This is an implementation of Realtime Multi-Person Pose Estimation with Chainer. The original project is here.

日本語版 README

 

This project is licensed under the terms of the license.

1. Converting caffe model
2. Testing
3. Training

• Python 3.4+
• Chainer 2.1+
• NumPy
• Matplotlib
• OpenCV

The authors of the original implementation provide trained caffe model which you can use to extract model weights. Execute the following commands to download the trained model and convert it to npz file:

cd models
wget http://posefs1.perception.cs.cmu.edu/Users/ZheCao/pose_iter_440000.caffemodel
python convert_model.py pose_iter_440000.caffemodel
cd ..

Execute the following command with the weight parameter file and the image file as arguments for inference. The resulting image will be saved as result.png.

python pose_detector.py posenet models/coco_posenet.npz --img data/person.png 

 

If the --img option is omitted, the real-time demonstration mode with the web camera is activated. Quit with the q key.

This is a full scratch training procedure using COCO 2017 dataset.

bash getData.sh

If you already downloaded the dataset by yourself, please skip this procedure and change coco_dir in entity.py to the dataset path that was already downloaded.

git clone https://github.com/cocodataset/cocoapi.git
cd cocoapi/PythonAPI/
make
python setup.py install
cd ../../

Execute the following command to check randomly generated training images by generator. Please confirm that you can see the correct PAFs, Heatmaps, and masks on the clipped image.

python coco_data_loader.py

For each 1000 iterations, the recent weight parameters are saved as a weight file model_iter_1000.

python train_coco_pose_estimation.py --gpu=0

Execute the following command with your own trained weight parameter file and the image file as arguments for inference. The resulting image will be saved as result.png.

python pose_detector.py posenet model_iter_1000 --img data/person.png 

• CVPR'16, Convolutional Pose Machines.
• CVPR'17, Realtime Multi-Person Pose Estimation.

Please cite the original 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}
  }