initial commit

Author: michalfaber

.gitignore

@@ -0,0 +1,10 @@
+model/**/*.caffemodel
+model/**/*.h5
+model/**/*.npy
+dataset/train2017
+dataset/val2017
+dataset/test2017
+dataset/annotations
+.idea
+__pycache__
+.ipynb_checkpoints

README.md

@@ -0,0 +1,66 @@
+# Realtime Multi-Person Pose Estimation
+This is a keras version of [Realtime Multi-Person Pose Estimation](https://github.com/ZheC/Realtime_Multi-Person_Pose_Estimation) project  
+
+## Introduction
+Code repo for reproducing [2017 CVPR](https://arxiv.org/abs/1611.08050) paper using keras.  
+
+## Results
+
+<p align="center">
+<img src="https://github.com/michalfaber/keras_Realtime_Multi-Person_Pose_Estimation/blob/master/readme/dance.gif", width="720">
+</p>
+
+<div align="center">
+<img src="https://github.com/michalfaber/keras_Realtime_Multi-Person_Pose_Estimation/blob/master/sample_images/ski.jpg", width="300", height="300">
+&nbsp;
+<img src="https://github.com/michalfaber/keras_Realtime_Multi-Person_Pose_Estimation/blob/master/readme/result.png", width="300", height="300">
+</div>
+
+
+## Contents
+1. [Converting caffe model](#converting-caffe-model-to-keras-model)
+2. [Testing](#testing-steps)
+3. [Training](#training-steps)
+
+## Require
+1. [Keras](https://keras.io/)
+2. [Caffe - docker](https://hub.docker.com/r/bvlc/caffe/) required if you would like to convert caffe model to keras model. You 
+ don't have to compile/install caffe on your local machine.
+
+## Converting Caffe model to Keras model
+Authors of [original implementation](https://github.com/ZheC/Realtime_Multi-Person_Pose_Estimation) released already trained caffe model 
+which you can use to extract weights data.   
+
+- Download caffe model `cd model; sh get_caffe_model.sh`
+- Dump caffe layers to numpy data `cd ..; docker run -v [absolute path to your keras_Realtime_Multi-Person_Pose_Estimation folder]:/workspace -it bvlc/caffe:cpu python dump_caffe_layers.py`
+  Note that docker accepts only absolute paths so you have to set the full path to the folder containing this project.
+- Convert caffe model (from numpy data) to keras model `python caffe_to_keras.py`  
+
+## Testing steps
+- Convert caffe model to keras model or download already converted keras model https://www.dropbox.com/s/llpxd14is7gyj0z/model.h5
+- Run the notebook `demo.ipynb`.
+- `python demo_image.py --image sample_images\ski.jpg` to run the picture demo. Result will be stored in the file result.png. You can use
+any image file as an input.
+- `python demo_camera.py` to run the web demo.
+
+## Training steps
+- Install gsutil `curl https://sdk.cloud.google.com | bash`. This is a really helpful tool for downloading large datasets. 
+- Download the data set (~25 GB) `cd dataset; sh get_dataset.sh`,
+- Download [COCO official toolbox](https://github.com/pdollar/coco) in `dataset/coco/` . 
+- `cd coco/PythonAPI; sudo python setup.py install` to install pycocotools.
+- Run `cd ../..; python train_pose.py` to start training. 
+
+## Related repository
+- CVPR'16, [Convolutional Pose Machines](https://github.com/shihenw/convolutional-pose-machines-release).
+- CVPR'17, [Realtime Multi-Person Pose Estimation](https://github.com/ZheC/Realtime_Multi-Person_Pose_Estimation).
+
+## Citation
+Please cite the paper in your publications if it helps your research:    
+
+    @InProceedings{cao2017realtime,
+      title = {Realtime Multi-Person 2D Pose Estimation using Part Affinity Fields},
+      author = {Zhe Cao and Tomas Simon and Shih-En Wei and Yaser Sheikh},
+      booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
+      year = {2017}
+      }
+	  

caffe_to_keras.py

@@ -0,0 +1,23 @@
+from model import get_model
+import numpy as np
+import os
+
+CAFFE_LAYERS_DIR = "model/caffe/layers"
+KERAS_MODEL_FILE = "model/keras/model.h5"
+
+m = get_model()
+
+for layer in m.layers:
+    layer_name = layer.name
+    if (os.path.exists(os.path.join(CAFFE_LAYERS_DIR, "W_%s.npy" % layer_name))):
+        w = np.load(os.path.join(CAFFE_LAYERS_DIR, "W_%s.npy" % layer_name))
+        b = np.load(os.path.join(CAFFE_LAYERS_DIR, "b_%s.npy" % layer_name))
+
+        w = np.transpose(w, (2, 3, 1, 0))
+
+        layer_weights = [w, b]
+        layer.set_weights(layer_weights)
+
+m.save_weights(KERAS_MODEL_FILE)
+
+print("Done !")

config

@@ -0,0 +1,36 @@
+[param]
+
+# CPU mode or GPU mode
+use_gpu = 1
+
+# GPU device number (doesn't matter for CPU mode)
+GPUdeviceNumber = 0
+
+# Select model (default: 1)
+modelID = 1
+
+# Look in matlab counterpart for explanation
+octave = 3
+starting_range = 0.8
+ending_range = 2
+scale_search = 0.5, 1, 1.5, 2
+thre1 = 0.1
+thre2 = 0.05 
+thre3 = 0.5 
+min_num = 4
+mid_num = 10
+crop_ratio = 2.5
+bbox_ratio = 0.25
+
+[models]
+## don't edit this part
+
+[[1]]
+caffemodel = './model/_trained_COCO/pose_iter_440000.caffemodel'
+deployFile = './model/_trained_COCO/pose_deploy.prototxt'
+description = 'COCO Pose56 Two-level Linevec'
+boxsize = 368
+padValue = 128
+np = 12
+stride = 8
+part_str = [nose, neck, Rsho, Relb, Rwri, Lsho, Lelb, Lwri, Rhip, Rkne, Rank, Lhip, Lkne, Lank, Leye, Reye, Lear, Rear, pt19]

config_reader.py

@@ -0,0 +1,33 @@
+from configobj import ConfigObj
+import numpy as np
+
+
+def config_reader():
+    config = ConfigObj('config')
+
+    param = config['param']
+    model_id = param['modelID']
+    model = config['models'][model_id]
+    model['boxsize'] = int(model['boxsize'])
+    model['stride'] = int(model['stride'])
+    model['padValue'] = int(model['padValue'])
+    #param['starting_range'] = float(param['starting_range'])
+    #param['ending_range'] = float(param['ending_range'])
+    param['octave'] = int(param['octave'])
+    param['use_gpu'] = int(param['use_gpu'])
+    param['starting_range'] = float(param['starting_range'])
+    param['ending_range'] = float(param['ending_range'])
+    param['scale_search'] = map(float, param['scale_search'])
+    param['thre1'] = float(param['thre1'])
+    param['thre2'] = float(param['thre2'])
+    param['thre3'] = float(param['thre3'])
+    param['mid_num'] = int(param['mid_num'])
+    param['min_num'] = int(param['min_num'])
+    param['crop_ratio'] = float(param['crop_ratio'])
+    param['bbox_ratio'] = float(param['bbox_ratio'])
+    param['GPUdeviceNumber'] = int(param['GPUdeviceNumber'])
+
+    return param, model
+
+if __name__ == "__main__":
+    config_reader()

dataset/coco/INFO

@@ -0,0 +1 @@
+Folder for https://github.com/pdollar/coco files 

dataset/get_dataset.sh

@@ -0,0 +1,23 @@
+#!/usr/bin/env bash
+
+# Install gsutil which provides tools for efficiently accessing datasets
+# without unzipping large files.
+# Install gsutil via:curl https://sdk.cloud.google.com | bash
+
+mkdir train2017
+mkdir val2017
+mkdir test2017
+mkdir annotations
+
+echo "Downloading train2017..."
+gsutil -m rsync gs://images.cocodataset.org/train2017 train2017
+
+echo "Downloading val2017..."
+gsutil -m rsync gs://images.cocodataset.org/val2017 val2017
+
+echo "Downloading test2017..."
+gsutil -m rsync gs://images.cocodataset.org/test2017 test2017
+
+echo "Downloading annotations..."
+gsutil -m rsync gs://images.cocodataset.org/annotations annotations
+