This repository contains Pytorch code for our paper Weakly Supervised Temporal Action Localization Using Deep Metric Learning which has been accepted to WACV 2020.
This package requires Pytorch 1.4. Other Pytorch versions should also work. You can create environment using conda:
conda env create -n wsad -f environment.yml
conda activate wsad
Download Thumos14-reduced dataset from here.
python main.py
python main.py --test --ckpt [pretrained weight]
If you find this work useful, please cite our work:
@misc{islam2020weakly,
title={Weakly Supervised Temporal Action Localization Using Deep Metric Learning},
author={Ashraful Islam and Richard J. Radke},
year={2020},
eprint={2001.07793},
archivePrefix={arXiv},
primaryClass={cs.CV}
}