Bottom-Up Temporal Action Localization with Mutual Regularization (ECCV2020) pdf

Update

2020-12-02 We also provide a pytorch implementation for proposed Mutual Regularization losses in Mutual_Regularization_Loss.py.

Environment Configuration

1. The code is based on tensorflow 1.5 and python3.5
2. Some required python packages: tqdm, matplotlib, pickle, json,

Data Preparation

We use the features provided by paper CMCS-Temporal-Action-Localization [1]. Download and use merge_feature.py in ./data folder to pre-process the features.

(I3D Features)

[1] Liu D, Jiang T, Wang Y. Completeness Modeling and Context Separation for Weakly Supervised Temporal Action Localization[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019: 1298-1307.

We also provide another feature download links：

THUMOS14 link: https://jbox.sjtu.edu.cn/l/pn3mvh pw: ibhg

ActivityNet1.3 link: https://jbox.sjtu.edu.cn/l/vuB3WW pw: yqgt

Training and Testing

step 1: Obtain the proposal results w/o additional proposal scoring.

python main.py

step 2: Obtain the proposal results w/ additional proposal scoring.

python main_pem.py

step 3: Obtain the detection results.

python main_detection.py

Citation

Please cite our paper if you use this code in your research:

@inproceedings{zhao2020bottom,
  title={Bottom-up temporal action localization with mutual regularization},
  author={Zhao, Peisen and Xie, Lingxi and Ju, Chen and Zhang, Ya and Wang, Yanfeng and Tian, Qi},
  booktitle={European Conference on Computer Vision},
  pages={539--555},
  year={2020},
  organization={Springer}
}