RAS

Official implementation of Context Enhancement with Reconstruction as Sequence for Unified Unsupervised Anomaly Detection, accepted by ECAI 2024.

Preparation

Datasets

Please download the MVTec-AD dataset, VisA dataset, BTAD dataset and MPDD dataset, and put them at ./data.

|-- data
	|-- btad
		|-- 01
		|-- 02
		|-- 03
	|-- mpdd
		|-- bracket_black
		|-- bracket_brown
		...
	|-- mvtec
		|-- bottle
		|-- cable
		...
	|-- visa
		|-- 1cls
			|-- candle
			|-- capsules
			...

Environment

conda create -n ras python=3.8
conda activate ras
conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
pip install -r requirements.txt

Training & Evaluation

train

sh ras_train.sh

eval

sh ras_eval.sh

Acknowledgement

We acknowledge the excellent implementation from UniAD.

Citation

If our code or models help your work, please cite our paper:

@incollection{yang2024context,
  title={Context Enhancement with Reconstruction as Sequence for Unified Unsupervised Anomaly Detection},
  author={Yang, Hui-Yue and Chen, Hui and Liu, Lihao and Lin, Zijia and Chen, Kai and Wang, Liejun and Han, Jungong and Ding, Guiguang},
  booktitle={ECAI 2024},
  pages={2098--2105},
  year={2024},
  publisher={IOS Press}
}