search

wanweilin / WinCLIP

Unofficial implementation of: WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation, CVPR 2023
MIT License
5 stars 1 forks source link

readme

WinCLIP

Unofficial implementation of: WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation, CVPR 2023 [Paper]

Installation

  1. Clone the repository to your local machine:

    git clone https://github.com/wanweilin/WinCLIP.git

  2. Navigate to the project directory:

    cd WinCLIP
pip install -r requirements.txt

    Usage Examples

python winclip_ac.py

Dataset

Mvtec AD

Todo

  1. Implement segmentation part

Experimental Results

Obj Type AUROC AUPR F1-Max
bottle 0.9833 0.9947 0.9593
cable 0.8802 0.9222 0.8515
capsule 0.6635 0.8956 0.9046
carpet 0.9888 0.9968 0.9773
grid 0.9879 0.9963 0.9825
hazelnut 0.9305 0.9625 0.8921
leather 1.0000 1.0000 1.0000
metal_nut 0.9428 0.9871 0.9278
pill 0.7908 0.9577 0.9156
screw 0.6958 0.8677 0.8561
tile 1.0000 1.0000 0.9941
toothbrush 0.8431 0.9389 0.8889
transistor 0.8946 0.8605 0.8193
wood 0.9838 0.9949 0.9677
zipper 0.8851 0.9667 0.9063
------------- --------- --------- ---------
Avg 0.8980 0.9561 0.9229

Citation

If you find this code useful, please consider citing the original paper:

@InProceedings{Jeong_2023_CVPR,
    author    = {Jeong, Jongheon and Zou, Yang and Kim, Taewan and Zhang, Dongqing and Ravichandran, Avinash and Dabeer, Onkar},
    title     = {WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2023},
    pages     = {19606-19616}
}

Acknowledgements

This project borrows some code from OpenCLip and DRAEM, thanks for their admiring contributions!