- Pytorch implementation of
WOAD: Weakly Supervised Online Action Detection in Untrimmed Videos
- The code is developed with CUDA V9.0, Python 3.6.3
- pip install -r requirements.txt
-
Download Thumos14 annotations from
here
-
Download Thumos14reduced-I3D-JOINTFeatures from
here
-
Put the downloaded annotations under Thumos14reduced-Annotations/ and features under data/
- THUMOS'14
weakly-supervised model
andstrongly-supervised model
python eval.py --pretrained-ckpt MODEL_NAME
python main.py --supervision SUPERVISION_TYPE --model-name NAME_TO_SAVE_MODEL
- If you find this codebase useful, please cite our paper:
@inproceedings{gao2021woad,
title = {WOAD: Weakly Supervised Online Action Detection in Untrimmed Videos},
author = {Mingfei Gao, Yingbo Zhou, Ran Xu, Richard Socher, Caiming Xiong},
booktitle = {CVPR},
year = {2021}
}
- Please send an email to mingfei.gao@salesforce.com if you have questions.
We referenced W-TALC for the code.