256, num_units=23, stride=2),\n resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v2(inputs, blocks, num_classes, is_training=is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v2_101.default_image_size = resnet_v2.default_image_size\n\n\ndef resnet_v2_152(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v2_152'):\n \"\"\"ResNet-152 model of [1]. See resnet_v2() for arg and return description.\"\"\"\n blocks = [\n resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v2_block('block2', base_depth=128, num_units=8, stride=2),\n resnet_v2_block('block3', base_depth=256, num_units=36, stride=2),\n resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v2(inputs, blocks, num_classes, is_training=is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v2_152.default_image_size = resnet_v2.default_image_size\n\n\ndef resnet_v2_200(inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v2_200'):\n \"\"\"ResNet-200 model of [2]. See resnet_v2() for arg and return description.\"\"\"\n blocks = [\n resnet_v2_block('block1', base_depth=64, num_units=3, stride=2),\n resnet_v2_block('block2', base_depth=128, num_units=24, stride=2),\n resnet_v2_block('block3', base_depth=256, num_units=36, stride=2),\n resnet_v2_block('block4', base_depth=512, num_units=3, stride=1),\n ]\n return resnet_v2(inputs, blocks, num_classes, is_training=is_training,\n global_pool=global_pool, output_stride=output_stride,\n include_root_block=True, spatial_squeeze=spatial_squeeze,\n reuse=reuse, scope=scope)\nresnet_v2_200.default_image_size = resnet_v2.default_image_size\n"
},
{
"path": "nets/resnet_v2_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for slim.nets.resnet_v2.\"\"\"\n\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport numpy as np\nimport tensorflow as tf\n\nfrom nets import resnet_utils\nfrom nets import resnet_v2\n\nslim = tf.contrib.slim\n\n\ndef create_test_input(batch_size, height, width, channels):\n \"\"\"Create test input tensor.\n\n Args:\n batch_size: The number of images per batch or `None` if unknown.\n height: The height of each image or `None` if unknown.\n width: The width of each image or `None` if unknown.\n channels: The number of channels per image or `None` if unknown.\n\n Returns:\n Either a placeholder `Tensor` of dimension\n [batch_size, height, width, channels] if any of the inputs are `None` or a\n constant `Tensor` with the mesh grid values along the spatial dimensions.\n \"\"\"\n if None in [batch_size, height, width, channels]:\n return tf.placeholder(tf.float32, (batch_size, height, width, channels))\n else:\n return tf.to_float(\n np.tile(\n np.reshape(\n np.reshape(np.arange(height), [height, 1]) +\n np.reshape(np.arange(width), [1, width]),\n [1, height, width, 1]),\n [batch_size, 1, 1, channels]))\n\n\nclass ResnetUtilsTest(tf.test.TestCase):\n\n def testSubsampleThreeByThree(self):\n x = tf.reshape(tf.to_float(tf.range(9)), [1, 3, 3, 1])\n x = resnet_utils.subsample(x, 2)\n expected = tf.reshape(tf.constant([0, 2, 6, 8]), [1, 2, 2, 1])\n with self.test_session():\n self.assertAllClose(x.eval(), expected.eval())\n\n def testSubsampleFourByFour(self):\n x = tf.reshape(tf.to_float(tf.range(16)), [1, 4, 4, 1])\n x = resnet_utils.subsample(x, 2)\n expected = tf.reshape(tf.constant([0, 2, 8, 10]), [1, 2, 2, 1])\n with self.test_session():\n self.assertAllClose(x.eval(), expected.eval())\n\n def testConv2DSameEven(self):\n n, n2 = 4, 2\n\n # Input image.\n x = create_test_input(1, n, n, 1)\n\n # Convolution kernel.\n w = create_test_input(1, 3, 3, 1)\n w = tf.reshape(w, [3, 3, 1, 1])\n\n tf.get_variable('Conv/weights', initializer=w)\n tf.get_variable('Conv/biases', initializer=tf.zeros([1]))\n tf.get_variable_scope().reuse_variables()\n\n y1 = slim.conv2d(x, 1, [3, 3], stride=1, scope='Conv')\n y1_expected = tf.to_float([[14, 28, 43, 26],\n [28, 48, 66, 37],\n [43, 66, 84, 46],\n [26, 37, 46, 22]])\n y1_expected = tf.reshape(y1_expected, [1, n, n, 1])\n\n y2 = resnet_utils.subsample(y1, 2)\n y2_expected = tf.to_float([[14, 43],\n [43, 84]])\n y2_expected = tf.reshape(y2_expected, [1, n2, n2, 1])\n\n y3 = resnet_utils.conv2d_same(x, 1, 3, stride=2, scope='Conv')\n y3_expected = y2_expected\n\n y4 = slim.conv2d(x, 1, [3, 3], stride=2, scope='Conv')\n y4_expected = tf.to_float([[48, 37],\n [37, 22]])\n y4_expected = tf.reshape(y4_expected, [1, n2, n2, 1])\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n self.assertAllClose(y1.eval(), y1_expected.eval())\n self.assertAllClose(y2.eval(), y2_expected.eval())\n self.assertAllClose(y3.eval(), y3_expected.eval())\n self.assertAllClose(y4.eval(), y4_expected.eval())\n\n def testConv2DSameOdd(self):\n n, n2 = 5, 3\n\n # Input image.\n x = create_test_input(1, n, n, 1)\n\n # Convolution kernel.\n w = create_test_input(1, 3, 3, 1)\n w = tf.reshape(w, [3, 3, 1, 1])\n\n tf.get_variable('Conv/weights', initializer=w)\n tf.get_variable('Conv/biases', initializer=tf.zeros([1]))\n tf.get_variable_scope().reuse_variables()\n\n y1 = slim.conv2d(x, 1, [3, 3], stride=1, scope='Conv')\n y1_expected = tf.to_float([[14, 28, 43, 58, 34],\n [28, 48, 66, 84, 46],\n [43, 66, 84, 102, 55],\n [58, 84, 102, 120, 64],\n [34, 46, 55, 64, 30]])\n y1_expected = tf.reshape(y1_expected, [1, n, n, 1])\n\n y2 = resnet_utils.subsample(y1, 2)\n y2_expected = tf.to_float([[14, 43, 34],\n [43, 84, 55],\n [34, 55, 30]])\n y2_expected = tf.reshape(y2_expected, [1, n2, n2, 1])\n\n y3 = resnet_utils.conv2d_same(x, 1, 3, stride=2, scope='Conv')\n y3_expected = y2_expected\n\n y4 = slim.conv2d(x, 1, [3, 3], stride=2, scope='Conv')\n y4_expected = y2_expected\n\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n self.assertAllClose(y1.eval(), y1_expected.eval())\n self.assertAllClose(y2.eval(), y2_expected.eval())\n self.assertAllClose(y3.eval(), y3_expected.eval())\n self.assertAllClose(y4.eval(), y4_expected.eval())\n\n def _resnet_plain(self, inputs, blocks, output_stride=None, scope=None):\n \"\"\"A plain ResNet without extra layers before or after the ResNet blocks.\"\"\"\n with tf.variable_scope(scope, values=[inputs]):\n with slim.arg_scope([slim.conv2d], outputs_collections='end_points'):\n net = resnet_utils.stack_blocks_dense(inputs, blocks, output_stride)\n end_points = slim.utils.convert_collection_to_dict('end_points')\n return net, end_points\n\n def testEndPointsV2(self):\n \"\"\"Test the end points of a tiny v2 bottleneck network.\"\"\"\n blocks = [\n resnet_v2.resnet_v2_block(\n 'block1', base_depth=1, num_units=2, stride=2),\n resnet_v2.resnet_v2_block(\n 'block2', base_depth=2, num_units=2, stride=1),\n ]\n inputs = create_test_input(2, 32, 16, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_plain(inputs, blocks, scope='tiny')\n expected = [\n 'tiny/block1/unit_1/bottleneck_v2/shortcut',\n 'tiny/block1/unit_1/bottleneck_v2/conv1',\n 'tiny/block1/unit_1/bottleneck_v2/conv2',\n 'tiny/block1/unit_1/bottleneck_v2/conv3',\n 'tiny/block1/unit_2/bottleneck_v2/conv1',\n 'tiny/block1/unit_2/bottleneck_v2/conv2',\n 'tiny/block1/unit_2/bottleneck_v2/conv3',\n 'tiny/block2/unit_1/bottleneck_v2/shortcut',\n 'tiny/block2/unit_1/bottleneck_v2/conv1',\n 'tiny/block2/unit_1/bottleneck_v2/conv2',\n 'tiny/block2/unit_1/bottleneck_v2/conv3',\n 'tiny/block2/unit_2/bottleneck_v2/conv1',\n 'tiny/block2/unit_2/bottleneck_v2/conv2',\n 'tiny/block2/unit_2/bottleneck_v2/conv3']\n self.assertItemsEqual(expected, end_points)\n\n def _stack_blocks_nondense(self, net, blocks):\n \"\"\"A simplified ResNet Block stacker without output stride control.\"\"\"\n for block in blocks:\n with tf.variable_scope(block.scope, 'block', [net]):\n for i, unit in enumerate(block.args):\n with tf.variable_scope('unit_%d' % (i + 1), values=[net]):\n net = block.unit_fn(net, rate=1, **unit)\n return net\n\n def testAtrousValuesBottleneck(self):\n \"\"\"Verify the values of dense feature extraction by atrous convolution.\n\n Make sure that dense feature extraction by stack_blocks_dense() followed by\n subsampling gives identical results to feature extraction at the nominal\n network output stride using the simple self._stack_blocks_nondense() above.\n \"\"\"\n block = resnet_v2.resnet_v2_block\n blocks = [\n block('block1', base_depth=1, num_units=2, stride=2),\n block('block2', base_depth=2, num_units=2, stride=2),\n block('block3', base_depth=4, num_units=2, stride=2),\n block('block4', base_depth=8, num_units=2, stride=1),\n ]\n nominal_stride = 8\n\n # Test both odd and even input dimensions.\n height = 30\n width = 31\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n with slim.arg_scope([slim.batch_norm], is_training=False):\n for output_stride in [1, 2, 4, 8, None]:\n with tf.Graph().as_default():\n with self.test_session() as sess:\n tf.set_random_seed(0)\n inputs = create_test_input(1, height, width, 3)\n # Dense feature extraction followed by subsampling.\n output = resnet_utils.stack_blocks_dense(inputs,\n blocks,\n output_stride)\n if output_stride is None:\n factor = 1\n else:\n factor = nominal_stride // output_stride\n\n output = resnet_utils.subsample(output, factor)\n # Make the two networks use the same weights.\n tf.get_variable_scope().reuse_variables()\n # Feature extraction at the nominal network rate.\n expected = self._stack_blocks_nondense(inputs, blocks)\n sess.run(tf.global_variables_initializer())\n output, expected = sess.run([output, expected])\n self.assertAllClose(output, expected, atol=1e-4, rtol=1e-4)\n\n\nclass ResnetCompleteNetworkTest(tf.test.TestCase):\n \"\"\"Tests with complete small ResNet v2 networks.\"\"\"\n\n def _resnet_small(self,\n inputs,\n num_classes=None,\n is_training=True,\n global_pool=True,\n output_stride=None,\n include_root_block=True,\n spatial_squeeze=True,\n reuse=None,\n scope='resnet_v2_small'):\n \"\"\"A shallow and thin ResNet v2 for faster tests.\"\"\"\n block = resnet_v2.resnet_v2_block\n blocks = [\n block('block1', base_depth=1, num_units=3, stride=2),\n block('block2', base_depth=2, num_units=3, stride=2),\n block('block3', base_depth=4, num_units=3, stride=2),\n block('block4', base_depth=8, num_units=2, stride=1),\n ]\n return resnet_v2.resnet_v2(inputs, blocks, num_classes,\n is_training=is_training,\n global_pool=global_pool,\n output_stride=output_stride,\n include_root_block=include_root_block,\n spatial_squeeze=spatial_squeeze,\n reuse=reuse,\n scope=scope)\n\n def testClassificationEndPoints(self):\n global_pool = True\n num_classes = 10\n inputs = create_test_input(2, 224, 224, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n logits, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n spatial_squeeze=False,\n scope='resnet')\n self.assertTrue(logits.op.name.startswith('resnet/logits'))\n self.assertListEqual(logits.get_shape().as_list(), [2, 1, 1, num_classes])\n self.assertTrue('predictions' in end_points)\n self.assertListEqual(end_points['predictions'].get_shape().as_list(),\n [2, 1, 1, num_classes])\n\n def testClassificationShapes(self):\n global_pool = True\n num_classes = 10\n inputs = create_test_input(2, 224, 224, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 28, 28, 4],\n 'resnet/block2': [2, 14, 14, 8],\n 'resnet/block3': [2, 7, 7, 16],\n 'resnet/block4': [2, 7, 7, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testFullyConvolutionalEndpointShapes(self):\n global_pool = False\n num_classes = 10\n inputs = create_test_input(2, 321, 321, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n spatial_squeeze=False,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 41, 41, 4],\n 'resnet/block2': [2, 21, 21, 8],\n 'resnet/block3': [2, 11, 11, 16],\n 'resnet/block4': [2, 11, 11, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testRootlessFullyConvolutionalEndpointShapes(self):\n global_pool = False\n num_classes = 10\n inputs = create_test_input(2, 128, 128, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n include_root_block=False,\n spatial_squeeze=False,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 64, 64, 4],\n 'resnet/block2': [2, 32, 32, 8],\n 'resnet/block3': [2, 16, 16, 16],\n 'resnet/block4': [2, 16, 16, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testAtrousFullyConvolutionalEndpointShapes(self):\n global_pool = False\n num_classes = 10\n output_stride = 8\n inputs = create_test_input(2, 321, 321, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n _, end_points = self._resnet_small(inputs,\n num_classes,\n global_pool=global_pool,\n output_stride=output_stride,\n spatial_squeeze=False,\n scope='resnet')\n endpoint_to_shape = {\n 'resnet/block1': [2, 41, 41, 4],\n 'resnet/block2': [2, 41, 41, 8],\n 'resnet/block3': [2, 41, 41, 16],\n 'resnet/block4': [2, 41, 41, 32]}\n for endpoint in endpoint_to_shape:\n shape = endpoint_to_shape[endpoint]\n self.assertListEqual(end_points[endpoint].get_shape().as_list(), shape)\n\n def testAtrousFullyConvolutionalValues(self):\n \"\"\"Verify dense feature extraction with atrous convolution.\"\"\"\n nominal_stride = 32\n for output_stride in [4, 8, 16, 32, None]:\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n with tf.Graph().as_default():\n with self.test_session() as sess:\n tf.set_random_seed(0)\n inputs = create_test_input(2, 81, 81, 3)\n # Dense feature extraction followed by subsampling.\n output, _ = self._resnet_small(inputs, None,\n is_training=False,\n global_pool=False,\n output_stride=output_stride)\n if output_stride is None:\n factor = 1\n else:\n factor = nominal_stride // output_stride\n output = resnet_utils.subsample(output, factor)\n # Make the two networks use the same weights.\n tf.get_variable_scope().reuse_variables()\n # Feature extraction at the nominal network rate.\n expected, _ = self._resnet_small(inputs, None,\n is_training=False,\n global_pool=False)\n sess.run(tf.global_variables_initializer())\n self.assertAllClose(output.eval(), expected.eval(),\n atol=1e-4, rtol=1e-4)\n\n def testUnknownBatchSize(self):\n batch = 2\n height, width = 65, 65\n global_pool = True\n num_classes = 10\n inputs = create_test_input(None, height, width, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n logits, _ = self._resnet_small(inputs, num_classes,\n global_pool=global_pool,\n spatial_squeeze=False,\n scope='resnet')\n self.assertTrue(logits.op.name.startswith('resnet/logits'))\n self.assertListEqual(logits.get_shape().as_list(),\n [None, 1, 1, num_classes])\n images = create_test_input(batch, height, width, 3)\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits, {inputs: images.eval()})\n self.assertEqual(output.shape, (batch, 1, 1, num_classes))\n\n def testFullyConvolutionalUnknownHeightWidth(self):\n batch = 2\n height, width = 65, 65\n global_pool = False\n inputs = create_test_input(batch, None, None, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n output, _ = self._resnet_small(inputs, None,\n global_pool=global_pool)\n self.assertListEqual(output.get_shape().as_list(),\n [batch, None, None, 32])\n images = create_test_input(batch, height, width, 3)\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(output, {inputs: images.eval()})\n self.assertEqual(output.shape, (batch, 3, 3, 32))\n\n def testAtrousFullyConvolutionalUnknownHeightWidth(self):\n batch = 2\n height, width = 65, 65\n global_pool = False\n output_stride = 8\n inputs = create_test_input(batch, None, None, 3)\n with slim.arg_scope(resnet_utils.resnet_arg_scope()):\n output, _ = self._resnet_small(inputs,\n None,\n global_pool=global_pool,\n output_stride=output_stride)\n self.assertListEqual(output.get_shape().as_list(),\n [batch, None, None, 32])\n images = create_test_input(batch, height, width, 3)\n with self.test_session() as sess:\n sess.run(tf.global_variables_initializer())\n output = sess.run(output, {inputs: images.eval()})\n self.assertEqual(output.shape, (batch, 9, 9, 32))\n\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "nets/vgg.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Contains model definitions for versions of the Oxford VGG network.\n\nThese model definitions were introduced in the following technical report:\n\n Very Deep Convolutional Networks For Large-Scale Image Recognition\n Karen Simonyan and Andrew Zisserman\n arXiv technical report, 2015\n PDF: http://arxiv.org/pdf/1409.1556.pdf\n ILSVRC 2014 Slides: http://www.robots.ox.ac.uk/~karen/pdf/ILSVRC_2014.pdf\n CC-BY-4.0\n\nMore information can be obtained from the VGG website:\nwww.robots.ox.ac.uk/~vgg/research/very_deep/\n\nUsage:\n with slim.arg_scope(vgg.vgg_arg_scope()):\n outputs, end_points = vgg.vgg_a(inputs)\n\n with slim.arg_scope(vgg.vgg_arg_scope()):\n outputs, end_points = vgg.vgg_16(inputs)\n\n@@vgg_a\n@@vgg_16\n@@vgg_19\n\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nslim = tf.contrib.slim\n\n\ndef vgg_arg_scope(weight_decay=0.0005):\n \"\"\"Defines the VGG arg scope.\n\n Args:\n weight_decay: The l2 regularization coefficient.\n\n Returns:\n An arg_scope.\n \"\"\"\n with slim.arg_scope([slim.conv2d, slim.fully_connected],\n activation_fn=tf.nn.relu,\n weights_regularizer=slim.l2_regularizer(weight_decay),\n biases_initializer=tf.zeros_initializer()):\n with slim.arg_scope([slim.conv2d], padding='SAME') as arg_sc:\n return arg_sc\n\n\ndef vgg_a(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.5,\n spatial_squeeze=True,\n scope='vgg_a',\n fc_conv_padding='VALID'):\n \"\"\"Oxford Net VGG 11-Layers version A Example.\n\n Note: All the fully_connected layers have been transformed to conv2d layers.\n To use in classification mode, resize input to 224x224.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether or not the model is being trained.\n dropout_keep_prob: the probability that activations are kept in the dropout\n layers during training.\n spatial_squeeze: whether or not should squeeze the spatial dimensions of the\n outputs. Useful to remove unnecessary dimensions for classification.\n scope: Optional scope for the variables.\n fc_conv_padding: the type of padding to use for the fully connected layer\n that is implemented as a convolutional layer. Use 'SAME' padding if you\n are applying the network in a fully convolutional manner and want to\n get a prediction map downsampled by a factor of 32 as an output.\n Otherwise, the output prediction map will be (input / 32) - 6 in case of\n 'VALID' padding.\n\n Returns:\n the last op containing the log predictions and end_points dict.\n \"\"\"\n with tf.variable_scope(scope, 'vgg_a', [inputs]) as sc:\n end_points_collection = sc.name + '_end_points'\n # Collect outputs for conv2d, fully_connected and max_pool2d.\n with slim.arg_scope([slim.conv2d, slim.max_pool2d],\n outputs_collections=end_points_collection):\n net = slim.repeat(inputs, 1, slim.conv2d, 64, [3, 3], scope='conv1')\n net = slim.max_pool2d(net, [2, 2], scope='pool1')\n net = slim.repeat(net, 1, slim.conv2d, 128, [3, 3], scope='conv2')\n net = slim.max_pool2d(net, [2, 2], scope='pool2')\n net = slim.repeat(net, 2, slim.conv2d, 256, [3, 3], scope='conv3')\n net = slim.max_pool2d(net, [2, 2], scope='pool3')\n net = slim.repeat(net, 2, slim.conv2d, 512, [3, 3], scope='conv4')\n net = slim.max_pool2d(net, [2, 2], scope='pool4')\n net = slim.repeat(net, 2, slim.conv2d, 512, [3, 3], scope='conv5')\n net = slim.max_pool2d(net, [2, 2], scope='pool5')\n # Use conv2d instead of fully_connected layers.\n net = slim.conv2d(net, 4096, [7, 7], padding=fc_conv_padding, scope='fc6')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout6')\n net = slim.conv2d(net, 4096, [1, 1], scope='fc7')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout7')\n net = slim.conv2d(net, num_classes, [1, 1],\n activation_fn=None,\n normalizer_fn=None,\n scope='fc8')\n # Convert end_points_collection into a end_point dict.\n end_points = slim.utils.convert_collection_to_dict(end_points_collection)\n if spatial_squeeze:\n net = tf.squeeze(net, [1, 2], name='fc8/squeezed')\n end_points[sc.name + '/fc8'] = net\n return net, end_points\nvgg_a.default_image_size = 224\n\n\ndef vgg_16(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.5,\n spatial_squeeze=True,\n scope='vgg_16',\n fc_conv_padding='VALID'):\n \"\"\"Oxford Net VGG 16-Layers version D Example.\n\n Note: All the fully_connected layers have been transformed to conv2d layers.\n To use in classification mode, resize input to 224x224.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether or not the model is being trained.\n dropout_keep_prob: the probability that activations are kept in the dropout\n layers during training.\n spatial_squeeze: whether or not should squeeze the spatial dimensions of the\n outputs. Useful to remove unnecessary dimensions for classification.\n scope: Optional scope for the variables.\n fc_conv_padding: the type of padding to use for the fully connected layer\n that is implemented as a convolutional layer. Use 'SAME' padding if you\n are applying the network in a fully convolutional manner and want to\n get a prediction map downsampled by a factor of 32 as an output.\n Otherwise, the output prediction map will be (input / 32) - 6 in case of\n 'VALID' padding.\n\n Returns:\n the last op containing the log predictions and end_points dict.\n \"\"\"\n with tf.variable_scope(scope, 'vgg_16', [inputs]) as sc:\n end_points_collection = sc.name + '_end_points'\n # Collect outputs for conv2d, fully_connected and max_pool2d.\n with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.max_pool2d],\n outputs_collections=end_points_collection):\n net = slim.repeat(inputs, 2, slim.conv2d, 64, [3, 3], scope='conv1')\n net = slim.max_pool2d(net, [2, 2], scope='pool1')\n net = slim.repeat(net, 2, slim.conv2d, 128, [3, 3], scope='conv2')\n net = slim.max_pool2d(net, [2, 2], scope='pool2')\n net = slim.repeat(net, 3, slim.conv2d, 256, [3, 3], scope='conv3')\n net = slim.max_pool2d(net, [2, 2], scope='pool3')\n net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3], scope='conv4')\n net = slim.max_pool2d(net, [2, 2], scope='pool4')\n net = slim.repeat(net, 3, slim.conv2d, 512, [3, 3], scope='conv5')\n net = slim.max_pool2d(net, [2, 2], scope='pool5')\n # Use conv2d instead of fully_connected layers.\n net = slim.conv2d(net, 4096, [7, 7], padding=fc_conv_padding, scope='fc6')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout6')\n net = slim.conv2d(net, 4096, [1, 1], scope='fc7')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout7')\n net = slim.conv2d(net, num_classes, [1, 1],\n activation_fn=None,\n normalizer_fn=None,\n scope='fc8')\n # Convert end_points_collection into a end_point dict.\n end_points = slim.utils.convert_collection_to_dict(end_points_collection)\n if spatial_squeeze:\n net = tf.squeeze(net, [1, 2], name='fc8/squeezed')\n end_points[sc.name + '/fc8'] = net\n return net, end_points\nvgg_16.default_image_size = 224\n\n\ndef vgg_19(inputs,\n num_classes=1000,\n is_training=True,\n dropout_keep_prob=0.5,\n spatial_squeeze=True,\n scope='vgg_19',\n fc_conv_padding='VALID'):\n \"\"\"Oxford Net VGG 19-Layers version E Example.\n\n Note: All the fully_connected layers have been transformed to conv2d layers.\n To use in classification mode, resize input to 224x224.\n\n Args:\n inputs: a tensor of size [batch_size, height, width, channels].\n num_classes: number of predicted classes.\n is_training: whether or not the model is being trained.\n dropout_keep_prob: the probability that activations are kept in the dropout\n layers during training.\n spatial_squeeze: whether or not should squeeze the spatial dimensions of the\n outputs. Useful to remove unnecessary dimensions for classification.\n scope: Optional scope for the variables.\n fc_conv_padding: the type of padding to use for the fully connected layer\n that is implemented as a convolutional layer. Use 'SAME' padding if you\n are applying the network in a fully convolutional manner and want to\n get a prediction map downsampled by a factor of 32 as an output.\n Otherwise, the output prediction map will be (input / 32) - 6 in case of\n 'VALID' padding.\n\n\n Returns:\n the last op containing the log predictions and end_points dict.\n \"\"\"\n with tf.variable_scope(scope, 'vgg_19', [inputs]) as sc:\n end_points_collection = sc.name + '_end_points'\n # Collect outputs for conv2d, fully_connected and max_pool2d.\n with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.max_pool2d],\n outputs_collections=end_points_collection):\n net = slim.repeat(inputs, 2, slim.conv2d, 64, [3, 3], scope='conv1')\n net = slim.max_pool2d(net, [2, 2], scope='pool1')\n net = slim.repeat(net, 2, slim.conv2d, 128, [3, 3], scope='conv2')\n net = slim.max_pool2d(net, [2, 2], scope='pool2')\n net = slim.repeat(net, 4, slim.conv2d, 256, [3, 3], scope='conv3')\n net = slim.max_pool2d(net, [2, 2], scope='pool3')\n net = slim.repeat(net, 4, slim.conv2d, 512, [3, 3], scope='conv4')\n net = slim.max_pool2d(net, [2, 2], scope='pool4')\n net = slim.repeat(net, 4, slim.conv2d, 512, [3, 3], scope='conv5')\n net = slim.max_pool2d(net, [2, 2], scope='pool5')\n # Use conv2d instead of fully_connected layers.\n net = slim.conv2d(net, 4096, [7, 7], padding=fc_conv_padding, scope='fc6')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout6')\n net = slim.conv2d(net, 4096, [1, 1], scope='fc7')\n net = slim.dropout(net, dropout_keep_prob, is_training=is_training,\n scope='dropout7')\n net = slim.conv2d(net, num_classes, [1, 1],\n activation_fn=None,\n normalizer_fn=None,\n scope='fc8')\n # Convert end_points_collection into a end_point dict.\n end_points = slim.utils.convert_collection_to_dict(end_points_collection)\n if spatial_squeeze:\n net = tf.squeeze(net, [1, 2], name='fc8/squeezed')\n end_points[sc.name + '/fc8'] = net\n return net, end_points\nvgg_19.default_image_size = 224\n\n# Alias\nvgg_d = vgg_16\nvgg_e = vgg_19\n"
},
{
"path": "nets/vgg_test.py",
"content": "# Copyright 2016 The TensorFlow Authors. All Rights Reserved.\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n# http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n# ==============================================================================\n\"\"\"Tests for slim.nets.vgg.\"\"\"\nfrom __future__ import absolute_import\nfrom __future__ import division\nfrom __future__ import print_function\n\nimport tensorflow as tf\n\nfrom nets import vgg\n\nslim = tf.contrib.slim\n\n\nclass VGGATest(tf.test.TestCase):\n\n def testBuild(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_a(inputs, num_classes)\n self.assertEquals(logits.op.name, 'vgg_a/fc8/squeezed')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testFullyConvolutional(self):\n batch_size = 1\n height, width = 256, 256\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_a(inputs, num_classes, spatial_squeeze=False)\n self.assertEquals(logits.op.name, 'vgg_a/fc8/BiasAdd')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, 2, 2, num_classes])\n\n def testEndPoints(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = vgg.vgg_a(inputs, num_classes)\n expected_names = ['vgg_a/conv1/conv1_1',\n 'vgg_a/pool1',\n 'vgg_a/conv2/conv2_1',\n 'vgg_a/pool2',\n 'vgg_a/conv3/conv3_1',\n 'vgg_a/conv3/conv3_2',\n 'vgg_a/pool3',\n 'vgg_a/conv4/conv4_1',\n 'vgg_a/conv4/conv4_2',\n 'vgg_a/pool4',\n 'vgg_a/conv5/conv5_1',\n 'vgg_a/conv5/conv5_2',\n 'vgg_a/pool5',\n 'vgg_a/fc6',\n 'vgg_a/fc7',\n 'vgg_a/fc8'\n ]\n self.assertSetEqual(set(end_points.keys()), set(expected_names))\n\n def testModelVariables(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n vgg.vgg_a(inputs, num_classes)\n expected_names = ['vgg_a/conv1/conv1_1/weights',\n 'vgg_a/conv1/conv1_1/biases',\n 'vgg_a/conv2/conv2_1/weights',\n 'vgg_a/conv2/conv2_1/biases',\n 'vgg_a/conv3/conv3_1/weights',\n 'vgg_a/conv3/conv3_1/biases',\n 'vgg_a/conv3/conv3_2/weights',\n 'vgg_a/conv3/conv3_2/biases',\n 'vgg_a/conv4/conv4_1/weights',\n 'vgg_a/conv4/conv4_1/biases',\n 'vgg_a/conv4/conv4_2/weights',\n 'vgg_a/conv4/conv4_2/biases',\n 'vgg_a/conv5/conv5_1/weights',\n 'vgg_a/conv5/conv5_1/biases',\n 'vgg_a/conv5/conv5_2/weights',\n 'vgg_a/conv5/conv5_2/biases',\n 'vgg_a/fc6/weights',\n 'vgg_a/fc6/biases',\n 'vgg_a/fc7/weights',\n 'vgg_a/fc7/biases',\n 'vgg_a/fc8/weights',\n 'vgg_a/fc8/biases',\n ]\n model_variables = [v.op.name for v in slim.get_model_variables()]\n self.assertSetEqual(set(model_variables), set(expected_names))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_a(eval_inputs, is_training=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n predictions = tf.argmax(logits, 1)\n self.assertListEqual(predictions.get_shape().as_list(), [batch_size])\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 2\n eval_batch_size = 1\n train_height, train_width = 224, 224\n eval_height, eval_width = 256, 256\n num_classes = 1000\n with self.test_session():\n train_inputs = tf.random_uniform(\n (train_batch_size, train_height, train_width, 3))\n logits, _ = vgg.vgg_a(train_inputs)\n self.assertListEqual(logits.get_shape().as_list(),\n [train_batch_size, num_classes])\n tf.get_variable_scope().reuse_variables()\n eval_inputs = tf.random_uniform(\n (eval_batch_size, eval_height, eval_width, 3))\n logits, _ = vgg.vgg_a(eval_inputs, is_training=False,\n spatial_squeeze=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [eval_batch_size, 2, 2, num_classes])\n logits = tf.reduce_mean(logits, [1, 2])\n predictions = tf.argmax(logits, 1)\n self.assertEquals(predictions.get_shape().as_list(), [eval_batch_size])\n\n def testForward(self):\n batch_size = 1\n height, width = 224, 224\n with self.test_session() as sess:\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_a(inputs)\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits)\n self.assertTrue(output.any())\n\n\nclass VGG16Test(tf.test.TestCase):\n\n def testBuild(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_16(inputs, num_classes)\n self.assertEquals(logits.op.name, 'vgg_16/fc8/squeezed')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testFullyConvolutional(self):\n batch_size = 1\n height, width = 256, 256\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_16(inputs, num_classes, spatial_squeeze=False)\n self.assertEquals(logits.op.name, 'vgg_16/fc8/BiasAdd')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, 2, 2, num_classes])\n\n def testEndPoints(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = vgg.vgg_16(inputs, num_classes)\n expected_names = ['vgg_16/conv1/conv1_1',\n 'vgg_16/conv1/conv1_2',\n 'vgg_16/pool1',\n 'vgg_16/conv2/conv2_1',\n 'vgg_16/conv2/conv2_2',\n 'vgg_16/pool2',\n 'vgg_16/conv3/conv3_1',\n 'vgg_16/conv3/conv3_2',\n 'vgg_16/conv3/conv3_3',\n 'vgg_16/pool3',\n 'vgg_16/conv4/conv4_1',\n 'vgg_16/conv4/conv4_2',\n 'vgg_16/conv4/conv4_3',\n 'vgg_16/pool4',\n 'vgg_16/conv5/conv5_1',\n 'vgg_16/conv5/conv5_2',\n 'vgg_16/conv5/conv5_3',\n 'vgg_16/pool5',\n 'vgg_16/fc6',\n 'vgg_16/fc7',\n 'vgg_16/fc8'\n ]\n self.assertSetEqual(set(end_points.keys()), set(expected_names))\n\n def testModelVariables(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n vgg.vgg_16(inputs, num_classes)\n expected_names = ['vgg_16/conv1/conv1_1/weights',\n 'vgg_16/conv1/conv1_1/biases',\n 'vgg_16/conv1/conv1_2/weights',\n 'vgg_16/conv1/conv1_2/biases',\n 'vgg_16/conv2/conv2_1/weights',\n 'vgg_16/conv2/conv2_1/biases',\n 'vgg_16/conv2/conv2_2/weights',\n 'vgg_16/conv2/conv2_2/biases',\n 'vgg_16/conv3/conv3_1/weights',\n 'vgg_16/conv3/conv3_1/biases',\n 'vgg_16/conv3/conv3_2/weights',\n 'vgg_16/conv3/conv3_2/biases',\n 'vgg_16/conv3/conv3_3/weights',\n 'vgg_16/conv3/conv3_3/biases',\n 'vgg_16/conv4/conv4_1/weights',\n 'vgg_16/conv4/conv4_1/biases',\n 'vgg_16/conv4/conv4_2/weights',\n 'vgg_16/conv4/conv4_2/biases',\n 'vgg_16/conv4/conv4_3/weights',\n 'vgg_16/conv4/conv4_3/biases',\n 'vgg_16/conv5/conv5_1/weights',\n 'vgg_16/conv5/conv5_1/biases',\n 'vgg_16/conv5/conv5_2/weights',\n 'vgg_16/conv5/conv5_2/biases',\n 'vgg_16/conv5/conv5_3/weights',\n 'vgg_16/conv5/conv5_3/biases',\n 'vgg_16/fc6/weights',\n 'vgg_16/fc6/biases',\n 'vgg_16/fc7/weights',\n 'vgg_16/fc7/biases',\n 'vgg_16/fc8/weights',\n 'vgg_16/fc8/biases',\n ]\n model_variables = [v.op.name for v in slim.get_model_variables()]\n self.assertSetEqual(set(model_variables), set(expected_names))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_16(eval_inputs, is_training=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n predictions = tf.argmax(logits, 1)\n self.assertListEqual(predictions.get_shape().as_list(), [batch_size])\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 2\n eval_batch_size = 1\n train_height, train_width = 224, 224\n eval_height, eval_width = 256, 256\n num_classes = 1000\n with self.test_session():\n train_inputs = tf.random_uniform(\n (train_batch_size, train_height, train_width, 3))\n logits, _ = vgg.vgg_16(train_inputs)\n self.assertListEqual(logits.get_shape().as_list(),\n [train_batch_size, num_classes])\n tf.get_variable_scope().reuse_variables()\n eval_inputs = tf.random_uniform(\n (eval_batch_size, eval_height, eval_width, 3))\n logits, _ = vgg.vgg_16(eval_inputs, is_training=False,\n spatial_squeeze=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [eval_batch_size, 2, 2, num_classes])\n logits = tf.reduce_mean(logits, [1, 2])\n predictions = tf.argmax(logits, 1)\n self.assertEquals(predictions.get_shape().as_list(), [eval_batch_size])\n\n def testForward(self):\n batch_size = 1\n height, width = 224, 224\n with self.test_session() as sess:\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_16(inputs)\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits)\n self.assertTrue(output.any())\n\n\nclass VGG19Test(tf.test.TestCase):\n\n def testBuild(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_19(inputs, num_classes)\n self.assertEquals(logits.op.name, 'vgg_19/fc8/squeezed')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n\n def testFullyConvolutional(self):\n batch_size = 1\n height, width = 256, 256\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_19(inputs, num_classes, spatial_squeeze=False)\n self.assertEquals(logits.op.name, 'vgg_19/fc8/BiasAdd')\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, 2, 2, num_classes])\n\n def testEndPoints(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n _, end_points = vgg.vgg_19(inputs, num_classes)\n expected_names = [\n 'vgg_19/conv1/conv1_1',\n 'vgg_19/conv1/conv1_2',\n 'vgg_19/pool1',\n 'vgg_19/conv2/conv2_1',\n 'vgg_19/conv2/conv2_2',\n 'vgg_19/pool2',\n 'vgg_19/conv3/conv3_1',\n 'vgg_19/conv3/conv3_2',\n 'vgg_19/conv3/conv3_3',\n 'vgg_19/conv3/conv3_4',\n 'vgg_19/pool3',\n 'vgg_19/conv4/conv4_1',\n 'vgg_19/conv4/conv4_2',\n 'vgg_19/conv4/conv4_3',\n 'vgg_19/conv4/conv4_4',\n 'vgg_19/pool4',\n 'vgg_19/conv5/conv5_1',\n 'vgg_19/conv5/conv5_2',\n 'vgg_19/conv5/conv5_3',\n 'vgg_19/conv5/conv5_4',\n 'vgg_19/pool5',\n 'vgg_19/fc6',\n 'vgg_19/fc7',\n 'vgg_19/fc8'\n ]\n self.assertSetEqual(set(end_points.keys()), set(expected_names))\n\n def testModelVariables(self):\n batch_size = 5\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n inputs = tf.random_uniform((batch_size, height, width, 3))\n vgg.vgg_19(inputs, num_classes)\n expected_names = [\n 'vgg_19/conv1/conv1_1/weights',\n 'vgg_19/conv1/conv1_1/biases',\n 'vgg_19/conv1/conv1_2/weights',\n 'vgg_19/conv1/conv1_2/biases',\n 'vgg_19/conv2/conv2_1/weights',\n 'vgg_19/conv2/conv2_1/biases',\n 'vgg_19/conv2/conv2_2/weights',\n 'vgg_19/conv2/conv2_2/biases',\n 'vgg_19/conv3/conv3_1/weights',\n 'vgg_19/conv3/conv3_1/biases',\n 'vgg_19/conv3/conv3_2/weights',\n 'vgg_19/conv3/conv3_2/biases',\n 'vgg_19/conv3/conv3_3/weights',\n 'vgg_19/conv3/conv3_3/biases',\n 'vgg_19/conv3/conv3_4/weights',\n 'vgg_19/conv3/conv3_4/biases',\n 'vgg_19/conv4/conv4_1/weights',\n 'vgg_19/conv4/conv4_1/biases',\n 'vgg_19/conv4/conv4_2/weights',\n 'vgg_19/conv4/conv4_2/biases',\n 'vgg_19/conv4/conv4_3/weights',\n 'vgg_19/conv4/conv4_3/biases',\n 'vgg_19/conv4/conv4_4/weights',\n 'vgg_19/conv4/conv4_4/biases',\n 'vgg_19/conv5/conv5_1/weights',\n 'vgg_19/conv5/conv5_1/biases',\n 'vgg_19/conv5/conv5_2/weights',\n 'vgg_19/conv5/conv5_2/biases',\n 'vgg_19/conv5/conv5_3/weights',\n 'vgg_19/conv5/conv5_3/biases',\n 'vgg_19/conv5/conv5_4/weights',\n 'vgg_19/conv5/conv5_4/biases',\n 'vgg_19/fc6/weights',\n 'vgg_19/fc6/biases',\n 'vgg_19/fc7/weights',\n 'vgg_19/fc7/biases',\n 'vgg_19/fc8/weights',\n 'vgg_19/fc8/biases',\n ]\n model_variables = [v.op.name for v in slim.get_model_variables()]\n self.assertSetEqual(set(model_variables), set(expected_names))\n\n def testEvaluation(self):\n batch_size = 2\n height, width = 224, 224\n num_classes = 1000\n with self.test_session():\n eval_inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_19(eval_inputs, is_training=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [batch_size, num_classes])\n predictions = tf.argmax(logits, 1)\n self.assertListEqual(predictions.get_shape().as_list(), [batch_size])\n\n def testTrainEvalWithReuse(self):\n train_batch_size = 2\n eval_batch_size = 1\n train_height, train_width = 224, 224\n eval_height, eval_width = 256, 256\n num_classes = 1000\n with self.test_session():\n train_inputs = tf.random_uniform(\n (train_batch_size, train_height, train_width, 3))\n logits, _ = vgg.vgg_19(train_inputs)\n self.assertListEqual(logits.get_shape().as_list(),\n [train_batch_size, num_classes])\n tf.get_variable_scope().reuse_variables()\n eval_inputs = tf.random_uniform(\n (eval_batch_size, eval_height, eval_width, 3))\n logits, _ = vgg.vgg_19(eval_inputs, is_training=False,\n spatial_squeeze=False)\n self.assertListEqual(logits.get_shape().as_list(),\n [eval_batch_size, 2, 2, num_classes])\n logits = tf.reduce_mean(logits, [1, 2])\n predictions = tf.argmax(logits, 1)\n self.assertEquals(predictions.get_shape().as_list(), [eval_batch_size])\n\n def testForward(self):\n batch_size = 1\n height, width = 224, 224\n with self.test_session() as sess:\n inputs = tf.random_uniform((batch_size, height, width, 3))\n logits, _ = vgg.vgg_19(inputs)\n sess.run(tf.global_variables_initializer())\n output = sess.run(logits)\n self.assertTrue(output.any())\n\nif __name__ == '__main__':\n tf.test.main()\n"
},
{
"path": "run_attack.sh",
"content": "#!/bin/bash\n#\n# run_attack.sh is a script which executes the attack\n#\n# Envoronment which runs attacks and defences calls it in a following way:\n# run_attack.sh INPUT_DIR OUTPUT_DIR MAX_EPSILON\n# where:\n# INPUT_DIR - directory with input PNG images\n# OUTPUT_DIR - directory where adversarial images should be written\n# MAX_EPSILON - maximum allowed L_{\\infty} norm of adversarial perturbation\n#\n\nINPUT_DIR=$1\nOUTPUT_DIR=$2\nMAX_EPSILON=$3\n\npython attack_iter.py \\\n --input_dir=\"${INPUT_DIR}\" \\\n --output_dir=\"${OUTPUT_DIR}\" \\\n --max_epsilon=\"${MAX_EPSILON}\" \\\n --checkpoint_path_inception_v3=inception_v3.ckpt \\\n --checkpoint_path_inception_v4=inception_v4.ckpt \\\n --checkpoint_path_inception_resnet_v2=inception_resnet_v2_2016_08_30.ckpt \\\n --checkpoint_path_resnet=resnet_v2_152.ckpt \\\n --num_iter=10 \\\n --momentum=1.0 \\\n --prob=0.7\n"
}
]