Home

Awesome

SSTAP

Pytorch implementation of the paper: "Self-Supervised Learning for Semi-Supervised Temporal Action Proposal" (CVPR-2021) [SSTAP-Paper]

Update

ActivityNet Challenge: [2021 champion solution--(PRN)], [2020 champion solution--(CBR-Net)]

Slowfast: [Slowfast Paper], [Slowfast Github]

Requirements

The code runs correctly with:

Other versions may also work.

Feature and model weights

Prepare

Generate labeled/unlabeled data (you can also use our files directly)

python gen_unlabel_videos.py

Training and Validation

bash SSTAP.sh | tee log_SSTAP.txt

Acknowledgement

BMN: Boundary-Matching Network

TSN-Feature

Citation

If our code is helpful for your reseach, please cite our paper:

@inproceedings{SSTAP,
  title={Self-Supervised Learning for Semi-Supervised Temporal Action Proposal},
  author={Wang, Xiang and Zhang, Shiwei and Qing, Zhiwu and Shao, Yuanjie and Gao, Changxin and Sang, Nong},
  booktitle={CVPR},
  year={2021}
}

@article{wang2020cbr,
  title={CBR-Net: Cascade Boundary Refinement Network for Action Detection: Submission to ActivityNet Challenge 2020 (Task 1)},
  author={Wang, Xiang and Ma, Baiteng and Qing, Zhiwu and Sang, Yongpeng and Gao, Changxin and Zhang, Shiwei and Sang, Nong},
  journal={arXiv preprint arXiv:2006.07526},
  year={2020}
}

@article{wang2021pro,
  title={Proposal Relation Network for Temporal Action Detection},
  author={Wang, Xiang and Qing, Zhiwu and Huang, Ziyuan and Feng, Yutong and Zhang, Shiwei and Jiang, Jianwen and Tang, Mingqian and Gao, Changxin and Sang, Nong},
  journal={arXiv preprint arXiv:2106.11812},
  year={2021}
}