PyTorch (re)implement of Convolutional Pose Machine [1] with weighted loss as an option. More information about weighted loss please refer to another repo of me.

Please notice that this is an underdeveloped implementation of Convolutional Pose Machine for my own experiment. But the architecture and training are very close to the original model, please feel free to use it for your own project.

• Python 3.6
• scipy
• sklearn
• pillow
• PyTorch 0.2
• torchvision 0.1.9
• tensorboardX (only if you need tensorboard summary)
• TensorFlow (for tensorboard web server)
• OpenCv > 3.0

I found that original link of the Leeds Sports Pose Dataset at University of Leeds has been removed. You can download the dataset here and the extended dataset here.

Please download the dataset and unzip it in data folder with a directory tree like this:

data
└── LSP
    ├── lsp_dataset
    │   ├── images
    │   └── visualized
    └── lspet_dataset
        └── images

python -W ignore::UserWarning cpm_train.py --lsp-root ./data/LSP --ckpt-dir ./model  --summary-dir ./summary --cuda

python -W ignore::UserWarning cpm_train.py --lsp-root ./data/LSP --ckpt-dir ./model  --summary-dir ./summary --cuda --wl

More argument for training please refer to cpm_train.py.

[1] Wei, Shih-En, et al. "Convolutional pose machines." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016.