Anomaly-Detection-using-Score-based-Perturbation-Resilience

This repository contains the implementation for Anomaly Detection using Score-based Perturbation Resilience [ICCV 2023]

Environments

• Python : 3.8
• CUDA : 11.3

Packages

Pillow==8.4.0
numpy==1.19.5
scikit-learn==0.24.2
torch-ema==0.3
torch==1.12.0
torchvision==0.13.0

Data Preparations

Download MVTEC dataset from [Link]

Train

python train.py --dataset_path ./mvtec/    \
               --save_path ./save/        \
               --class_name all

Test

python test.py --num_iter 3                                    \
               --perturbed_t 1e-3                              \
               --dataset_path ./mvtec/                         \
               --pretrained_weights_path ./save/models/        \
               --class_name all

Pretrained weights

Download pretrained weights from [Google Drive]

Citation

@inproceedings{Anomaly-Detection-using-Score-based-Perturbation-Resilience,
  title={Anomaly Detection using Score-based Perturbation Resilience},
  author={Shin, Woosang and Lee, Jonghyeon and Lee, Taehan and Lee, Sangmoon and Yun, Jong Pil},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={23372--23382},
  year={2023}
}

Acknowledgement

Our repository is inspired by the following repositories. Thank you for their contribution.

• Score SDE
• MahalanobisAD
• Pytorch-UNet