search

sungwool / CFA_for_anomaly_localization

104 stars 22 forks source link

readme

CFA for Target-Oriented Anomaly Localization

PWC

PyTorch implementation of CFA: Coupled-hypersphere-based Feature Adaptation for Target-Oriented Anomaly Localization (CFA).

Getting Started

Install packages with:

$ pip install -r requirements.txt

Dataset

Prepare industrial image as:

train data:
    dataset_path/class_name/train/good/any_filename.png
    [...]

test data:
    dataset_path/class_name/test/good/any_filename.png
    [...]

    dataset_path/class_name/test/defect_type/any_filename.png
    [...]

How to train

Example

python trainer_cfa.py --class_name all --data_path [/path/to/dataset/] --cnn wrn50_2 --size 224 --gamma_c 1 --gamma_d 1

Performance

WideResNet-50

R : resize. C : crop

R+C R CFA++
bottle 100 / 98.6 100 / 98.9 100 / 98.9
cable 99.8 / 98.7 99.8 / 99.0 99.8 / 99.0
capsule 97.3 / 98.9 99.2 / 99.1 99.2 / 99.1
carpet 99.5 / 98.7 99.4 / 99.0 99.5 / 99.0
grid 99.2 / 97.8 99.9 / 98.1 99.9 / 98.1
hazelnut 100 / 98.6 100 / 98.9 100 / 98.9
leather 100 / 99.1 100 / 99.3 100 / 99.3
metalnut 100 / 98.8 100 / 99.1 100 / 99.1
pill 97.9 / 98.6 97.9 / 98.8 97.9 / 98.8
screw 97.3 / 99.0 93.5 / 98.8 97.3 / 99.0
tile 99.4 / 95.8 100 / 96.3 100 / 96.3
toothbrush 100 / 98.8 97.2 / 99.1 100 / 99.1
transistor 100 / 98.3 100 / 98.4 100 / 98.4
wood 99.7 / 94.8 99.2 / 95.0 99.7 / 95.0
zipper 99.6 / 98.6 99.5 / 99.0 99.6 / 99.0
avg. 99.3 / 98.2 99.0 / 98.5 99.5 / 98.5

Reference

[1] https://github.com/byungjae89/SPADE-pytorch

[2] https://github.com/xiahaifeng1995/PaDiM-Anomaly-Detection-Localization-master

[3] https://github.com/pytorch/vision/tree/main/torchvision/models

[4] https://github.com/lukasruff/Deep-SVDD-PyTorch

Citation

@article{lee2022cfa,
  title={CFA: Coupled-hypersphere-based Feature Adaptation for Target-Oriented Anomaly Localization},
  author={Lee, Sungwook and Lee, Seunghyun and Song, Byung Cheol},
  journal={arXiv preprint arXiv:2206.04325},
  year={2022}
}