This repository contains the framework for Action Recognition in the Dark.
This code is based on PyTorch, you may need to install the following packages:
PyTorch >= 0.4 (perfer 1.2 and above)
opencv-python (pip install)
PILLOW (pip install) (optional for optical flow)
scikit-video (optional for optical flow)
Train with initialization from pre-trained models:
python train_arid11.py --network <Network Name> --is-dark
There are a number of parameters that can be further tuned. We recommend a batch size of 16 per GPU.
We provide several networks that can be utilized, and can be found in the /network folder, change the --network parameter to toggle through the networks
Evaluate the trained model:
cd test
python evaluate_video.py
If models with optical flow is used, the following command is used instead:
cd test
python evaluate_flow.py
- To download the dataset, please write to xuyu0014@e.ntu.edu.sg for the download link. Thank you! [Update!]
- To view our paper, go to this arxiv link
- If you find our paper useful, please cite our paper:
@article{xu2020arid,
title={ARID: A New Dataset for Recognizing Action in the Dark},
author={Xu, Yuecong and Yang, Jianfei and Cao, Haozhi and Mao, Kezhi and Yin, Jianxiong and See, Simon},
journal={arXiv preprint arXiv:2006.03876},
year={2020}
}
- Our code base is adapted from Multi-Fiber Network for Video Recognition, we would like to thank the authors for providing the code base.
- You may contact me through xuyu0014@e.ntu.edu.sg
- This work is licensed under a Creative Commons Attribution 4.0 International License.
