[
{
"path": ".gitignore",
"content": "# Mac\n.DS_Store\n\n# Matlab\n*.m~\n*.asv\n\n# Vim\n*~\n\n# Python binary\n*.pyc\n\n# Temp dir\ntmp\n\n# Build and CMake\nbuild\n\n# openFrameworks\nbin\nobj\n"
},
{
"path": "LICENSE",
"content": "MIT License\n\nCopyright (c) 2016 Saiwen Wang\n\nPermission is hereby granted, free of charge, to any person obtaining a copy\nof this software and associated documentation files (the \"Software\"), to deal\nin the Software without restriction, including without limitation the rights\nto use, copy, modify, merge, publish, distribute, sublicense, and/or sell\ncopies of the Software, and to permit persons to whom the Software is\nfurnished to do so, subject to the following conditions:\n\nThe above copyright notice and this permission notice shall be included in all\ncopies or substantial portions of the Software.\n\nTHE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\nIMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\nFITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\nAUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\nLIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\nOUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\nSOFTWARE.\n"
},
{
"path": "README.md",
"content": "# deep-soli\n\nGesture Recognition Using Neural Networks with Google's Project Soli Sensor\n\n## Update\n\nDataset and trained model are now available.\n\n## Introduction\n\nThis is the open source evaluation code base of our paper:\n\n**Interacting with Soli: Exploring Fine-Grained Dynamic Gesture Recognition\nin the Radio-Frequency Spectrum**
\nSaiwen Wang, Jie Song, Jamie Lien, Poupyrev Ivan, Otmar Hilliges
\n(link to the [paper](http://bit.ly/2ftSRcn))\n\nPlease cite the paper if you find the code useful. Thank you.\n([BibTex](https://github.com/simonwsw/deep-soli/blob/master/cite.bib))\n\nThis project uses Google's [Project Soli](atap.google.com/soli) sensor.\n\n> Soli is a new sensing technology that uses miniature radar to\n> detect touchless gesture interactions.\n>\n> \n>\n> Soli sensor technology works by emitting electromagnetic waves in a\n> broad beam. Objects within the beam scatter this energy, reflecting\n> some portion back towards the radar antenna. Properties of the\n> reflected signal, such as energy, time delay, and frequency shift\n> capture rich information about the object’s characteristics and\n> dynamics, including size, shape, orientation, material, distance,\n> and velocity.\n\nOur paper uses a light-weight end-to-end trained Convolutional Neural Networks\nand Recurrent Neural Networks architecture, recognizes 11 in-air gestures\nwith 87% per-frame accuracy, and can perform realtime predictions at 140Hz\non commodity hardware. (link to the [paper video](http://bit.ly/2fDd9iJ))\n\n## Pre-request\n\n- Python 2: HDF5, OpenCV 2 interfaces for python.\n- C++: HDF5, OpenCV 2, Boost\n- Lua JIT and Torch 7.\n- Torch 7 packages: `class`, GPU support `cunn` and `cutorch`, Matlab\n support `mattorch`, JSON support `lunajson`, Torch image library `image`\n- Please note that `mattorch` is an outdated packages which is no\n longer maintained.\n\n## Quick start\n\n- Make image binary\n\n```\ncd image\nmkdir bin\nmake\n```\n\n- Preprocessing (HDF5 to images):\n\n```\npython pre/main.py --op image --file [dataset folder]\n--target [target image folder] --channel 4 --originsize 32 --outsize 32\n```\n\n- Preprocessing (generate mean file):\n\n```\npython pre/main.py --op mean --file [image folder]\n--target [mean file name] --channel 4 --outsize 32\n```\n\n- Load model and evaluate:\n\n```\nth net/main.lua --file [image folder] --list [train/test sequence split file]\n--load [model file] --inputsize 32 --inputch 4 --label 13 --datasize 32\n--datach 4 --batch 16 --maxseq 40 --cuda --cudnn\n```\n\n## Dataset\n\n- Download [dataset](https://polybox.ethz.ch/index.php/s/wG93iTUdvRU8EaT)\n (please let me know when the link doesn't work).\n- Train/test split file (in JSON format) we used is stored in the repo\n `config/file_half.json`.\n- The dataset contains multiple preprocessed Range-Doppler Image sequences.\n Each sequence is saved as a single HDF5 format data file. File names are\n defined as `[gesture ID]_[session ID]_[instance ID].h5`. Range-Doppler Image\n data of a specific channel can be accessed by dataset name `ch[channel ID]`.\n Label can be accessed by dataset name `label`. Range-Doppler Image\n data array has shape of `[number of frame] * 1024` (can be reshape back to 2D Range-Doppler Image to `32 * 32`)\n- Simple Python code to access the data:\n\n```python\n# Demo code to extract data in python\nimport h5py\n\nuse_channel = 0\nwith h5py.File(file_name, 'r') as f:\n # Data and label are numpy arrays\n data = f['ch{}'.format(use_channel)][()]\n label = f['label'][()]\n```\n\n- Dataset session arrangement for evaluation.\n - 11 (gestures) * 25 (instances) * 10 (users) for cross user evaluation:\n session 2 (25), 3 (25), 5 (25), 6 (25), 8 (25), 9 (25), 10 (25), 11 (25),\n 12 (25), 13 (25).\n - 11 (gestures) * (50 (instances) * 4 (sessions) +\n 25 (instances) * 2 (sessions)) for single user\n cross session evaluation: session 0 (50), 1 (50), 4 (50), 7 (50),\n 13 (25), 14 (25).\n - Please refer to the paper for the gesture collecting\n campaign details.\n- The gestures are listed in the table below. Each column represents\n one gesture and we snapshot three important steps for each gestures.\n The gesture label is indicated by the number in the circle above. Please\n notice that the gesture label order is different than the paper, as\n we regroup gestures in the paper. Sequences with gesture ID 11 are\n background signals with no presence of hand.\n\n\n\n## Pre-trained model\n\n- Download [model](https://polybox.ethz.ch/index.php/s/0SEdZqkn433dbEh)\n- Trained proposed model, please refer to the paper for model detail.\n- Simple Lua (Torch 7) code to load the model:\n\n```\nloadFile = 'uni_image_np_50.t7'\nnet = torch.load(loadFile)\nprint(net)\n```\n\n- The model uses layers support `cudnn`.\n\n## Comments on the code base\n\nThis is a simplified version of the original code base we used for all the\nexperiments in the paper. The complex Torch based class hierarchies in\nthe `net` reflects varies model architectures we tried during the\nexperiments. For simplicity, we only make the evaluation part public.\nThe model detail can be found both in the paper and the model file.\n\n## License\n\nThis project is licensed under the terms of the MIT license.\n"
},
{
"path": "cite.bib",
"content": "@inproceedings{wang2016interacting,\n title={Interacting with soli: Exploring fine-grained dynamic gesture recognition in the radio-frequency spectrum},\n author={Wang, Saiwen and Song, Jie and Lien, Jaime and Poupyrev, Ivan and Hilliges, Otmar},\n booktitle={Proceedings of the 29th Annual Symposium on User Interface Software and Technology},\n pages={851--860},\n year={2016},\n organization={ACM}\n}\n"
},
{
"path": "config/file_half.json",
"content": "{\"train\": [\"0_8_4\", \"0_3_1\", \"0_2_0\", \"0_13_5\", \"0_2_19\", \"0_3_2\", \"0_12_1\", \"0_9_7\", \"0_3_22\", \"0_2_2\", \"0_2_3\", \"0_5_8\", \"0_8_21\", \"0_9_4\", \"0_2_17\", \"0_13_13\", \"0_2_22\", \"0_9_15\", \"0_9_17\", \"0_11_14\", \"0_10_14\", \"0_2_16\", \"0_6_7\", \"0_2_10\", \"0_12_19\", \"0_8_24\", \"0_8_3\", \"0_11_4\", \"0_2_11\", \"0_3_3\", \"0_12_14\", \"0_5_7\", \"0_11_3\", \"0_10_1\", \"0_9_14\", \"0_11_23\", \"0_12_13\", \"0_8_19\", \"0_8_16\", \"0_10_16\", \"0_9_2\", \"0_8_13\", \"0_3_9\", \"0_9_12\", \"0_13_12\", \"0_5_15\", \"0_6_15\", \"0_6_0\", \"0_5_0\", \"0_10_7\", \"0_9_3\", \"0_3_11\", \"0_3_6\", \"0_13_18\", \"0_6_19\", \"0_9_5\", \"0_10_11\", \"0_12_7\", \"0_12_16\", \"0_11_16\", \"0_8_14\", \"0_11_20\", \"0_10_0\", \"0_10_13\", \"0_6_17\", \"0_6_11\", \"0_11_17\", \"0_10_2\", \"0_11_2\", \"0_12_18\", \"0_12_4\", \"0_11_9\", \"0_2_14\", \"0_6_6\", \"0_5_9\", \"0_10_8\", \"0_8_23\", \"0_11_0\", \"0_3_20\", \"0_11_7\", \"0_11_13\", \"0_2_7\", \"0_11_5\", \"0_8_12\", \"0_10_21\", \"0_2_9\", \"0_12_5\", \"0_3_23\", \"0_3_19\", \"0_8_7\", \"0_10_10\", \"0_5_2\", \"0_13_22\", \"0_8_5\", \"0_11_22\", \"0_11_24\", \"0_9_24\", \"0_3_12\", \"0_2_6\", \"0_8_20\", \"0_5_18\", \"0_5_3\", \"0_3_8\", \"0_9_9\", \"0_11_19\", \"0_13_23\", \"0_6_5\", \"0_11_18\", \"0_3_13\", \"0_3_16\", \"0_9_1\", \"0_3_4\", \"0_5_6\", \"0_6_22\", \"0_13_20\", \"0_9_10\", \"0_3_17\", \"0_6_10\", \"0_2_24\", \"0_10_3\", \"0_5_5\", \"0_6_18\", \"0_6_4\", \"0_2_1\", \"0_3_14\", \"0_10_5\", \"1_8_3\", \"1_2_24\", \"1_2_16\", \"1_9_12\", \"1_6_14\", \"1_13_24\", \"1_3_6\", \"1_3_1\", \"1_9_7\", \"1_3_24\", \"1_5_14\", \"1_9_6\", \"1_13_6\", \"1_13_3\", \"1_2_6\", \"1_6_10\", \"1_10_10\", \"1_2_12\", \"1_3_4\", \"1_5_12\", \"1_12_7\", \"1_12_24\", \"1_5_21\", \"1_12_12\", \"1_11_9\", \"1_11_10\", \"1_13_10\", \"1_11_19\", \"1_6_4\", \"1_5_1\", \"1_3_16\", \"1_5_6\", \"1_6_15\", \"1_5_18\", \"1_2_13\", \"1_9_21\", \"1_13_19\", \"1_3_0\", \"1_13_1\", \"1_10_12\", \"1_12_5\", \"1_9_13\", \"1_11_22\", \"1_3_10\", \"1_3_14\", \"1_8_11\", \"1_5_15\", \"1_6_2\", \"1_5_9\", \"1_9_24\", \"1_8_10\", \"1_5_10\", \"1_6_0\", \"1_10_8\", \"1_10_2\", \"1_8_19\", \"1_13_5\", \"1_12_11\", \"1_12_22\", \"1_9_17\", \"1_2_18\", \"1_11_4\", \"1_2_0\", \"1_5_22\", \"1_3_15\", \"1_10_20\", \"1_2_5\", \"1_2_8\", \"1_13_2\", \"1_10_5\", \"1_11_3\", \"1_10_24\", \"1_10_21\", \"1_6_11\", \"1_11_14\", \"1_9_1\", \"1_8_22\", \"1_13_21\", \"1_10_14\", \"1_9_19\", \"1_6_16\", \"1_2_10\", \"1_3_21\", \"1_8_14\", \"1_10_1\", \"1_12_14\", \"1_2_17\", \"1_5_19\", \"1_13_9\", \"1_8_12\", \"1_2_11\", \"1_13_11\", \"1_3_17\", \"1_11_17\", \"1_12_2\", \"1_12_15\", \"1_3_9\", \"1_10_6\", \"1_11_15\", \"1_2_23\", \"1_2_14\", \"1_8_16\", \"1_6_3\", \"1_3_5\", \"1_11_24\", \"1_13_14\", \"1_13_20\", \"1_3_2\", \"1_10_18\", \"1_9_16\", \"1_9_23\", \"1_8_2\", \"1_12_18\", \"1_3_8\", \"1_13_7\", \"1_2_4\", \"1_2_22\", \"1_5_8\", \"1_6_24\", \"1_2_9\", \"1_2_2\", \"1_11_12\", \"1_11_21\", \"1_11_8\", \"1_8_5\", \"1_11_18\", \"2_6_6\", \"2_9_20\", \"2_2_7\", \"2_3_16\", \"2_8_24\", \"2_8_9\", \"2_8_21\", \"2_11_17\", \"2_8_23\", \"2_9_8\", \"2_3_8\", \"2_2_11\", \"2_8_18\", \"2_13_11\", \"2_9_5\", \"2_11_14\", \"2_13_9\", \"2_12_14\", \"2_8_16\", \"2_10_21\", \"2_2_13\", \"2_13_18\", \"2_2_23\", \"2_12_15\", \"2_8_12\", \"2_11_0\", \"2_5_5\", \"2_5_19\", \"2_8_13\", \"2_5_12\", \"2_13_21\", \"2_3_13\", \"2_9_22\", \"2_2_9\", \"2_10_20\", \"2_6_2\", \"2_5_3\", \"2_10_17\", \"2_11_24\", \"2_10_12\", \"2_11_16\", \"2_11_5\", \"2_3_0\", \"2_12_19\", \"2_3_2\", \"2_10_0\", \"2_10_18\", \"2_13_2\", \"2_12_17\", \"2_12_11\", \"2_13_10\", \"2_5_17\", \"2_13_7\", \"2_11_8\", \"2_5_6\", \"2_13_20\", \"2_8_7\", \"2_6_7\", \"2_12_22\", \"2_6_16\", \"2_5_21\", \"2_3_5\", \"2_5_11\", \"2_13_17\", \"2_3_1\", \"2_6_12\", \"2_10_24\", \"2_3_3\", \"2_8_2\", \"2_10_23\", \"2_3_7\", \"2_5_1\", \"2_8_11\", \"2_5_15\", \"2_6_23\", \"2_2_5\", \"2_12_23\", \"2_5_23\", \"2_10_22\", \"2_2_3\", \"2_13_1\", \"2_9_7\", \"2_12_10\", \"2_12_9\", \"2_11_6\", \"2_2_18\", \"2_12_2\", \"2_10_10\", \"2_2_2\", \"2_11_2\", \"2_3_20\", \"2_13_6\", \"2_12_5\", \"2_11_18\", \"2_6_14\", \"2_5_20\", \"2_5_8\", \"2_8_3\", \"2_10_5\", \"2_8_4\", \"2_9_17\", \"2_2_14\", \"2_10_19\", \"2_2_22\", \"2_2_12\", \"2_11_21\", \"2_12_7\", \"2_6_5\", \"2_6_13\", \"2_12_24\", \"2_5_14\", \"2_6_8\", \"2_8_0\", \"2_8_1\", \"2_10_7\", \"2_12_3\", \"2_3_22\", \"2_9_24\", \"2_3_23\", \"2_12_20\", \"2_10_11\", \"2_3_19\", \"2_5_13\", \"2_11_9\", \"2_9_23\", \"2_2_20\", \"3_5_21\", \"3_8_14\", \"3_9_1\", \"3_10_21\", \"3_9_14\", \"3_12_1\", \"3_8_24\", \"3_3_4\", \"3_12_3\", \"3_9_24\", \"3_13_4\", \"3_2_2\", \"3_12_14\", \"3_8_4\", \"3_6_22\", \"3_10_16\", \"3_10_2\", \"3_5_4\", \"3_3_13\", \"3_2_21\", \"3_11_0\", \"3_6_16\", \"3_3_2\", \"3_8_12\", \"3_12_18\", \"3_5_2\", \"3_6_12\", \"3_11_13\", \"3_10_6\", \"3_6_8\", \"3_6_21\", \"3_10_19\", \"3_11_2\", \"3_3_5\", \"3_12_19\", \"3_13_18\", \"3_12_12\", \"3_13_22\", \"3_3_22\", \"3_10_8\", \"3_6_17\", \"3_11_10\", \"3_2_10\", \"3_5_14\", \"3_6_18\", \"3_2_9\", \"3_5_22\", \"3_9_3\", \"3_3_8\", \"3_12_20\", \"3_2_12\", \"3_10_1\", \"3_13_9\", \"3_13_12\", \"3_11_20\", \"3_12_10\", \"3_6_24\", \"3_13_2\", \"3_5_6\", \"3_12_21\", \"3_10_17\", \"3_10_18\", \"3_9_23\", \"3_3_12\", \"3_12_11\", \"3_8_17\", \"3_11_6\", \"3_8_8\", \"3_11_7\", \"3_3_19\", \"3_10_22\", \"3_10_14\", \"3_10_20\", \"3_13_15\", \"3_5_8\", \"3_3_10\", \"3_8_20\", \"3_11_9\", \"3_9_13\", \"3_6_11\", \"3_6_19\", \"3_11_18\", \"3_13_5\", \"3_8_9\", \"3_13_21\", \"3_9_21\", \"3_2_3\", \"3_9_7\", \"3_2_11\", \"3_3_20\", \"3_5_15\", \"3_12_7\", \"3_5_5\", \"3_3_17\", \"3_10_13\", \"3_11_19\", \"3_9_4\", \"3_3_14\", \"3_10_7\", \"3_9_11\", \"3_10_11\", \"3_10_10\", \"3_13_17\", \"3_5_19\", \"3_2_23\", \"3_6_5\", \"3_3_11\", \"3_13_14\", \"3_8_10\", \"3_5_11\", \"3_9_22\", \"3_13_13\", \"3_3_16\", \"3_11_5\", \"3_3_18\", \"3_6_23\", \"3_5_17\", \"3_13_1\", \"3_9_20\", \"3_5_24\", \"3_12_22\", \"3_6_6\", \"3_12_16\", \"3_12_15\", \"3_10_3\", \"3_5_16\", \"4_3_14\", \"4_3_7\", \"4_3_20\", \"4_3_5\", \"4_5_8\", \"4_2_15\", \"4_6_10\", \"4_13_9\", \"4_13_23\", \"4_10_16\", \"4_5_16\", \"4_10_3\", \"4_3_1\", \"4_10_24\", \"4_2_13\", \"4_5_11\", \"4_6_23\", \"4_9_1\", \"4_10_23\", \"4_6_20\", \"4_2_22\", \"4_11_18\", \"4_5_15\", \"4_9_22\", \"4_10_20\", \"4_11_7\", \"4_12_4\", \"4_3_0\", \"4_9_15\", \"4_9_17\", \"4_9_5\", \"4_9_21\", \"4_9_4\", \"4_13_24\", \"4_9_24\", \"4_10_11\", \"4_3_9\", \"4_6_2\", \"4_9_6\", \"4_3_3\", \"4_8_15\", \"4_3_18\", \"4_6_14\", \"4_13_11\", \"4_8_19\", \"4_5_4\", \"4_3_16\", \"4_6_6\", \"4_5_7\", \"4_8_14\", \"4_11_12\", \"4_12_14\", \"4_11_22\", \"4_11_17\", \"4_8_24\", \"4_6_1\", \"4_13_20\", \"4_12_21\", \"4_8_7\", \"4_3_22\", \"4_12_17\", \"4_10_9\", \"4_11_19\", \"4_3_17\", \"4_2_16\", \"4_10_19\", \"4_13_21\", \"4_9_20\", \"4_8_10\", \"4_5_2\", \"4_9_3\", \"4_9_16\", \"4_2_7\", \"4_13_3\", \"4_10_0\", \"4_8_22\", \"4_11_23\", \"4_5_0\", \"4_6_0\", \"4_11_6\", \"4_9_10\", \"4_2_6\", \"4_9_18\", \"4_8_23\", \"4_5_19\", \"4_2_14\", \"4_12_19\", \"4_11_4\", \"4_8_13\", \"4_8_2\", \"4_5_18\", \"4_13_15\", \"4_13_19\", \"4_8_5\", \"4_11_3\", \"4_3_19\", \"4_2_23\", \"4_2_8\", \"4_3_12\", \"4_6_18\", \"4_12_9\", \"4_2_4\", \"4_11_0\", \"4_12_5\", \"4_12_23\", \"4_6_15\", \"4_11_5\", \"4_3_13\", \"4_5_24\", \"4_12_18\", \"4_13_7\", \"4_5_12\", \"4_8_9\", \"4_6_8\", \"4_2_3\", \"4_10_17\", \"4_5_1\", \"4_6_24\", \"4_9_14\", \"4_3_8\", \"4_9_19\", \"4_10_18\", \"4_5_14\", \"4_10_1\", \"4_9_23\", \"4_12_13\", \"5_10_0\", \"5_3_24\", \"5_12_1\", \"5_12_4\", \"5_2_10\", \"5_8_22\", \"5_2_21\", \"5_9_7\", \"5_6_21\", \"5_12_14\", \"5_13_3\", \"5_12_5\", \"5_11_8\", \"5_10_11\", \"5_9_20\", \"5_11_1\", \"5_8_1\", \"5_12_9\", \"5_12_22\", \"5_5_17\", \"5_6_15\", \"5_9_23\", \"5_8_16\", \"5_6_19\", \"5_2_11\", \"5_13_4\", \"5_11_5\", \"5_3_20\", \"5_13_0\", \"5_8_13\", \"5_6_5\", \"5_2_19\", \"5_9_3\", \"5_5_10\", \"5_5_0\", \"5_9_16\", \"5_6_0\", \"5_2_15\", \"5_5_12\", \"5_2_4\", \"5_2_7\", \"5_3_2\", \"5_3_11\", \"5_11_11\", \"5_3_8\", \"5_11_13\", \"5_13_21\", \"5_5_15\", \"5_12_24\", \"5_3_7\", \"5_3_3\", \"5_5_21\", \"5_13_13\", \"5_10_9\", \"5_13_6\", \"5_9_17\", \"5_5_9\", \"5_10_20\", \"5_12_17\", \"5_11_18\", \"5_6_16\", \"5_9_11\", \"5_2_18\", \"5_2_9\", \"5_8_18\", \"5_13_18\", \"5_9_5\", \"5_3_18\", \"5_2_13\", \"5_12_3\", \"5_10_16\", \"5_3_13\", \"5_2_17\", \"5_10_8\", \"5_5_8\", \"5_13_2\", \"5_13_14\", \"5_13_5\", \"5_12_6\", \"5_6_1\", \"5_11_4\", \"5_3_5\", \"5_12_0\", \"5_6_2\", \"5_12_18\", \"5_6_13\", \"5_13_1\", \"5_9_13\", \"5_10_24\", \"5_3_6\", \"5_5_23\", \"5_12_15\", \"5_12_7\", \"5_13_9\", \"5_2_6\", \"5_10_14\", \"5_8_11\", \"5_10_15\", \"5_6_3\", \"5_9_0\", \"5_9_22\", \"5_8_10\", \"5_8_12\", \"5_8_19\", \"5_2_3\", \"5_2_2\", \"5_13_8\", \"5_10_2\", \"5_9_6\", \"5_8_4\", \"5_10_6\", \"5_6_14\", \"5_9_10\", \"5_3_21\", \"5_11_12\", \"5_8_5\", \"5_10_22\", \"5_13_12\", \"5_8_23\", \"5_5_4\", \"5_10_13\", \"5_5_19\", \"5_8_8\", \"5_6_17\", \"5_3_17\", \"5_5_11\", \"6_11_2\", \"6_11_21\", \"6_3_5\", \"6_6_19\", \"6_12_4\", \"6_3_24\", \"6_13_0\", \"6_6_16\", \"6_3_18\", \"6_3_13\", \"6_9_20\", \"6_9_24\", \"6_9_21\", \"6_2_12\", \"6_13_20\", \"6_6_22\", \"6_9_13\", \"6_8_3\", \"6_2_20\", \"6_2_4\", \"6_10_0\", \"6_9_6\", \"6_10_13\", \"6_3_11\", \"6_11_11\", \"6_8_13\", \"6_8_8\", \"6_9_9\", \"6_12_10\", \"6_10_7\", \"6_12_16\", \"6_3_14\", \"6_8_17\", \"6_6_11\", \"6_9_2\", \"6_8_0\", \"6_10_1\", \"6_5_5\", \"6_3_4\", \"6_2_16\", \"6_8_2\", \"6_6_2\", \"6_10_8\", \"6_9_3\", \"6_9_14\", \"6_10_12\", \"6_2_1\", \"6_9_0\", \"6_2_15\", \"6_9_15\", \"6_10_14\", \"6_5_13\", \"6_3_9\", \"6_12_24\", \"6_3_2\", \"6_6_10\", \"6_5_20\", \"6_9_22\", \"6_6_9\", \"6_9_1\", \"6_6_24\", \"6_13_7\", \"6_5_2\", \"6_10_20\", \"6_12_9\", \"6_10_3\", \"6_9_8\", \"6_2_8\", \"6_12_23\", \"6_10_5\", \"6_8_15\", \"6_13_12\", \"6_9_16\", \"6_10_10\", \"6_5_14\", \"6_13_1\", \"6_11_3\", \"6_13_13\", \"6_3_20\", \"6_9_11\", \"6_5_12\", \"6_12_11\", \"6_9_18\", \"6_6_21\", \"6_8_12\", \"6_5_10\", \"6_8_23\", \"6_3_6\", \"6_9_12\", \"6_8_9\", \"6_5_1\", \"6_6_20\", \"6_11_19\", \"6_10_19\", \"6_8_20\", \"6_5_11\", \"6_5_8\", \"6_9_4\", \"6_2_21\", \"6_11_1\", \"6_6_13\", \"6_10_15\", \"6_2_7\", \"6_8_14\", \"6_3_3\", \"6_9_10\", \"6_5_0\", \"6_6_5\", \"6_3_0\", \"6_10_4\", \"6_6_23\", \"6_12_12\", \"6_3_15\", \"6_8_11\", \"6_10_21\", \"6_3_10\", \"6_13_5\", \"6_6_15\", \"6_12_6\", \"6_10_9\", \"6_11_23\", \"6_5_4\", \"6_6_7\", \"6_8_21\", \"6_12_21\", \"6_8_18\", \"7_12_6\", \"7_8_1\", \"7_11_17\", \"7_2_3\", \"7_10_14\", \"7_6_9\", \"7_5_9\", \"7_6_23\", \"7_5_14\", \"7_3_13\", \"7_2_5\", \"7_13_20\", \"7_10_24\", \"7_13_21\", \"7_5_20\", \"7_8_20\", \"7_12_18\", \"7_8_18\", \"7_11_16\", \"7_3_17\", \"7_10_21\", \"7_5_13\", \"7_2_13\", \"7_11_3\", \"7_5_4\", \"7_2_20\", \"7_9_10\", \"7_6_24\", \"7_13_15\", \"7_2_21\", \"7_2_4\", \"7_8_7\", \"7_13_9\", \"7_8_4\", \"7_5_5\", \"7_13_8\", \"7_6_2\", \"7_13_10\", \"7_9_13\", \"7_8_23\", \"7_9_11\", \"7_8_11\", \"7_9_19\", \"7_3_12\", \"7_10_6\", \"7_8_10\", \"7_3_3\", \"7_11_15\", \"7_3_7\", \"7_6_14\", \"7_2_9\", \"7_11_13\", \"7_12_7\", \"7_3_10\", \"7_5_8\", \"7_2_24\", \"7_10_12\", \"7_11_7\", \"7_11_10\", \"7_13_16\", \"7_10_1\", \"7_6_15\", \"7_8_5\", \"7_3_20\", \"7_6_0\", \"7_9_8\", \"7_12_17\", \"7_12_24\", \"7_2_0\", \"7_10_20\", \"7_10_2\", \"7_9_1\", \"7_11_5\", \"7_12_16\", \"7_11_12\", \"7_9_9\", \"7_3_2\", \"7_6_21\", \"7_13_5\", \"7_10_23\", \"7_11_14\", \"7_9_3\", \"7_2_15\", \"7_6_7\", \"7_2_17\", \"7_6_3\", \"7_5_22\", \"7_9_21\", \"7_12_23\", \"7_10_18\", \"7_3_14\", \"7_8_2\", \"7_5_23\", \"7_12_5\", \"7_13_23\", \"7_2_16\", \"7_12_22\", \"7_9_4\", \"7_13_17\", \"7_6_4\", \"7_10_11\", \"7_10_7\", \"7_3_1\", \"7_5_15\", \"7_9_14\", \"7_2_1\", \"7_9_12\", \"7_5_11\", \"7_11_9\", \"7_6_20\", \"7_6_19\", \"7_12_0\", \"7_13_2\", \"7_6_1\", \"7_13_11\", \"7_11_21\", \"7_8_19\", \"7_11_23\", \"7_8_13\", \"7_12_20\", \"7_2_22\", \"7_11_22\", \"7_10_22\", \"7_9_0\", \"7_3_19\", \"7_3_5\", \"8_11_7\", \"8_3_20\", \"8_13_6\", \"8_2_5\", \"8_13_16\", \"8_13_2\", \"8_10_14\", \"8_13_20\", \"8_8_0\", \"8_2_20\", \"8_2_24\", \"8_2_7\", \"8_11_2\", \"8_2_12\", \"8_2_14\", \"8_2_17\", \"8_10_11\", \"8_6_22\", \"8_10_19\", \"8_12_17\", \"8_2_11\", \"8_9_9\", \"8_9_0\", \"8_11_11\", \"8_13_0\", \"8_11_6\", \"8_5_24\", \"8_10_5\", \"8_5_9\", \"8_13_4\", \"8_3_2\", \"8_6_16\", \"8_13_5\", \"8_8_11\", \"8_5_19\", \"8_10_7\", \"8_12_13\", \"8_13_14\", \"8_8_3\", \"8_12_18\", \"8_3_16\", \"8_12_23\", \"8_10_8\", \"8_11_17\", \"8_8_16\", \"8_3_9\", \"8_8_22\", \"8_11_3\", \"8_10_24\", \"8_3_23\", \"8_6_13\", \"8_5_17\", \"8_2_4\", \"8_13_21\", \"8_12_1\", \"8_10_17\", \"8_6_2\", \"8_6_7\", \"8_11_24\", \"8_3_10\", \"8_12_20\", \"8_3_3\", \"8_12_14\", \"8_5_6\", \"8_8_18\", \"8_11_13\", \"8_2_23\", \"8_2_1\", \"8_12_12\", \"8_8_6\", \"8_12_16\", \"8_5_4\", \"8_3_6\", \"8_12_2\", \"8_10_3\", \"8_5_23\", \"8_12_4\", \"8_11_12\", \"8_6_5\", \"8_13_24\", \"8_3_19\", \"8_10_18\", \"8_12_21\", \"8_9_16\", \"8_9_3\", \"8_3_22\", \"8_8_9\", \"8_3_17\", \"8_9_7\", \"8_5_22\", \"8_8_10\", \"8_9_11\", \"8_12_3\", \"8_9_22\", \"8_2_13\", \"8_6_23\", \"8_6_20\", \"8_6_18\", \"8_12_7\", \"8_10_0\", \"8_8_7\", \"8_12_19\", \"8_8_20\", \"8_12_9\", \"8_9_14\", \"8_12_15\", \"8_3_4\", \"8_11_1\", \"8_6_14\", \"8_6_15\", \"8_9_13\", \"8_10_10\", \"8_3_12\", \"8_5_12\", \"8_2_21\", \"8_13_7\", \"8_10_20\", \"8_6_3\", \"8_8_8\", \"8_6_12\", \"8_5_3\", \"8_6_8\", \"8_11_0\", \"8_8_19\", \"8_11_19\", \"8_9_5\", \"9_8_3\", \"9_9_18\", \"9_8_13\", \"9_9_16\", \"9_8_5\", \"9_11_23\", \"9_6_1\", \"9_9_17\", \"9_3_20\", \"9_2_23\", \"9_9_10\", \"9_12_21\", \"9_5_21\", \"9_10_10\", \"9_8_22\", \"9_6_7\", \"9_3_23\", \"9_9_21\", \"9_11_8\", \"9_2_17\", \"9_9_12\", \"9_11_15\", \"9_11_6\", \"9_2_21\", \"9_2_0\", \"9_13_1\", \"9_9_20\", \"9_5_22\", \"9_12_20\", \"9_2_14\", \"9_10_23\", \"9_5_7\", \"9_12_0\", \"9_3_11\", \"9_10_5\", \"9_2_16\", \"9_5_12\", \"9_8_16\", \"9_13_9\", \"9_12_15\", \"9_12_24\", \"9_5_14\", \"9_9_13\", \"9_13_0\", \"9_6_4\", \"9_9_23\", \"9_13_19\", \"9_13_12\", \"9_9_4\", \"9_9_5\", \"9_8_12\", \"9_13_23\", \"9_5_9\", \"9_8_14\", \"9_13_4\", \"9_3_10\", \"9_5_11\", \"9_3_9\", \"9_3_3\", \"9_11_14\", \"9_2_22\", \"9_9_15\", \"9_13_3\", \"9_10_19\", \"9_11_12\", \"9_12_22\", \"9_8_1\", \"9_11_5\", \"9_10_16\", \"9_3_21\", \"9_10_18\", \"9_12_17\", \"9_10_15\", \"9_3_7\", \"9_5_18\", \"9_5_19\", \"9_2_4\", \"9_12_3\", \"9_3_17\", \"9_2_5\", \"9_13_24\", \"9_6_17\", \"9_6_11\", \"9_11_21\", \"9_10_0\", \"9_11_18\", \"9_10_11\", \"9_8_24\", \"9_12_18\", \"9_10_2\", \"9_2_8\", \"9_9_3\", \"9_11_9\", \"9_5_23\", \"9_5_15\", \"9_12_6\", \"9_9_22\", \"9_2_10\", \"9_2_3\", \"9_10_12\", \"9_11_17\", \"9_3_14\", \"9_13_14\", \"9_8_0\", \"9_3_8\", \"9_5_1\", \"9_13_17\", \"9_12_12\", \"9_6_19\", \"9_3_15\", \"9_8_11\", \"9_3_4\", \"9_10_21\", \"9_6_21\", \"9_10_9\", \"9_8_2\", \"9_6_10\", \"9_8_4\", \"9_2_13\", \"9_3_1\", \"9_6_20\", \"9_10_14\", \"9_8_7\", \"9_3_13\", \"9_5_0\", \"9_5_17\", \"10_10_16\", \"10_10_7\", \"10_3_23\", \"10_5_0\", \"10_6_16\", \"10_10_11\", \"10_6_2\", \"10_5_9\", \"10_12_7\", \"10_10_19\", \"10_13_15\", \"10_10_3\", \"10_6_7\", \"10_11_17\", \"10_9_0\", \"10_6_5\", \"10_3_11\", \"10_13_0\", \"10_13_2\", \"10_10_13\", \"10_11_20\", \"10_3_19\", \"10_13_14\", \"10_5_15\", \"10_9_19\", \"10_9_22\", \"10_11_3\", \"10_2_1\", \"10_6_13\", \"10_12_11\", \"10_9_20\", \"10_8_6\", \"10_10_20\", \"10_8_2\", \"10_8_19\", \"10_9_14\", \"10_2_14\", \"10_13_18\", \"10_9_16\", \"10_2_13\", \"10_9_7\", \"10_3_1\", \"10_9_4\", \"10_2_15\", \"10_9_6\", \"10_8_4\", \"10_12_13\", \"10_13_10\", \"10_12_1\", \"10_12_2\", \"10_13_13\", \"10_8_0\", \"10_10_6\", \"10_12_20\", \"10_5_16\", \"10_2_20\", \"10_6_8\", \"10_12_15\", \"10_2_2\", \"10_9_1\", \"10_13_19\", \"10_9_15\", \"10_8_20\", \"10_8_22\", \"10_2_8\", \"10_2_7\", \"10_11_23\", \"10_10_1\", \"10_9_17\", \"10_10_5\", \"10_10_18\", \"10_11_8\", \"10_6_17\", \"10_11_12\", \"10_3_15\", \"10_13_20\", \"10_3_17\", \"10_11_0\", \"10_2_24\", \"10_12_12\", \"10_12_10\", \"10_13_4\", \"10_8_16\", \"10_8_11\", \"10_8_3\", \"10_6_18\", \"10_8_10\", \"10_5_14\", \"10_12_17\", \"10_6_23\", \"10_3_13\", \"10_13_12\", \"10_5_19\", \"10_6_20\", \"10_11_21\", \"10_11_7\", \"10_12_4\", \"10_13_5\", \"10_2_12\", \"10_11_6\", \"10_10_15\", \"10_2_19\", \"10_10_22\", \"10_13_9\", \"10_5_5\", \"10_5_1\", \"10_5_24\", \"10_2_9\", \"10_6_22\", \"10_12_16\", \"10_5_6\", \"10_13_11\", \"10_13_21\", \"10_10_0\", \"10_9_10\", \"10_10_8\", \"10_9_3\", \"10_5_12\", \"10_12_19\", \"10_8_9\", \"10_3_12\", \"10_11_11\", \"10_11_16\", \"10_5_4\", \"10_6_19\", \"10_3_7\"], \"eval\": [\"0_12_20\", \"0_10_18\", \"0_6_8\", \"0_3_0\", \"0_12_22\", \"0_13_10\", \"0_12_3\", \"0_5_14\", \"0_10_15\", \"0_2_5\", \"0_3_5\", \"0_3_18\", \"0_12_21\", \"0_5_21\", \"0_10_20\", \"0_8_22\", \"0_8_11\", \"0_13_6\", \"0_9_0\", \"0_3_7\", \"0_8_8\", \"0_5_10\", \"0_10_6\", \"0_9_20\", \"0_6_20\", \"0_13_15\", \"0_6_2\", \"0_9_13\", \"0_13_17\", \"0_12_23\", \"0_2_4\", \"0_12_10\", \"0_11_12\", \"0_13_21\", \"0_13_9\", \"0_12_6\", \"0_5_12\", \"0_9_16\", \"0_2_18\", \"0_13_4\", \"0_6_1\", \"0_10_12\", \"0_11_1\", \"0_11_6\", \"0_13_16\", \"0_2_13\", \"0_2_12\", \"0_5_16\", \"0_8_10\", \"0_13_19\", \"0_12_2\", \"0_12_17\", \"0_9_19\", \"0_6_3\", \"0_13_2\", \"0_9_6\", \"0_5_13\", \"0_12_9\", \"0_6_21\", \"0_5_19\", \"0_10_9\", \"0_9_23\", \"0_2_15\", \"0_9_22\", \"0_3_15\", \"0_11_21\", \"0_8_18\", \"0_12_12\", \"0_12_8\", \"0_5_22\", \"0_10_23\", \"0_5_17\", \"0_13_24\", \"0_6_12\", \"0_6_13\", \"0_10_24\", \"0_8_0\", \"0_5_24\", \"0_12_0\", \"0_8_6\", \"0_8_15\", \"0_5_11\", \"0_2_20\", \"0_8_2\", \"0_6_16\", \"0_9_21\", \"0_6_14\", \"0_3_21\", \"0_10_4\", \"0_13_8\", \"0_10_22\", \"0_9_18\", \"0_6_24\", \"0_13_1\", \"0_11_10\", \"0_5_4\", \"0_8_1\", \"0_13_11\", \"0_10_19\", \"0_2_23\", \"0_13_3\", \"0_11_8\", \"0_8_9\", \"0_12_15\", \"0_11_15\", \"0_3_24\", \"0_8_17\", \"0_3_10\", \"0_5_1\", \"0_2_21\", \"0_10_17\", \"0_2_8\", \"0_9_8\", \"0_13_7\", \"0_6_9\", \"0_5_20\", \"0_12_24\", \"0_12_11\", \"0_13_0\", \"0_9_11\", \"0_5_23\", \"0_11_11\", \"0_6_23\", \"0_13_14\", \"1_6_23\", \"1_9_5\", \"1_11_16\", \"1_10_16\", \"1_13_22\", \"1_3_18\", \"1_6_19\", \"1_8_23\", \"1_9_20\", \"1_8_9\", \"1_8_24\", \"1_9_15\", \"1_12_9\", \"1_5_2\", \"1_5_24\", \"1_6_17\", \"1_10_3\", \"1_10_22\", \"1_6_7\", \"1_12_1\", \"1_9_10\", \"1_12_10\", \"1_6_5\", \"1_8_21\", \"1_5_17\", \"1_3_20\", \"1_13_23\", \"1_2_15\", \"1_8_7\", \"1_11_1\", \"1_2_7\", \"1_11_23\", \"1_13_8\", \"1_6_13\", \"1_2_21\", \"1_3_13\", \"1_13_15\", \"1_6_18\", \"1_9_9\", \"1_2_19\", \"1_12_17\", \"1_11_11\", \"1_5_20\", \"1_5_16\", \"1_12_6\", \"1_3_7\", \"1_2_20\", \"1_11_7\", \"1_6_1\", \"1_9_22\", \"1_10_0\", \"1_6_22\", \"1_9_14\", \"1_2_3\", \"1_6_6\", \"1_8_17\", \"1_3_19\", \"1_11_0\", \"1_8_18\", \"1_10_15\", \"1_9_8\", \"1_9_18\", \"1_5_0\", \"1_10_23\", \"1_5_23\", \"1_10_13\", \"1_10_7\", \"1_6_8\", \"1_11_6\", \"1_8_0\", \"1_8_6\", \"1_13_17\", \"1_12_4\", \"1_13_13\", \"1_9_0\", \"1_6_20\", \"1_5_4\", \"1_3_3\", \"1_3_23\", \"1_12_20\", \"1_10_4\", \"1_11_2\", \"1_8_20\", \"1_12_3\", \"1_8_8\", \"1_12_8\", \"1_12_0\", \"1_5_11\", \"1_13_12\", \"1_9_11\", \"1_2_1\", \"1_12_13\", \"1_10_17\", \"1_3_22\", \"1_6_9\", \"1_3_12\", \"1_12_16\", \"1_8_15\", \"1_12_23\", \"1_5_5\", \"1_11_5\", \"1_11_20\", \"1_13_16\", \"1_8_4\", \"1_5_13\", \"1_13_18\", \"1_8_13\", \"1_8_1\", \"1_10_9\", \"1_13_4\", \"1_6_12\", \"1_5_7\", \"1_6_21\", \"1_9_2\", \"1_3_11\", \"1_5_3\", \"1_10_19\", \"1_9_4\", \"1_9_3\", \"1_13_0\", \"1_12_19\", \"1_10_11\", \"1_11_13\", \"1_12_21\", \"2_6_24\", \"2_10_13\", \"2_12_18\", \"2_5_2\", \"2_2_10\", \"2_12_4\", \"2_2_15\", \"2_10_14\", \"2_9_11\", \"2_9_19\", \"2_8_5\", \"2_9_14\", \"2_10_6\", \"2_3_18\", \"2_13_12\", \"2_2_0\", \"2_11_19\", \"2_12_21\", \"2_5_0\", \"2_8_19\", \"2_9_18\", \"2_13_5\", \"2_3_12\", \"2_3_6\", \"2_12_6\", \"2_9_21\", \"2_9_4\", \"2_3_14\", \"2_11_10\", \"2_11_1\", \"2_9_6\", \"2_2_17\", \"2_5_16\", \"2_6_9\", \"2_2_16\", \"2_6_11\", \"2_9_9\", \"2_9_12\", \"2_8_15\", \"2_2_1\", \"2_6_10\", \"2_13_16\", \"2_12_1\", \"2_9_13\", \"2_10_3\", \"2_6_4\", \"2_6_1\", \"2_11_12\", \"2_12_12\", \"2_11_7\", \"2_3_17\", \"2_6_17\", \"2_9_0\", \"2_2_8\", \"2_3_24\", \"2_9_3\", \"2_8_6\", \"2_13_0\", \"2_9_2\", \"2_5_10\", \"2_13_4\", \"2_3_10\", \"2_6_3\", \"2_9_1\", \"2_3_11\", \"2_9_10\", \"2_13_14\", \"2_5_7\", \"2_6_19\", \"2_13_19\", \"2_10_1\", \"2_2_19\", \"2_11_20\", \"2_13_24\", \"2_10_4\", \"2_6_18\", \"2_9_15\", \"2_6_15\", \"2_5_22\", \"2_13_3\", \"2_10_16\", \"2_11_11\", \"2_8_17\", \"2_10_2\", \"2_6_0\", \"2_3_21\", \"2_3_15\", \"2_13_8\", \"2_6_22\", \"2_10_9\", \"2_12_16\", \"2_5_18\", \"2_8_10\", \"2_10_15\", \"2_8_22\", \"2_13_13\", \"2_12_8\", \"2_13_15\", \"2_11_22\", \"2_2_4\", \"2_11_13\", \"2_3_4\", \"2_5_4\", \"2_11_4\", \"2_8_14\", \"2_2_6\", \"2_9_16\", \"2_11_23\", \"2_12_13\", \"2_3_9\", \"2_8_20\", \"2_6_20\", \"2_11_15\", \"2_6_21\", \"2_13_22\", \"2_5_9\", \"2_10_8\", \"2_12_0\", \"2_11_3\", \"2_2_24\", \"2_5_24\", \"2_13_23\", \"2_8_8\", \"2_2_21\", \"3_13_24\", \"3_2_17\", \"3_11_3\", \"3_5_9\", \"3_2_0\", \"3_3_23\", \"3_10_23\", \"3_3_6\", \"3_10_9\", \"3_8_18\", \"3_5_12\", \"3_11_17\", \"3_8_5\", \"3_2_15\", \"3_6_14\", \"3_9_12\", \"3_9_9\", \"3_13_10\", \"3_11_14\", \"3_8_3\", \"3_2_8\", \"3_11_8\", \"3_8_1\", \"3_3_15\", \"3_5_0\", \"3_13_0\", \"3_8_2\", \"3_6_7\", \"3_13_3\", \"3_11_1\", \"3_2_20\", \"3_8_16\", \"3_5_3\", \"3_2_13\", \"3_2_7\", \"3_2_4\", \"3_5_18\", \"3_6_3\", \"3_3_7\", \"3_9_18\", \"3_6_15\", \"3_9_15\", \"3_2_24\", \"3_11_22\", \"3_2_18\", \"3_10_12\", \"3_8_0\", \"3_8_13\", \"3_8_21\", \"3_9_6\", \"3_3_21\", \"3_12_13\", \"3_3_1\", \"3_8_15\", \"3_11_4\", \"3_3_0\", \"3_13_19\", \"3_11_16\", \"3_6_9\", \"3_2_19\", \"3_9_17\", \"3_13_11\", \"3_5_20\", \"3_13_7\", \"3_3_3\", \"3_13_23\", \"3_9_16\", \"3_9_8\", \"3_5_13\", \"3_5_23\", \"3_10_15\", \"3_8_19\", \"3_12_5\", \"3_12_4\", \"3_9_19\", \"3_6_20\", \"3_11_21\", \"3_6_1\", \"3_11_12\", \"3_2_22\", \"3_2_5\", \"3_6_13\", \"3_12_6\", \"3_2_16\", \"3_9_0\", \"3_8_23\", \"3_9_10\", \"3_6_0\", \"3_11_15\", \"3_12_9\", \"3_5_1\", \"3_13_20\", \"3_13_8\", \"3_13_6\", \"3_10_5\", \"3_10_24\", \"3_8_22\", \"3_6_10\", \"3_8_6\", \"3_8_7\", \"3_9_5\", \"3_11_11\", \"3_2_14\", \"3_6_2\", \"3_5_10\", \"3_11_23\", \"3_8_11\", \"3_10_0\", \"3_2_1\", \"3_11_24\", \"3_12_23\", \"3_12_17\", \"3_12_0\", \"3_12_24\", \"3_12_2\", \"3_9_2\", \"3_3_9\", \"3_5_7\", \"3_13_16\", \"3_10_4\", \"3_12_8\", \"3_6_4\", \"3_2_6\", \"3_3_24\", \"4_12_3\", \"4_6_12\", \"4_10_13\", \"4_3_21\", \"4_6_21\", \"4_3_15\", \"4_11_16\", \"4_13_5\", \"4_3_4\", \"4_10_2\", \"4_6_7\", \"4_9_11\", \"4_10_8\", \"4_2_20\", \"4_3_6\", \"4_3_2\", \"4_11_11\", \"4_8_16\", \"4_2_0\", \"4_2_2\", \"4_5_6\", \"4_12_1\", \"4_6_19\", \"4_8_0\", \"4_12_8\", \"4_13_10\", \"4_2_21\", \"4_3_24\", \"4_12_22\", \"4_11_8\", \"4_5_17\", \"4_13_4\", \"4_8_1\", \"4_11_14\", \"4_11_20\", \"4_10_21\", \"4_13_2\", \"4_13_1\", \"4_8_12\", \"4_9_0\", \"4_12_11\", \"4_12_2\", \"4_2_24\", \"4_3_11\", \"4_6_16\", \"4_12_7\", \"4_6_3\", \"4_9_13\", \"4_9_2\", \"4_13_6\", \"4_6_9\", \"4_8_18\", \"4_12_24\", \"4_2_11\", \"4_2_18\", \"4_9_8\", \"4_5_22\", \"4_5_10\", \"4_11_24\", \"4_13_22\", \"4_13_17\", \"4_8_11\", \"4_13_14\", \"4_8_20\", \"4_11_13\", \"4_2_1\", \"4_12_16\", \"4_11_15\", \"4_13_0\", \"4_2_5\", \"4_12_20\", \"4_5_5\", \"4_10_12\", \"4_8_21\", \"4_13_13\", \"4_10_6\", \"4_9_7\", \"4_2_19\", \"4_10_15\", \"4_10_22\", \"4_11_1\", \"4_10_7\", \"4_12_0\", \"4_2_12\", \"4_12_15\", \"4_9_9\", \"4_8_17\", \"4_8_8\", \"4_12_10\", \"4_3_23\", \"4_6_22\", \"4_8_4\", \"4_11_10\", \"4_2_9\", \"4_5_23\", \"4_6_4\", \"4_5_3\", \"4_13_12\", \"4_3_10\", \"4_10_4\", \"4_12_6\", \"4_6_13\", \"4_10_10\", \"4_9_12\", \"4_8_6\", \"4_13_16\", \"4_2_17\", \"4_2_10\", \"4_13_8\", \"4_11_21\", \"4_12_12\", \"4_11_9\", \"4_8_3\", \"4_13_18\", \"4_5_9\", \"4_5_20\", \"4_10_14\", \"4_6_5\", \"4_10_5\", \"4_11_2\", \"4_6_11\", \"4_5_13\", \"4_5_21\", \"4_6_17\", \"5_8_21\", \"5_10_19\", \"5_2_8\", \"5_5_16\", \"5_9_21\", \"5_2_20\", \"5_6_18\", \"5_6_8\", \"5_9_18\", \"5_12_11\", \"5_10_17\", \"5_13_11\", \"5_8_3\", \"5_5_13\", \"5_11_7\", \"5_12_8\", \"5_2_0\", \"5_3_23\", \"5_8_17\", \"5_5_6\", \"5_10_5\", \"5_8_9\", \"5_10_21\", \"5_12_10\", \"5_3_16\", \"5_5_14\", \"5_9_1\", \"5_11_21\", \"5_11_16\", \"5_5_22\", \"5_2_24\", \"5_13_15\", \"5_8_0\", \"5_10_1\", \"5_8_14\", \"5_8_20\", \"5_12_19\", \"5_8_6\", \"5_6_7\", \"5_5_1\", \"5_11_14\", \"5_3_19\", \"5_11_9\", \"5_5_24\", \"5_6_20\", \"5_8_24\", \"5_3_12\", \"5_12_2\", \"5_5_2\", \"5_12_16\", \"5_3_0\", \"5_11_10\", \"5_11_20\", \"5_13_19\", \"5_5_5\", \"5_3_15\", \"5_9_14\", \"5_9_24\", \"5_13_24\", \"5_11_23\", \"5_13_7\", \"5_13_23\", \"5_6_9\", \"5_6_24\", \"5_2_12\", \"5_5_7\", \"5_3_22\", \"5_12_12\", \"5_5_18\", \"5_12_23\", \"5_13_22\", \"5_2_23\", \"5_11_24\", \"5_11_17\", \"5_2_14\", \"5_3_4\", \"5_11_6\", \"5_11_0\", \"5_6_6\", \"5_10_4\", \"5_5_20\", \"5_11_15\", \"5_10_3\", \"5_3_10\", \"5_11_2\", \"5_9_4\", \"5_9_2\", \"5_3_14\", \"5_10_7\", \"5_10_23\", \"5_11_3\", \"5_6_4\", \"5_9_19\", \"5_9_8\", \"5_6_22\", \"5_8_7\", \"5_6_11\", \"5_9_9\", \"5_6_23\", \"5_9_15\", \"5_3_1\", \"5_12_20\", \"5_5_3\", \"5_12_21\", \"5_2_22\", \"5_13_16\", \"5_8_15\", \"5_10_18\", \"5_13_20\", \"5_9_12\", \"5_6_10\", \"5_10_10\", \"5_13_10\", \"5_8_2\", \"5_13_17\", \"5_10_12\", \"5_12_13\", \"5_3_9\", \"5_6_12\", \"5_2_16\", \"5_11_19\", \"5_2_5\", \"5_2_1\", \"5_11_22\", \"6_2_11\", \"6_2_17\", \"6_11_13\", \"6_2_24\", \"6_13_18\", \"6_5_19\", \"6_6_14\", \"6_12_1\", \"6_13_23\", \"6_5_15\", \"6_10_6\", \"6_12_19\", \"6_12_22\", \"6_6_12\", \"6_6_1\", \"6_10_17\", \"6_3_23\", \"6_13_9\", \"6_9_5\", \"6_13_19\", \"6_6_4\", \"6_5_22\", \"6_12_15\", \"6_5_21\", \"6_8_1\", \"6_12_7\", \"6_2_9\", \"6_11_16\", \"6_2_5\", \"6_2_18\", \"6_5_7\", \"6_6_6\", \"6_11_7\", \"6_11_6\", \"6_12_14\", \"6_11_9\", \"6_12_20\", \"6_8_16\", \"6_10_18\", \"6_2_19\", \"6_10_2\", \"6_2_22\", \"6_3_22\", \"6_6_18\", \"6_11_22\", \"6_13_22\", \"6_5_24\", \"6_3_19\", \"6_12_5\", \"6_5_17\", \"6_8_7\", \"6_8_19\", \"6_8_6\", \"6_13_8\", \"6_13_2\", \"6_9_19\", \"6_11_17\", \"6_11_4\", \"6_9_7\", \"6_13_21\", \"6_5_6\", \"6_11_0\", \"6_5_18\", \"6_2_10\", \"6_13_3\", \"6_13_24\", \"6_8_10\", \"6_12_0\", \"6_10_24\", \"6_13_11\", \"6_5_23\", \"6_2_3\", \"6_13_16\", \"6_13_6\", \"6_3_12\", \"6_12_17\", \"6_11_15\", \"6_11_24\", \"6_12_2\", \"6_3_21\", \"6_12_18\", \"6_9_23\", \"6_8_24\", \"6_2_13\", \"6_6_17\", \"6_6_0\", \"6_6_3\", \"6_6_8\", \"6_13_15\", \"6_3_8\", \"6_11_10\", \"6_10_16\", \"6_8_22\", \"6_2_2\", \"6_5_9\", \"6_8_5\", \"6_2_23\", \"6_11_20\", \"6_10_11\", \"6_11_14\", \"6_2_6\", \"6_2_0\", \"6_3_1\", \"6_3_17\", \"6_13_4\", \"6_13_10\", \"6_10_22\", \"6_11_8\", \"6_5_16\", \"6_8_4\", \"6_13_17\", \"6_2_14\", \"6_10_23\", \"6_5_3\", \"6_11_5\", \"6_12_8\", \"6_11_12\", \"6_13_14\", \"6_9_17\", \"6_11_18\", \"6_12_3\", \"6_12_13\", \"6_3_16\", \"6_3_7\", \"7_6_5\", \"7_12_19\", \"7_11_18\", \"7_2_6\", \"7_9_23\", \"7_3_21\", \"7_3_23\", \"7_12_2\", \"7_12_14\", \"7_6_16\", \"7_12_3\", \"7_8_9\", \"7_10_4\", \"7_2_12\", \"7_12_8\", \"7_5_12\", \"7_9_15\", \"7_9_16\", \"7_6_11\", \"7_9_5\", \"7_11_4\", \"7_8_0\", \"7_5_21\", \"7_5_19\", \"7_5_10\", \"7_9_7\", \"7_2_7\", \"7_8_22\", \"7_9_24\", \"7_10_5\", \"7_10_15\", \"7_5_7\", \"7_10_13\", \"7_6_17\", \"7_2_23\", \"7_9_20\", \"7_11_20\", \"7_13_14\", \"7_12_12\", \"7_9_2\", \"7_3_22\", \"7_3_18\", \"7_5_3\", \"7_9_6\", \"7_11_0\", \"7_8_15\", \"7_6_6\", \"7_5_1\", \"7_13_24\", \"7_3_4\", \"7_9_17\", \"7_5_2\", \"7_12_1\", \"7_13_0\", \"7_9_22\", \"7_12_9\", \"7_5_17\", \"7_12_10\", \"7_13_3\", \"7_2_10\", \"7_6_8\", \"7_2_18\", \"7_6_18\", \"7_13_13\", \"7_11_1\", \"7_12_21\", \"7_12_15\", \"7_13_18\", \"7_8_21\", \"7_13_7\", \"7_3_9\", \"7_3_8\", \"7_2_11\", \"7_8_12\", \"7_2_2\", \"7_12_4\", \"7_6_12\", \"7_10_16\", \"7_13_4\", \"7_5_16\", \"7_10_9\", \"7_11_19\", \"7_3_16\", \"7_3_15\", \"7_2_8\", \"7_11_2\", \"7_3_24\", \"7_6_13\", \"7_10_19\", \"7_10_0\", \"7_8_6\", \"7_10_17\", \"7_10_3\", \"7_8_14\", \"7_13_22\", \"7_2_19\", \"7_8_8\", \"7_3_11\", \"7_3_6\", \"7_5_24\", \"7_5_0\", \"7_3_0\", \"7_8_17\", \"7_12_11\", \"7_5_18\", \"7_11_6\", \"7_8_3\", \"7_6_22\", \"7_11_11\", \"7_9_18\", \"7_5_6\", \"7_13_19\", \"7_10_8\", \"7_11_24\", \"7_12_13\", \"7_8_24\", \"7_13_6\", \"7_13_12\", \"7_8_16\", \"7_10_10\", \"7_13_1\", \"7_6_10\", \"7_11_8\", \"7_2_14\", \"8_11_15\", \"8_12_10\", \"8_13_11\", \"8_10_6\", \"8_6_21\", \"8_5_10\", \"8_9_10\", \"8_13_17\", \"8_10_22\", \"8_9_20\", \"8_8_12\", \"8_8_2\", \"8_5_2\", \"8_11_23\", \"8_13_22\", \"8_3_21\", \"8_12_8\", \"8_2_16\", \"8_3_18\", \"8_8_17\", \"8_9_15\", \"8_2_10\", \"8_5_1\", \"8_9_23\", \"8_9_2\", \"8_8_1\", \"8_6_10\", \"8_5_11\", \"8_2_3\", \"8_5_7\", \"8_2_6\", \"8_11_8\", \"8_8_4\", \"8_6_9\", \"8_5_5\", \"8_9_8\", \"8_10_1\", \"8_13_1\", \"8_9_1\", \"8_3_0\", \"8_3_13\", \"8_11_21\", \"8_8_5\", \"8_13_13\", \"8_6_24\", \"8_6_11\", \"8_2_9\", \"8_5_15\", \"8_5_13\", \"8_2_8\", \"8_13_15\", \"8_10_12\", \"8_11_10\", \"8_10_2\", \"8_5_0\", \"8_9_24\", \"8_9_21\", \"8_3_15\", \"8_2_19\", \"8_2_22\", \"8_10_13\", \"8_5_8\", \"8_5_16\", \"8_9_6\", \"8_9_19\", \"8_6_17\", \"8_3_7\", \"8_11_5\", \"8_6_19\", \"8_12_0\", \"8_2_2\", \"8_2_0\", \"8_10_4\", \"8_8_23\", \"8_10_16\", \"8_11_18\", \"8_13_3\", \"8_11_9\", \"8_3_14\", \"8_3_5\", \"8_13_8\", \"8_5_18\", \"8_10_23\", \"8_2_18\", \"8_13_23\", \"8_9_18\", \"8_13_12\", \"8_8_14\", \"8_12_24\", \"8_12_5\", \"8_9_12\", \"8_3_1\", \"8_11_16\", \"8_6_6\", \"8_12_11\", \"8_8_15\", \"8_8_24\", \"8_6_4\", \"8_2_15\", \"8_11_20\", \"8_11_14\", \"8_13_19\", \"8_10_15\", \"8_10_9\", \"8_13_10\", \"8_12_22\", \"8_13_9\", \"8_11_22\", \"8_5_20\", \"8_11_4\", \"8_3_24\", \"8_8_13\", \"8_9_17\", \"8_6_0\", \"8_12_6\", \"8_3_11\", \"8_6_1\", \"8_9_4\", \"8_13_18\", \"8_5_14\", \"8_10_21\", \"8_8_21\", \"8_5_21\", \"8_3_8\", \"9_2_1\", \"9_9_14\", \"9_10_1\", \"9_9_1\", \"9_13_2\", \"9_3_0\", \"9_11_24\", \"9_11_0\", \"9_12_23\", \"9_2_24\", \"9_6_8\", \"9_11_3\", \"9_6_9\", \"9_2_18\", \"9_6_24\", \"9_11_16\", \"9_2_12\", \"9_2_15\", \"9_2_20\", \"9_13_13\", \"9_13_6\", \"9_6_14\", \"9_2_9\", \"9_2_11\", \"9_6_22\", \"9_3_16\", \"9_9_2\", \"9_9_19\", \"9_13_16\", \"9_9_8\", \"9_12_4\", \"9_13_7\", \"9_12_7\", \"9_13_10\", \"9_3_19\", \"9_8_10\", \"9_10_13\", \"9_13_15\", \"9_8_19\", \"9_12_1\", \"9_11_4\", \"9_12_10\", \"9_13_5\", \"9_10_3\", \"9_12_19\", \"9_3_18\", \"9_5_16\", \"9_3_22\", \"9_11_22\", \"9_5_8\", \"9_5_20\", \"9_13_22\", \"9_10_17\", \"9_12_16\", \"9_10_24\", \"9_13_8\", \"9_8_15\", \"9_13_18\", \"9_3_24\", \"9_2_2\", \"9_8_20\", \"9_8_17\", \"9_10_20\", \"9_11_2\", \"9_9_6\", \"9_8_9\", \"9_8_23\", \"9_5_10\", \"9_8_18\", \"9_12_9\", \"9_10_7\", \"9_3_2\", \"9_13_11\", \"9_6_3\", \"9_9_24\", \"9_6_18\", \"9_6_0\", \"9_10_8\", \"9_9_9\", \"9_6_6\", \"9_11_7\", \"9_6_5\", \"9_11_13\", \"9_13_20\", \"9_11_10\", \"9_9_11\", \"9_10_22\", \"9_11_19\", \"9_6_16\", \"9_12_13\", \"9_6_12\", \"9_12_5\", \"9_9_7\", \"9_5_6\", \"9_5_5\", \"9_12_14\", \"9_9_0\", \"9_6_23\", \"9_10_4\", \"9_3_5\", \"9_6_2\", \"9_3_6\", \"9_6_13\", \"9_11_11\", \"9_5_13\", \"9_2_19\", \"9_12_11\", \"9_11_20\", \"9_8_21\", \"9_11_1\", \"9_5_4\", \"9_2_6\", \"9_5_3\", \"9_5_2\", \"9_8_8\", \"9_6_15\", \"9_2_7\", \"9_5_24\", \"9_12_8\", \"9_13_21\", \"9_3_12\", \"9_8_6\", \"9_10_6\", \"9_12_2\", \"10_13_24\", \"10_13_16\", \"10_12_8\", \"10_3_2\", \"10_12_18\", \"10_12_3\", \"10_9_2\", \"10_10_4\", \"10_9_9\", \"10_9_5\", \"10_8_18\", \"10_2_6\", \"10_13_1\", \"10_9_21\", \"10_11_1\", \"10_10_17\", \"10_13_17\", \"10_2_17\", \"10_13_6\", \"10_12_24\", \"10_2_5\", \"10_9_11\", \"10_13_7\", \"10_9_8\", \"10_10_2\", \"10_5_8\", \"10_12_23\", \"10_6_11\", \"10_3_24\", \"10_5_3\", \"10_6_15\", \"10_11_22\", \"10_9_24\", \"10_10_23\", \"10_9_18\", \"10_8_23\", \"10_6_6\", \"10_2_21\", \"10_10_21\", \"10_8_12\", \"10_2_3\", \"10_8_7\", \"10_6_9\", \"10_8_14\", \"10_5_17\", \"10_3_16\", \"10_11_5\", \"10_8_8\", \"10_3_21\", \"10_3_18\", \"10_3_10\", \"10_2_10\", \"10_8_15\", \"10_8_5\", \"10_6_4\", \"10_10_14\", \"10_2_4\", \"10_3_0\", \"10_11_18\", \"10_9_12\", \"10_6_24\", \"10_12_5\", \"10_13_3\", \"10_3_8\", \"10_3_9\", \"10_12_0\", \"10_6_10\", \"10_2_23\", \"10_9_23\", \"10_11_10\", \"10_11_9\", \"10_11_14\", \"10_2_16\", \"10_13_8\", \"10_3_22\", \"10_5_13\", \"10_5_22\", \"10_12_9\", \"10_3_6\", \"10_5_21\", \"10_5_2\", \"10_10_9\", \"10_11_19\", \"10_5_20\", \"10_3_14\", \"10_5_10\", \"10_11_15\", \"10_10_10\", \"10_8_21\", \"10_12_22\", \"10_11_24\", \"10_6_14\", \"10_12_14\", \"10_5_7\", \"10_11_13\", \"10_8_24\", \"10_6_12\", \"10_6_21\", \"10_6_1\", \"10_5_23\", \"10_6_0\", \"10_5_18\", \"10_11_4\", \"10_12_21\", \"10_3_3\", \"10_2_11\", \"10_11_2\", \"10_5_11\", \"10_6_3\", \"10_10_24\", \"10_9_13\", \"10_8_13\", \"10_8_17\", \"10_3_20\", \"10_13_22\", \"10_2_22\", \"10_13_23\", \"10_8_1\", \"10_10_12\", \"10_2_18\", \"10_3_4\", \"10_2_0\", \"10_12_6\", \"10_3_5\"]}\n"
},
{
"path": "image/Makefile",
"content": "CC=c++\n\n# c flags\nCFLAGS=-c -Wall -std=c++11 -O3\nCFLAGS_BOOST=-DBOOST_SYSTEM_NO_DEPRECATED\nCFLAGS_CV=`pkg-config --cflags opencv`\nCFLAGS_H5=-I/usr/include -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE \\\n-D_BSD_SOURCE -D_FORTIFY_SOURCE=2 -g -fstack-protector \\\n--param=ssp-buffer-size=4 -Wformat -Werror=format-security\n\n# linking flags\nLDFLAGS_BOOST=-lboost_system -lboost_program_options\nLDFLAGS_H5=-lhdf5 -lhdf5_cpp\nLDFLAGS_CV=-L/usr/local/cuda-7.5/lib64 `pkg-config --libs opencv`\n\nBIN=bin\nSOURCES=$(wildcard *.cpp)\nOBJECTS=$(patsubst %.cpp, %.o, $(SOURCES))\nEXECUTABLE=$(BIN)/image\n\nall: $(SOURCES) $(EXECUTABLE)\n\n$(BIN)/image: $(BIN)/main.o $(BIN)/h5Data.o $(BIN)/seq.o\n\t$(CC) $(BIN)/main.o $(BIN)/h5Data.o $(BIN)/seq.o \\\n\t$(LDFLAGS_BOOST) $(LDFLAGS_H5) $(LDFLAGS_CV) -o $@\n\n$(BIN)/main.o: main.cpp\n\t$(CC) $(CFLAGS) $(CFLAGS_BOOST) $< -o $@\n\n$(BIN)/h5Data.o: h5Data.cpp\n\t$(CC) $(CFLAGS) $(CFLAGS_H5) $< -o $@\n\n$(BIN)/seq.o: seq.cpp\n\t$(CC) $(CFLAGS) $(CFLAGS_CV) $< -o $@\n\nclean:\n\trm -r -f $(BIN)/*\n"
},
{
"path": "image/h5Data.cpp",
"content": "#include \"h5Data.hpp\"\n\nh5Data::h5Data(std::string f, int n) : fileName(f), numChannel(n) {\n\t// turn off the auto-printing when failure occurs\n\tdataFile = new H5::H5File(fileName, H5F_ACC_RDONLY);\n\n\tdatasetLa = load(datasetNameLa, -1);\n\tfor (int i = 0; i < numChannel; i++) {\n\t\tdatasetCh[i] = load(datasetNameCh + std::to_string(i), i);\n\t}\n}\n\nh5Data::~h5Data() {\n\t// delete array of label and channels with []\n\tdelete[] label;\n\tdelete datasetLa;\n\tfor (int i = 0; i < numChannel; i++) {\n\t\tdelete[] ch[i];\n\t\tdelete datasetCh[i];\n\t}\n\n\tdelete dataFile;\n}\n\nH5::DataSet* h5Data::load(H5std_string datasetName, int whichCh) {\n\t// open dataset, dataspace and property list\n\ttry {\n\t\tH5::Exception::dontPrint();\n\t\thsize_t dimsr[2];\n\t\tH5::DataSet* dataset = new H5::DataSet(dataFile->openDataSet(datasetName));\n\t\tH5::DataSpace *filespace = new H5::DataSpace(dataset->getSpace());\n\t\tH5::DSetCreatPropList prop = dataset->getCreatePlist();\n\n\t\t// get information to obtain memory dataspace.\n\t\tint rank = filespace->getSimpleExtentNdims();\n\t\tfilespace->getSimpleExtentDims(dimsr);\n\t\tH5::DataSpace* memspace = new H5::DataSpace(rank, dimsr, NULL);\n\n\t\tlength = dimsr[0];\n\n\t\t// load labels and channels\n\t\tif (whichCh == -1) {\n\t\t\tlabel = new int[length];\n\t\t\tdataset->read(label, H5::PredType::NATIVE_INT, *memspace, *filespace);\n\t\t} else {\n\t\t\tch[whichCh] = new float[length * dimsr[1]];\n\t\t\tdataset->read(ch[whichCh], H5::PredType::NATIVE_FLOAT, *memspace,\n\t\t\t\t*filespace);\n\t\t}\n\n\t\tprop.close();\n\t\tdelete filespace;\n\t\tdelete memspace;\n\n\t\treturn dataset;\n\t} catch (H5::FileIException error) {\n\t\terror.printError();\n\t\treturn NULL;\n\t} catch (H5::DataSetIException error) {\n\t\terror.printError();\n\t\treturn NULL;\n\t} catch (H5::DataSpaceIException error) {\n\t\terror.printError();\n\t\treturn NULL;\n\t}\n}\n"
},
{
"path": "image/h5Data.hpp",
"content": "#ifndef H5DATA_H\n#define H5DATA_H\n\n#include \n#include \n\n// hdf5 includes\n#include \"H5Cpp.h\"\n\n// h5 tags\nnamespace {\n\tconst H5std_string datasetNameLa = \"label\";\n\tconst H5std_string datasetNameCh = \"ch\";\n\n\tconst hsize_t dimsLa[2] = {1, 1};\n\tconst hsize_t dimsCh[2] = {1, 32 * 32};\n\n\tconst int backgroundClass = 0;\n}\n\n#endif\n\n// data class for 5f file\nclass h5Data {\nprivate:\n\tstd::string fileName;\n\tint numChannel;\n\tint length;\n\n\t// store dataset\n\tH5::H5File *dataFile = NULL;\n\tH5::DataSet *datasetLa = NULL;\n\tH5::DataSet *datasetCh[8];\n\n\t// store label and channels data\n\tint *label;\n\tfloat *ch[8];\n\npublic:\n\th5Data(std::string f, int n);\n\t~h5Data();\n\n\tH5::DataSet* load(H5std_string datasetName, int whichCh);\n\n\tinline float* getCh(int whichCh) { return ch[whichCh]; }\n\tinline int* getLa() { return label; }\n\tinline int getLength() { return length; }\n};\n"
},
{
"path": "image/main.cpp",
"content": "#include \n#include \n\n// boost options\n#include \nnamespace po = boost::program_options;\n\n#include \"seq.hpp\"\n\nint main(int argc, char *argv[]) {\n\ttry {\n\t\t// generate argument parser\n\t\tpo::options_description desc(\"Image: allowed options\");\n\t\tdesc.add_options()\n\t\t\t(\"help\", \"Print help messages\")\n\t\t\t(\"source\", po::value(), \"h5 source file\")\n\t\t\t(\"target\", po::value(), \"Flow target directory\")\n\t\t\t(\"channel\", po::value(), \"Channel number\")\n\t\t\t(\"origin\", po::value(), \"Origin size\")\n\t\t\t(\"out\", po::value(), \"Output size\")\n\t\t\t(\"label\", po::value(), \"Print label in output file name\")\n\t\t\t(\"bound\", po::value(), \"Bound for normalization\");\n\n\t\tpo::variables_map vm;\n\t\tpo::store(po::parse_command_line(argc, argv, desc), vm);\n\t\tpo::notify(vm);\n\n\t\tstd::string source, target;\n\t\tint numChannel = 1, origin = 32, out = 32;\n\t\tbool label = false;\n\t\tdouble bound = 15.0;\n\n\t\t// assign argument to local variables\n\t\tif (vm.count(\"help\")) {\n\t\t\tstd::cout << \"Image\" << std::endl << desc << std::endl;\n\t\t\treturn 0;\n\t\t} else {\n\t\t\tif (vm.count(\"source\")) source = vm[\"source\"].as();\n\t\t\tif (vm.count(\"target\")) target = vm[\"target\"].as();\n\t\t\tif (vm.count(\"channel\")) numChannel = vm[\"channel\"].as();\n\t\t\tif (vm.count(\"origin\")) origin = vm[\"origin\"].as();\n\t\t\tif (vm.count(\"out\")) out = vm[\"out\"].as();\n\t\t\tif (vm.count(\"label\")) label = vm[\"label\"].as();\n\t\t\tif (vm.count(\"bound\")) bound = vm[\"bound\"].as();\n\n\t\t\t// generate flow for different channels\n\t\t\tSeq seq = Seq(source, target, numChannel, label);\n\t\t\tfor (int i = 0; i < numChannel; i++) {\n\t\t\t\tseq.flow(i, origin, origin, out, out, bound);\n\t\t\t}\n\t\t}\n\t} catch (std::exception& e) {\n\t\t\tstd::cerr << \"Error: \" << e.what() << std::endl;\n\t\t\treturn 1;\n\t} catch (...) {\n\t\t\tstd::cerr << \"Exception of unknown type\" << std::endl;\n\t\t\treturn 1;\n\t}\n\n\treturn 0;\n}\n"
},
{
"path": "image/seq.cpp",
"content": "#include \"seq.hpp\"\n\nSeq::Seq(std::string s, std::string t, int n, bool l) :\n\t\tsource(s), target(t), numChannel(n), printLabel(l) {\n\t// read h5 file and get the data\n\th5file = new h5Data(source, numChannel);\n\tlabel = h5file->getLa();\n\tfor (int i = 0; i < numChannel; i++) {\n\t\tch[i] = h5file->getCh(i);\n\t}\n\tlength = h5file->getLength();\n}\n\nSeq::~Seq() {\n\t// delete files with pointer at the end, release channel data in the end\n\tdelete h5file;\n}\n\n// convert float (0-1) value images to int (0-255) images with bound limit\nvoid Seq::float2Image(const cv::Mat &floatMat, cv::Mat &imageMat,\n\t\tdouble l, double u) {\n\tfor (int i = 0; i < floatMat.rows; i++) {\n\t\tfor (int j = 0; j < floatMat.cols; j++) {\n\t\t\tfloat x = floatMat.at(i, j);\n\t\t\timageMat.at(i, j) = cast(x, l, u);\n\t\t}\n\t}\n}\n\n// load array data image with offset to int images with resize\ncv::Mat Seq::loadOrigin(float *image, int originHeight, int originWidth,\n\t\tint outHeight, int outWidth) {\n\t// convert array data from float to int image\n\tcv::Mat originFloat = cv::Mat(originHeight, originWidth, CV_32F, image);\n\tcv::Mat originInt = cv::Mat(cv::Size(originWidth, originHeight), CV_8UC1);\n\tfloat2Image(originFloat, originInt, 0.0, 1.0);\n\n\t// resize image if necessary\n\tcv::Mat resize = cv::Mat(cv::Size(outWidth, outHeight), CV_8UC1);\n\tif (originHeight == outHeight && originWidth == outWidth) {\n\t\toriginInt.copyTo(resize);\n\t} else {\n\t\tcv::resize(originInt, resize, cv::Size(outWidth, outHeight),\n\t\t\t0, 0, cv::INTER_LINEAR);\n\t}\n\n\treturn resize;\n}\n\n// generate flow and original image\nvoid Seq::flow(int whichCh, int originHeight, int originWidth,\n\t\tint outHeight, int outWidth, double bound) {\n\t// following frames (start with one to support original flow computation)\n\tfor (int i = 1; i < length; i++) {\n\t\t// load next image\n\t\tstd::string imageName;\n\t\tif (printLabel) {\n\t\t\timageName = target + \"/ch\" + std::to_string(whichCh) + \"_\" +\n\t\t\t\tstd::to_string(i - 1) + \"_\" + std::to_string(label[i - 1]) + \"_\";\n\t\t} else {\n\t\t\timageName = target + \"/ch\" + std::to_string(whichCh) + \"_\" +\n\t\t\t\tstd::to_string(i - 1) + \"_\";\n\t\t}\n\n\t\t// write the original image and save to file\n\t\tcv::Mat image = loadOrigin(ch[whichCh] + i * originHeight * originWidth,\n\t\t\toriginHeight, originWidth, outHeight, outWidth); // resize image for out\n\t\tcv::imwrite(imageName + \"image.jpg\", image);\n\t}\n}\n"
},
{
"path": "image/seq.hpp",
"content": "#ifndef SEQ_H\n#define SEQ_H\n\n#include \n#include \n\n// opencv\n#include \n\n#include \"h5Data.hpp\"\n\n#endif\n\nclass Seq {\nprivate:\n\tstd::string source, target;\n\tint numChannel;\n\tint length;\n\tbool printLabel;\n\n\t// use file pointer to avoid call destructor of h5Data too soon\n\th5Data *h5file = NULL;\n\n\t// not in charge of manage memory\n\tint *label;\n\tfloat *ch[8];\n\n\tvoid resize();\n\t// void flow();\npublic:\n\tSeq(std::string s, std::string t, int n, bool l);\n\t~Seq();\n\n\tinline float cast(float v, float l, float u) {\n\t\tif (v > u) return 255;\n\t\telse if (v < l) return 0;\n\t\telse return cvRound(255 * (v - l) / (u - l));\n\t}\n\n\tvoid float2Image(const cv::Mat &floatMat, cv::Mat &imageMat,\n\t\tdouble lBound, double uBound);\n\tcv::Mat loadOrigin(float *image, int originHeight, int originWidth,\n\t\tint outHeight, int outWidth);\n\n\tvoid flow(int whichCh, int originHeight, int originWidth,\n\t\tint outHeight, int outWidth, double bound);\n};\n"
},
{
"path": "net/data.lua",
"content": "require 'class'\nrequire 'paths'\nrequire 'mattorch'\nlocal json = require 'lunajson'\n\nrequire 'net.util'\n\nlocal Data = torch.class('Data')\n\nfunction Data:__init(fileDir, listFile, meanFile, listType,\n inputSize, inputCh, labelSize, dataSize, dataCh, ch, useCuda)\n self.seqList = self:getSeq(fileDir, listFile, listType)\n self.mean = self:loadMean(meanFile)\n self.inputSize = inputSize\n self.inputCh = inputCh\n self.labelSize = labelSize\n self.dataSize = dataSize\n self.dataCh = dataCh\n self.ch = ch\n self.useCuda = useCuda\nend\n\nfunction Data:getSeq(fileDir, listFile, listType)\n -- read file into json\n local f = io.open(listFile, 'rb')\n local jsonString = f:read('*all')\n local jsonObject = json.decode(jsonString)\n f:close()\n -- arrange it according to different label\n local seqList = {}\n for i in pairs(jsonObject[listType]) do\n seqList[#seqList + 1] = paths.concat(fileDir, jsonObject[listType][i])\n end\n return seqList\nend\n\n-- random shuffle a list\nfunction Data:shuffleList(list)\n local n, random = #list, math.random\n for i = 1, n do\n local j, k = random(n), random(n)\n list[j], list[k] = list[k], list[j]\n end\n return list\nend\n\n-- load mat mean file\nfunction Data:loadMean(meanFile)\n return mattorch.load(meanFile)\nend\n\nfunction Data:initBatch(batchSize, seqLength)\n -- initialize inputs and targets, zero mask will ignore null target\n local inputs, targets = {}, {}\n for l = 1, seqLength do\n -- init with 4d data (cnn and uni)\n inputs[l] = torch.Tensor(batchSize,\n self.inputCh, self.inputSize, self.inputSize):zero()\n targets[l] = torch.Tensor(batchSize):fill(self.labelSize)\n end\n return inputs, targets\nend\n"
},
{
"path": "net/imageseq.lua",
"content": "require 'class'\nrequire 'paths'\nrequire 'image'\nlocal json = require 'lunajson'\n\nrequire 'net.seq'\nrequire 'net.util'\n\nlocal ImageSeq = torch.class('ImageSeq', 'Seq')\n\nfunction ImageSeq:__init(seqName, mean, dataSize, dataCh, ch, useCuda)\n Seq.__init(self, seqName, mean, dataSize, dataCh, ch, useCuda)\n self.info = self:getSeqInfo()\n -- load frame depends on image type, lib image support only float tensor\n -- reset tensor type back to float tensor when using cuda\n torch.setdefaulttensortype('torch.FloatTensor')\n self.imageFrame = self:loadFrame()\n -- set tensor type back to cuda tensor when using cuda\n if self.useCuda then\n torch.setdefaulttensortype('torch.CudaTensor')\n end\nend\n\n-- read label file\nfunction ImageSeq:getSeqInfo()\n -- read file into json\n local f = io.open(paths.concat(self.name, 'label.json'), 'rb')\n local jsonString = f:read('*all')\n local jsonObject = json.decode(jsonString)\n f:close()\n -- convert into num based index, 0-based to 1-based\n -- label is also 1-based: 0-4 => 1-5\n local seqInfo = {}\n for i in pairs(jsonObject) do\n seqInfo[i + 1] = jsonObject[i] + 1\n end\n return seqInfo\nend\n\nfunction ImageSeq:loadFrame()\n local frame = {}\n local m = self.mean[string.format('ch%d_image', self.ch)]\n :permute(3, 2, 1) -- the matlab order is reversed\n -- access each frame, 1-based to 0-based image name\n for f = 1, #self.info do\n local name = string.format('ch%d_%d_image.jpg', self.ch, f - 1)\n -- for image tensor, first map from 0-1 to 0-255,\n -- then minus mean (also in 0-255)\n frame[#frame + 1] = (image.load(paths.concat(self.name, name), 1) * 256) - m\n end\n return frame\nend\n\nfunction ImageSeq:getImage(f)\n -- appending later frames\n local mat = self.imageFrame[f]\n for i = 1, self.dataCh - 1 do\n mat = torch.cat(mat, self.imageFrame[math.min(#self.info, f + i)], 1)\n end\n return mat\nend\n\nfunction ImageSeq:getFrame(f)\n -- load data depends on image or flow\n local matrix = self:getImage(f)\n return matrix\nend\n"
},
{
"path": "net/main.lua",
"content": "require 'net.util'\n\n-- parse cmd parameters\nlocal cmd = torch.CmdLine()\n\ncmd:option('--file', '../collect_image/origin_trans', 'Image dir')\ncmd:option('--list', 'tmp/file.json', 'List file json')\ncmd:option('--load', '', 'Load model file')\n\ncmd:option('--inputsize', 32, 'Input size')\ncmd:option('--inputch', 4, 'Input channel')\ncmd:option('--label', 3, 'Label size')\n\ncmd:option('--datasize', 32, 'Data size')\ncmd:option('--datach', 4, 'Data channel (number of stack)')\n\ncmd:option('--batch', 4, 'Batch size')\ncmd:option('--maxseq', 40, 'Sequence length')\n\ncmd:option('--cuda', false, 'Use CUDA')\ncmd:option('--cudnn', false, 'Use CUDNN')\nlocal opt = cmd:parse(arg or {})\n\n--- other parameters\nlocal meanFile = 'tmp/mean_32.mat'\nlocal ch = 1\n\n-- default tensor\nif opt.cuda then\n require 'cutorch'\n require 'cunn'\n torch.setdefaulttensortype('torch.CudaTensor')\nelse\n torch.setdefaulttensortype('torch.FloatTensor')\nend\n\n-- train net, switch train type\nlocal trainObject\nrequire 'net.rnntrain'\ntrainObject = RnnTrain(opt.file, opt.list, meanFile,\n opt.inputsize, opt.inputch, opt.label,\n opt.datasize, opt.datach, ch,\n opt.load, opt.cuda, opt.cudnn)\n\ntrainObject:train(opt.batch, opt.maxseq)\n"
},
{
"path": "net/net.lua",
"content": "require 'class'\n\nlocal Net = torch.class('Net')\n\nfunction Net:__init(useCuda, useCudnn)\n -- cuda support\n self.useCuda = useCuda\n self.useCudnn = useCudnn\n if self.useCuda then\n require 'cutorch'\n require 'cunn'\n if self.useCudnn then\n require 'cudnn'\n end\n end\nend\n\n-- load already exist net\nfunction Net:loadNet(loadFile)\n print('[net] loading model ' .. loadFile)\n net = torch.load(loadFile)\n print(net)\n return net\nend\n\n-- cuda and cudnn support\nfunction Net:cudaNet(net)\n if self.useCuda then\n print('[net] with cuda')\n net = net:cuda()\n if self.useCudnn then\n print('[net] with cudnn')\n cudnn.convert(net, cudnn)\n end\n else\n print('[net] without cuda')\n end\n print(net)\n return net\nend\n"
},
{
"path": "net/rnndata.lua",
"content": "require 'class'\n\nrequire 'net.data'\nrequire 'net.imageseq'\n\nlocal RnnData = torch.class('RnnData', 'Data')\n\nfunction RnnData:__init(fileDir, listFile, meanFile, listType,\n inputSize, inputCh, labelSize, dataSize, dataCh, ch, useCuda)\n Data.__init(self, fileDir, listFile, meanFile, listType,\n inputSize, inputCh, labelSize, dataSize, dataCh, ch, useCuda)\nend\n\nfunction RnnData:frameSeq(seq, maxSeq)\n -- load label for each frame, label is 1-based, f is frame id\n -- offset introduce zero paddings at the beginning of the sequence\n local frames = {}\n if #seq.info < maxSeq then\n local offset = maxSeq - #seq.info\n for f = 1, #seq.info do\n frames[offset + f] = f\n end\n else\n for f = 1, maxSeq do\n frames[f] = math.floor(f * (#seq.info / maxSeq))\n end\n end\n return frames\nend\n\n-- load batch to tensor\nfunction RnnData:loadBatch(batchNum, batchSize, maxSeq)\n local inputs, targets = self:initBatch(batchSize, maxSeq)\n -- load data into inputs and targets\n local batchStart = (batchNum - 1) * batchSize\n for b = 1, batchSize do\n -- get seq info\n local seq\n seq = ImageSeq(self.seqList[batchStart + b], self.mean,\n self.dataSize, self.dataCh, self.ch, self.useCuda)\n -- generate frame sequence\n local frames = self:frameSeq(seq, maxSeq)\n for i, v in pairs(frames) do\n inputs[i][b] = seq:getFrame(v)\n targets[i][b] = seq.info[v]\n end\n end\n\n return inputs, targets\nend\n"
},
{
"path": "net/rnntrain.lua",
"content": "require 'class'\n\nrequire 'net.train'\nrequire 'net.util'\n\nlocal RnnTrain = torch.class('RnnTrain', 'Train')\n\nfunction RnnTrain:__init(fileDir, listFile, meanFile,\n inputSize, inputCh, labelSize, dataSize, dataCh, ch,\n loadFile, useCuda, useCudnn)\n -- call base train constructor\n Train.__init(self, labelSize, useCuda, useCudnn)\n -- load data\n require 'net.rnndata'\n self.evalData = RnnData(fileDir, listFile, meanFile, 'eval',\n inputSize, inputCh, labelSize, dataSize, dataCh, ch, useCuda)\n print(string.format('[eval] data with %d seq', #self.evalData.seqList))\n -- build net\n local netObject\n require 'net.uninet'\n netObject = UniNet(useCuda, useCudnn)\n self.net = netObject:loadNet(loadFile)\nend\n\nfunction RnnTrain:batchEval(b, inputs)\n -- evaluate forward as whole sequence\n self.net:evaluate()\n local outputs = self.net:forward(self:convertCuda(inputs))\n self.net:forget()\n return outputs\nend\n"
},
{
"path": "net/seq.lua",
"content": "require 'class'\nrequire 'paths'\nrequire 'image'\nlocal json = require 'lunajson'\n\nrequire 'net.util'\n\nlocal Seq = torch.class('Seq')\n\nfunction Seq:__init(seqName, mean, dataSize, dataCh, ch, useCuda)\n -- load info\n self.name = seqName\n self.mean = mean\n self.dataSize = dataSize\n self.dataCh = dataCh\n self.ch = ch\n self.useCuda = useCuda\nend\n"
},
{
"path": "net/stat.lua",
"content": "require 'class'\nlocal json = require 'lunajson'\n\nlocal Stat = torch.class('Stat')\n\nfunction Stat:__init(labelSize, batchSize, maxSeq)\n self.labelSize = labelSize\n self.batchSize = batchSize\n self.maxSeq = maxSeq\n -- init confusion matrix (correct label (size) * predict label (size))\n self.confus = self:initMat(labelSize, labelSize)\nend\n\nfunction Stat:initMat(height, width)\n local mat = {}\n for h = 1, height do\n mat[h] = {}\n for w = 1, width do\n mat[h][w] = 0\n end\n end\n return mat\nend\n\nfunction Stat:updateStat(outputs, targets)\n -- calculate one sequence by one sequence\n for o = 1, #outputs do\n local _, indices = torch.max(outputs[o], 2)\n for b = 1, self.batchSize do\n local right, pred = targets[o][b], indices[b][1]\n -- update confusion matrix\n self.confus[right][pred] = self.confus[right][pred] + 1\n end\n end\nend\n\nfunction Stat:sum(list, length)\n local result = 0\n -- ignore null label in sum\n for i = 1, length do\n result = result + list[i]\n end\n return result\nend\n\n-- calculate per label stat\nfunction Stat:calPerLabel()\n local right, count = {}, {}\n -- iterate for all label\n for l = 1, self.labelSize do\n right[l] = self.confus[l][l]\n count[l] = self:sum(self.confus[l], #self.confus[l])\n end\n return right, count\nend\n\n-- print matrix\nfunction Stat:printMat(str, right, count, matRight, matCount)\n for l = 1, #right do\n if count[l] > 0 then\n local accLine = right[l] * 100 / count[l]\n io.write(string.format(' %s %02d: %02.0f [', str, l, accLine))\n for ll = 1, #matRight[1] do\n -- print each element depends on the type and value\n if type(matCount[l]) == 'number' and matCount[l] > 0 then\n local acc = matRight[l][ll] * 100 / matCount[l]\n io.write(string.format(' %02.0f ', acc))\n else\n io.write(' -- ')\n end\n end\n io.write(']\\n')\n else\n io.write(string.format(' %s %02d: --\\n', str, l))\n end\n end\nend\n\nfunction Stat:print(type)\n local right, count = self:calPerLabel()\n -- calculate sum of label\n local accEpoch = self:sum(right, #right - 1) / self:sum(count, #count - 1)\n io.write(string.format('[%s] accuracy %f\\n', type, accEpoch))\n -- print per label and confusion matrix\n self:printMat('label', right, count, self.confus, count)\n -- return accuracy for lr update\n return accEpoch\nend\n"
},
{
"path": "net/train.lua",
"content": "require 'class'\n\nlocal Train = torch.class('Train')\n\nfunction Train:__init(labelSize, useCuda, useCudnn)\n -- require and cuda support\n require 'net.stat'\n self.useCuda = useCuda\n self.useCudnn = useCudnn\n if useCuda then\n require 'cutorch'\n require 'cunn'\n end\n -- other parameters\n self.labelSize = labelSize\nend\n\n-- one step of evaluate to all data\nfunction Train:epochEval(batchSize, maxSeq)\n local stat = Stat(self.labelSize, batchSize, maxSeq)\n for b = 1, math.floor(#self.evalData.seqList / batchSize) do\n -- load data, start evaluating, depends on net type\n local inputs, targets = self.evalData:loadBatch(b, batchSize, maxSeq)\n local outputs = self:batchEval(b, inputs)\n -- update confusion matrix\n stat:updateStat(outputs, targets)\n end\n return stat:print('eval')\nend\n\n-- convert to cuda\nfunction Train:convertCuda(list)\n if self.cuda then\n -- convert table one by one to cuda\n for l = 1, #list do\n list[l] = list[l]:cuda()\n end\n return list\n else\n return list\n end\nend\n\nfunction Train:train(batchSize, maxSeq)\n -- evaluate by epoch\n self:epochEval(batchSize, maxSeq)\n print('Finished')\nend\n"
},
{
"path": "net/uninet.lua",
"content": "require 'class'\nrequire 'rnn'\n\nrequire 'net.net'\n\nlocal UniNet = torch.class('UniNet', 'Net')\n\nfunction UniNet:__init(useCuda, useCudnn)\n Net.__init(self, useCuda, useCudnn)\nend\n"
},
{
"path": "net/util.lua",
"content": "-- helper function to check if string start with some sub-string\nfunction string:startsWith(start)\n return string.sub(self, 1, string.len(start)) == start\nend\n\nfunction string:split(sep)\n local sep, fields = sep or \":\", {}\n local pattern = string.format(\"([^%s]+)\", sep)\n self:gsub(pattern, function(c) fields[#fields+1] = c end)\n return fields\nend\n"
},
{
"path": "pre/image.py",
"content": "import cv2\nimport numpy as np\n\nclass image:\n def __init__(self, file_name):\n self.file_name = file_name\n\n # load image and resize image with size\n def load(self, height, width):\n frame = cv2.imread(self.file_name, cv2.IMREAD_GRAYSCALE)\n if height and width:\n frame = cv2.resize(frame, (width, height))\n frame = np.expand_dims(frame, axis=0).astype('float32')\n return frame\n"
},
{
"path": "pre/main.py",
"content": "import argparse\n\nfrom seq import Seq\n\n# call different method depends on op\ndef main(op, file_dir, target_dir, num_channel=1, redo=False,\n origin_size=32, out_size=32):\n s = Seq()\n\n if op == 'image':\n s.generate_image(file_dir, target_dir, num_channel,\n origin_size, out_size, redo)\n elif op == 'clean':\n s.clean_image(file_dir)\n elif op == 'mean':\n s.generate_mean(file_dir, target_dir, num_channel, out_size)\n else:\n raise NameError('The operation type {} is undefined'.format(op))\n\nif __name__ == \"__main__\":\n parser = argparse.ArgumentParser(description=\"Generate image\")\n parser.add_argument('--op', type=str)\n parser.add_argument('--file', type=str)\n parser.add_argument('--target', type=str)\n parser.add_argument('--channel', type=int)\n parser.add_argument('--redo', action='store_true')\n parser.add_argument('--originsize', type=int)\n parser.add_argument('--outsize', type=int)\n\n args = parser.parse_args()\n\n main(args.op, args.file, args.target, args.channel, args.redo,\n args.originsize, args.outsize)\n"
},
{
"path": "pre/op.py",
"content": "import os, subprocess\nimport numpy as np\nimport h5py\nimport json\n\nfrom image import image\n\nclass Operation:\n def __init__(self):\n pass\n\n # generate image for each sequence\n def generate_image(self, source, target, num_channel, origin_size=32,\n out_size=32, bound=15.0, bin_image='image/bin/image'):\n if not os.path.exists(target): os.makedirs(target)\n # run flow command\n p = subprocess.Popen([bin_image,\n '--source', source, '--target', target,\n '--channel', str(num_channel),\n '--origin', str(origin_size),\n '--out', str(out_size),\n '--bound', str(bound)])\n p.wait()\n # generate label file\n self._generate_label(source, target)\n\n # load labels\n def _generate_label(self, source, target, label_file='label.json'):\n with h5py.File(source, 'r') as hf:\n labels = [int(l) for l in hf['label'][()]]\n # save labels as a dict in json file\n labels_dict = {i: l\n for i, l in zip(range(len(labels) - 1), labels[1:])}\n with open(os.path.join(target, label_file), 'w') as jf:\n json.dump(labels_dict, jf)\n\n # delete generated image files\n def clean_image(self, file_dir, label_file='label.json'):\n print 'Cleaning', file_dir\n # delete images in dir\n image_files = [os.path.join(file_dir, f)\n for f in os.listdir(file_dir)\n if os.path.isfile(os.path.join(file_dir, f)) and 'jpg' in f]\n for image_file in image_files:\n try:\n os.remove(image_file)\n except Exception, e:\n print e\n # delete label file\n try:\n os.remove(os.path.join(file_dir, label_file))\n except Exception, e:\n print e\n # delete dir\n try:\n os.rmdir(file_dir)\n except Exception, e:\n print e\n\n # get frame count\n def _get_frame_num(self, source, label_file='label.json'):\n with open(os.path.join(source, label_file), 'r') as jf:\n labels_dict = json.load(jf)\n return len(labels_dict)\n\n # setup mean counter and accumulator at the beginning\n def setup_mean(self, num_channel, out_size):\n self.image_sums = {}\n self.count = 0.0\n for c in range(num_channel):\n self.image_sums['ch%i_image' % (c,)] = \\\n np.zeros((1, out_size, out_size), dtype='float32')\n\n # accumulate mean for each sequence\n def accum_mean(self, source, num_channel, out_size):\n print 'Loading mean', source\n frame_num = self._get_frame_num(source)\n self.count += frame_num\n for c in range(num_channel):\n for i in range(frame_num):\n image_name = os.path.join(source,\n 'ch%i_%i_image.jpg' % (c, i))\n self.image_sums['ch%i_image' % (c,)] += \\\n image(image_name).load(out_size, out_size)\n\n # save accumulated mean to file\n def save_mean(self, mean_file, num_channel):\n # store file as hdf5\n if mean_file.endswith('h5'):\n print 'Save as hdf5'\n with h5py.File(mean_file, 'w') as f:\n for c in range(num_channel):\n f.create_dataset('ch%i_image' % (c,),\n data=self.image_sums['ch%i_image' % (c,)]\n / self.count)\n # store file as matlab data\n elif mean_file.endswith('mat'):\n import scipy.io as sio\n print 'Save as mat'\n data = {}\n for c in range(num_channel):\n data['ch%i_image' % (c,)] = \\\n self.image_sums['ch%i_image' % (c,)] / self.count\n sio.savemat(mean_file, data)\n"
},
{
"path": "pre/seq.py",
"content": "import os, re\nimport numpy as np\nimport h5py\n\nfrom op import Operation\n\nclass Seq:\n def __init__(self):\n pass\n\n # get files in label directories\n def get_h5(self, file_dir):\n return [os.path.join(file_dir, f)\n for f in os.listdir(file_dir)\n if os.path.isfile(os.path.join(file_dir, f)) and 'h5' in f]\n\n # get list of target image dir names\n def get_dir(self, file_dir):\n return [os.path.join(file_dir, d)\n for d in os.listdir(file_dir)\n if os.path.isdir(os.path.join(file_dir, d))]\n\n # generate image\n def generate_image(self, file_dir, target_dir, num_channel,\n origin_size, out_size, redo):\n source_files = self.get_h5(file_dir)\n o = Operation()\n for s in source_files:\n t = os.path.join(target_dir, re.sub('.h5', '',\n os.path.basename(s)))\n if redo or not os.path.exists(t):\n print 'Generating', s\n o.generate_image(s, t, num_channel, origin_size, out_size)\n\n # clean image\n def clean_image(self, file_dir):\n file_dirs = self.get_dir(file_dir)\n o = Operation()\n for d in file_dirs:\n o.clean_image(d)\n\n # generate mean\n def generate_mean(self, file_dir, target, num_channel, out_size):\n o = Operation()\n o.setup_mean(num_channel, out_size)\n source_dir = self.get_dir(file_dir)\n for s in source_dir:\n o.accum_mean(s, num_channel, out_size)\n o.save_mean(target, num_channel)\n"
}
]