Weakly and Semi Supervised Human Body Part Parsing via Pose-Guided Knowledge Transfer

[arXiv]

Transferring human body part parsing labels to raw images by exploiting the anatomical similarity. Some transferred results:

<p align="center"> <img src="imgs/transferred.jpg", width="700px"> </p>

These results are used as extra training samples for the parsing network and can improve the part segmentation results:

<p align="center"> <img src="imgs/comparison.gif", width="400px"> </p>

Getting Started

Demo video

Check out our demo video here.

Parsing Network

Checkout parsing_network for training\testing\demo code of our parsing network.

Data generation

Checkout data_generation for code of using keypoints similarity to transfer parsing knowledge and generate synthetic training labels.

Feedback

If you get any problems during usage, please open an issue.

Citation

If you use this code for your research, please cite our paper:

@article{fang2018wshp,
  title={Weakly and Semi Supervised Human Body Part Parsing via Pose-Guided Knowledge Transfer},
  author={Fang, Hao-Shu and Lu, Guansong and Fang, Xiaolin and Xie, Jianwen and Tai, Yu-Wing and Lu, Cewu},
  journal={CVPR},
  year={2018}
}