Mean-Shifted Contrastive Loss for Anomaly Detection

Official PyTorch implementation of “Mean-Shifted Contrastive Loss for Anomaly Detection” (AAAI 2023).

Virtual Environment

Use the following commands:

cd path-to-directory
virtualenv venv --python python3
source venv/bin/activate
pip install -r requirements.txt

Experiments

To replicate the results on CIFAR-10 for a specific normal class:

python main.py --dataset=cifar10 --label=n

Where n indicates the id of the normal class.

To replicate the results on CIFAR-10 with ResNet18 for a specific normal class:

python main.py --dataset=cifar10 --label=n --backbone=18

Where n indicates the id of the normal class.

Use the --angular flag to jointly optimize the mean-shifted contrastive loss and the angular center loss.

To run experiments on different datasets, please set the path in utils.py to the desired dataset.

Video Anomaly Detection

See our new paper “Attribute-based Representations for Accurate and Interpretable Video Anomaly Detection” which achieves state-of-the-art video anomaly detection performance on multiple benchmarks including 85.9% ROC-AUC on the ShanghaiTech dataset.

GitHub Repository

Citation

If you find this useful, please cite our paper:

@inproceedings{reiss2023mean,
  title={Mean-shifted contrastive loss for anomaly detection},
  author={Reiss, Tal and Hoshen, Yedid},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={37},
  number={2},
  pages={2155--2162},
  year={2023}
}