Awesome

STFPM

Official pytorch implementation for the paper entitled "Student-Teacher Feature Pyramid Matching for Anomaly Detection" (BMVC 2021) https://arxiv.org/abs/2103.04257v3

plot

Dataset

Download dataset from MvTec website.

Training

Train a model:

python main.py train --mvtec-ad mvtec_anomaly_detection --category carpet --epochs 200

After running this command, a directory snapshots/carpet should be created.

Testing

Evaluate a model:

python main.py test --mvtec-ad mvtec_anomaly_detection --category carpet --checkpoint snapshots/carpet/best.pth.tar

This command will evaluate the model specified by --checkpoint argument. You may download the pre-trained models here.

For per-region-overlap (PRO) calculation, you may refer to here. Note that it might take a long time for PRO calculation.

Results

You are expected to obtain the same numbers given the pre-trained models.

Category	AUC-ROC(pixel)	AUC-ROC (image)	PRO
carpet	0.990292	0.964286	0.966061
grid	0.989622	0.982456	0.963767
leather	0.990707	0.950747	0.956661
tile	0.969067	0.982323	0.896640
wood	0.964588	0.996491	0.900518
bottle	0.987894	1.000000	0.959157
cable	0.957504	0.935532	0.894954
capsule	0.985730	0.893498	0.895790
hazelnut	0.984715	1.000000	0.952182
meta_nut	0.971789	0.983382	0.948197
pill	0.975505	0.951173	0.965973
screw	0.988481	0.894651	0.948661
toothbrush	0.989551	0.897222	0.926844
transistor	0.819404	0.939167	0.880923
zipper	0.987756	0.961397	0.868873
<b>average</b>	<b>0.970174</b>	<b>0.955488</b>	<b>0.9283467</b>

Citation

If you find the work useful in your research, please cite our papar.

@inproceedings{wang2021student_teacher,
    title={Student-Teacher Feature Pyramid Matching for Anomaly Detection},
    author={Wang, Guodong and Han, Shumin and Ding, Errui and Huang, Di},
    booktitle={The British Machine Vision Conference (BMVC)},
    year={2021}
}