Reducing Footskate in Human Motion Reconstruction with Ground Contact Constraints

Official implementation for Reducing Footskate in Human Motion Reconstruction with Ground Contact Constraints.

You would need different python environments to finish each step, please check the corresponding README.md file in each folder. You should also follow the instructions in those files to set up dataset path.

Please see the project page for more details.

NOTE

We use OpenPose (version 1.2) to get 2D pose (Body_25 model), please follow their instructions to install it.

Example

For video demo/dance.mp4, you can follow these steps to get our motion reconstruction result. But you might need to modify some lines to specify the path.

• Check openpose folder: Use OpenPose to get per-frame detection json files. You need to modify some lines to specify the path. I have also put the raw OpenPose results in demo/dance.
• Check ground_detector folder: We need to convert json files to npy format. See op2npy.py.
• Check ground_detector folder: Get ground contact detection results. See inference.py.
• Check motion_reconstruct folder: We need to convert json files to h5 format. See run_openpose.py.
• Check motion_reconstruct folder: Last step, see refine_video.py

I have also put all the intermediate result files in the corresponding path, you should be able to run the last step directly (if you have specify the path correctly).

TODO

• (Probably after ECCV submission) Code clean up for motion reconstruction part

Citation

If you find this code useful for your research, please consider citing the following paper:

@inproceedings{zou2020reducing,
    author    = {Zou, Yuliang and Yang, Jimei and Ceylan, Duygu and Zhang, Jianming and Perazzi, Federico and Huang, Jia-Bin}, 
    title     = {Reducing Footskate in Human Motion Reconstruction with Ground Contact Constraints}, 
    booktitle = {Winter Conference on Applications of Computer Vision},
    year      = {2020}
}