LANP-UVAD

Official implementation of "Learning Anomalies with Normality Prior for Unsupervised Video Anomaly Detection, ECCV-2024".

1. Dependencies

python==3.7
pytorch==1.9.0
scikit-learn==0.24.2
scipy==1.7.3
tqdm==4.65.0

Platform: NVIDIA GeForce RTX 2080 Ti

2. Usage

Setup

• Please download the extracted features for ShanghaiTech and UCF-Crime dataset from links: ShanghaiTech features, UCF-Crime features. The above features use the RexNext-101 to extract from this repo

• Please download ShanghaiTech dataset from this repo and put the testing folder to data/shanghaitech/

• Change the file paths to the download datasets above in config/config_sh.yaml and config/config_ucf.yaml

Train and test the LANP-UVAD

After the setup, simply run the following commands:

python main.py --load_config config/config_sh.yaml

python main.py --load_config config/config_ucf.yaml

Citation

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

@inproceedings{park2020learning,
  title={Learning Anomalies with Normality Prior for Unsupervised Video Anomaly Detection},
  author={Shi, Haoyue and Wang, Le and Zhou, Sanping and Hua, Gang and Tang, Wei},
  booktitle={European Conference on Computer Vision},
  year={2024}
}