The official implementation of "Weakly supervised temporal action localization with actionness-guided false positive suppression".
| Dataset | 0.1 | 0.2 | 0.3 | 0.4 | 0.5 | 0.6 | 0.7 | AVG(0.1:0.5) | AVG(0.1:0.7) |
|---|---|---|---|---|---|---|---|---|---|
| THUMOS14 | 73.5 | 68.8 | 60.8 | 51.3 | 41.0 | 27.5 | 16.5 | 59.1 | 48.5 |
| Dataset | 0.5 | 0.75 | 0.95 | AVG(0.5:0.95) |
|---|---|---|---|---|
| ActivityNet 1.2 | 48.6 | 29.6 | 6.4 | 29.9 |
| ActivityNet 1.3 | 43.9 | 27.1 | 6.3 | 27.3 |
CUDA Version: 11.3
Pytorch: 1.12.0
Numpy: 1.23.5
Python: 3.9.7
Dataset: Download the two-stream I3D features for THUMOS'14 to "DATA_PATH". You can download them from Google Drive.
Update the data_path in "./scripts/train.sh" and "./scripts/inference.sh".
You can train your own model by executing the following command.
bash ./scripts/train.sh
You can download our trained model from here. Then you need to put the model folder "thumos_AFPS" into the "./outputs" folder. You can reproduce the results of our experiment by executing the following command.
bash ./scripts/inference.sh
If this work is helpful for your research, please consider citing our works.
@article{li2024weakly,
title={Weakly supervised temporal action localization with actionness-guided false positive suppression},
author={Li, Zhilin and Wang, Zilei and Liu, Qinying},
journal={Neural Networks},
volume={175},
pages={106307},
year={2024},
publisher={Elsevier}
}