bfshi/DGAM-Weakly-Supervised-Action-Localization

Code for our paper "Weakly-Supervised Action Localization by Generative Attention Modeling" (CVPR2020)

DGAM-Weakly-Supervised-Action-Localization

Code for our paper "Weakly-Supervised Action Localization by Generative Attention Modeling" by Baifeng Shi, Qi Dai, Yadong Mu, Jingdong Wang, CVPR2020.

Requirements

Required packges are listed in requirements.txt. You can install by running:

pip install -r requirements.txt

Dataset

We provide extracted features and corresponsing annotations for

• THUMOS14: Google Drive / Baidu Wangpan (pwd: pvfj)
• ActivityNet1.2: Google Drive / Baidu Wangpan (pwd: zhqy)

Before running the code, please download the target dataset and unzip it under data/.

Running

You can train your own model by running:

python train_all.py

Note that you can configure the hyperparameters in /lib/core/config.py.

To test your model, you shall first go to the file /lib/core/config.py and change the entries config.TEST.STATE_DICT_RGB and config.TEST.STATE_DICT_FLOW, then run:

python test.py

Citation

If you find our code useful, please consider citing:

@article{shi2020weakly,
  title={Weakly-Supervised Action Localization by Generative Attention Modeling},
  author={Shi, Baifeng and Dai, Qi and Mu, Yadong and Wang, Jingdong},
  journal={arXiv preprint arXiv:2003.12424},
  year={2020}
}