RTFM

This repo contains the Pytorch implementation of our paper:

Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learning

Yu Tian, Guansong Pang, Yuanhong Chen, Rajvinder Singh, Johan W. Verjans, Gustavo Carneiro.

• Accepted at ICCV 2021.

• SOTA on 4 benchmarks. Check out Papers With Code for Video Anomaly Detection.

Training

Setup

Please download the extracted I3d features for ShanghaiTech and UCF-Crime dataset from links below:

ShanghaiTech train i3d onedirve

ShanghaiTech test i3d onedrive

ShanghaiTech train features on Google dirve

ShanghaiTech test features on Google dirve

checkpoint for ShanghaiTech

Extracted I3d features for UCF-Crime dataset

UCF-Crime train i3d onedirve

UCF-Crime test i3d onedrive

UCF-Crime train I3d features on Google drive

UCF-Crime test I3d features on Google drive

checkpoint for Ucf-crime

The above features use the resnet50 I3D to extract from this repo.

Follow previous works, we also apply 10-crop augmentations.

The following files need to be adapted in order to run the code on your own machine:

• Change the file paths to the download datasets above in list/shanghai-i3d-test-10crop.list and list/shanghai-i3d-train-10crop.list.
• Feel free to change the hyperparameters in option.py

Train and test the RTFM

After the setup, simply run the following commands:

python -m visdom.server
python main.py

Citation

If you find this repo useful for your research, please consider citing our paper:

@article{tian2021weakly,
  title={Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learning},
  author={Tian, Yu and Pang, Guansong and Chen, Yuanhong and Singh, Rajvinder and Verjans, Johan W and Carneiro, Gustavo},
  journal={arXiv preprint arXiv:2101.10030},
  year={2021}
}