Skip to content
/ PANDA Public

PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation (CVPR 2021)

License

Notifications You must be signed in to change notification settings

talreiss/PANDA

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PANDA

Official PyTorch implementation of “PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation” (CVPR 2021).

Virtual Environment

Use the following commands:

cd path-to-PANDA-directory
virtualenv venv --python python3
source venv/bin/activate
pip install -r requirements.txt --find-links https://download.pytorch.org/whl/torch_stable.html

Data Preparation

Use the following commands:

cd path-to-PANDA-directory
mkdir data

Download:

Extract these files into path-to-PANDA-directory/data and unzip tiny.zip

Experiments

To replicate the results on CIFAR10, FMNIST for a specific normal class with EWC:

python panda.py --dataset=cifar10 --label=n --ewc --epochs=50
python panda.py --dataset=fashion --label=n --ewc --epochs=50

To replicate the results on CIFAR10, FMNIST for a specific normal class with early stopping:

python panda.py --dataset=cifar10 --label=n
python panda.py --dataset=fashion --label=n

Where n indicates the id of the normal class.

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

OE Experiments

To replicate the results on CIFAR10 for a specific normal class:

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

Where n indicates the id of the normal class.

Further work

See our new paper “Mean-Shifted Contrastive Loss for Anomaly Detection” which achieves state-of-the-art anomaly detection performance on multiple benchmarks including 97.5% ROC-AUC on the CIFAR-10 dataset.

GitHub Repository

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{reiss2021panda,
  title={PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation},
  author={Reiss, Tal and Cohen, Niv and Bergman, Liron and Hoshen, Yedid},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={2806--2814},
  year={2021}
}

About

PANDA: Adapting Pretrained Features for Anomaly Detection and Segmentation (CVPR 2021)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages