Keypoint Extraction with HRNet on the Volleyball / Collective Activity Dataset

The project is an official implementation of the keypoint extraction part of our ECCV 2022 paper COMPOSER: Compositional Reasoning of Group Activity in Videos with Keypoint-Only Modality.

If you find our repo useful in your research, please use the following BibTeX entry for citation.

@article{zhou2022composer,
  title={COMPOSER: Compositional Reasoning of Group Activity in Videos with Keypoint-Only Modality},
  author={Zhou, Honglu and Kadav, Asim and Shamsian, Aviv and Geng, Shijie and Lai, Farley and Zhao, Long and Liu, Ting and Kapadia, Mubbasir and Graf, Hans Peter},
  journal={Proceedings of the 17th European Conference on Computer Vision (ECCV 2022)},
  year={2022}
}

Installation

Please follow the instructions here for installation.

Preparation after Installation

• Step 1: please clone the official HRNet repo.

git clone https://github.com/leoxiaobin/deep-high-resolution-net.pytorch

• Step 2: download the pretrained HRNet model checkpoints using this link, and put the downloaded folder models inside the deep-high-resolution-net.pytorch/ directory.
• Step 3: replace the demo folder in the original deep-high-resolution-net.pytorch repository with the demo folder from this repository.
  We have also put the code together and uploaded to this link which you can refer to if you have any questions on the preparation.

Keypoint Extraction on Volleyball

deep-high-resolution-net.pytorch$ cd demo
deep-high-resolution-net.pytorch/demo$ python volleyball_joint_feature_extraction.py --dataset_path path_to_volleyball_videos --track_path path_to_tracks_normalized --save_path path_to_save_extracted_keypoints

Keypoint Extraction on Collective Activity

deep-high-resolution-net.pytorch$ cd demo
deep-high-resolution-net.pytorch/demo$ python collective_joint_feature_extraction.py --dataset_path path_to_collective_activity_videos --track_path path_to_tracks_normalized --save_path path_to_save_extracted_keypoints