Supervised Anomaly Detection for Complex Industrial Images

Official code for our CVPR 2024 paper

VAD repository

Get Started

Enviroment

pip install -r requirements.txt

Data

1. Download segmentation maps for VisA.
2. Download anomaly maps for EfficientAD.
3. Download anomaly maps for RD4AD.
4. Data structure should look as following:

data
|-- visa_segm
|-- anomaly_maps
|-----|--efficient_ad
|-----|--rd4ad

Train and evaluate

Only VisA dataset is available for now. List of available models: ["efficient_ad", "rd4ad", "all_ad"]. "all_ad" includes both EfficientAD and RD4AD.

python main.py --model efficient_ad

Results

Cl. AUROC (image-level) for SegAD with different sources of anomaly maps.

model	mean	candle	capsules	cashew	chewinggum	fryum	macaroni1	macaroni2	pcb1	pcb2	pcb3	pcb4	pipe_fryum
RD4AD + SegAD	95.3	98.5	80.2	98.9	99.4	96.1	97.4	90.7	96.4	96.3	94.1	99.9	95.8
EfficientAD + SegAD	98.3	98.7	89.7	98.6	99.9	98.6	99.5	98.1	99.5	99.7	98.4	99.3	99.2
All AD + SegAD	98.4	99.0	90.7	99.0	99.9	98.5	99.4	98.1	99.2	99.7	98.3	99.8	99.1

Acknowledgement

We use EfficientAD and Anomalib for baseline anomaly detection models. We are thankful for their amazing work!

Citation

Please cite this paper if it helps your project:

@misc{baitieva2024supervised,
      title={Supervised Anomaly Detection for Complex Industrial Images}, 
      author={Aimira Baitieva and David Hurych and Victor Besnier and Olivier Bernard},
      year={2024},
      eprint={2405.04953},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}