Towards Better Zero-Shot Anomaly Detection under Distribution Shift with CLIP


Jiyao Gao (Sichuan University), Chengxin He (Sichuan University), Lei Duan (Sichuan University), Jie Zuo (Sichuan University)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)Supplementary

Abstract

Industrial anomaly detection is one of the important computer vision applications in the real world, aiming at identifying anomalous products during testing. In this paper, we investigate a more challenging and practical scenario, anomaly detection under distribution shift, where the test set contains samples from different distributions. The distribution shift can be introduced by many environmental conditions such as lighting conditions or shooting angles. To tackle this issue, we utilize CLIP for generating synthetic text samples that mimic images from diverse real-world distributions. A classifier is then trained on these samples to better identify anomalies from various distributions during testing. Extensive experiments on three benchmark datasets show that our approach outperforms the existing state-of-the-art zero-shot anomaly detection methods, achieving a promising performance of anomaly detection under distribution shift.

Citation

@inproceedings{Gao_2024_BMVC,
author    = {Jiyao Gao and Chengxin He and Lei Duan and Jie Zuo},
title     = {Towards Better Zero-Shot Anomaly Detection under Distribution Shift with CLIP},
booktitle = {35th British Machine Vision Conference 2024, {BMVC} 2024, Glasgow, UK, November 25-28, 2024},
publisher = {BMVA},
year      = {2024},
url       = {https://papers.bmvc2024.org/0330.pdf}
}


Copyright © 2024 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection