FSLC: Fast Scoring with Learnable Coreset for Zero-shot Industrial Anomaly Detection


Songtao Ni (Shanghai Jiaotong University), Yuxin Li (Shanghai Jiaotong University), Xu Zhao (Shanghai Jiao Tong University)
The 35th British Machine Vision Conference
PDFPosterVideo (Right click to download)Supplementary

Abstract

This paper presents an efficient approach for zero-shot anomaly classification (AC) and segmentation (AS) in industrial applications. While existing zero-shot anomaly detection methods often rely on supplemental prior knowledge or trade computational speed for performance, our method eliminates dependence on external data and accelerates detection. We propose a dynamic coreset strategy that learns directly from test data and repurposes it for anomaly scoring. The coreset is initially constructed from diverse image patches to comprehensively capture potential data patterns across all test samples. Through iterative expansion and score-based filtration, the coreset progressively refines its representation of normal data distributions. This adaptive process enables quantitative evaluation of anomaly severity based on deviations from the learned norms. Experimental validation across multiple benchmarks demonstrates the method’s effectiveness. On MVTec AD, we achieve state-of-the-art average AUROC scores of 93.52\% (AC) and 96.55\% (AS), while maintaining processing speeds of 5 to 7 frames per second. These results highlight the ability of our framework to balance accuracy and efficiency in practical industrial deployments.

Citation

@inproceedings{Ni_2025_BMVC,
author    = {Songtao Ni and Yuxin Li and Xu Zhao},
title     = {FSLC: Fast Scoring with Learnable Coreset for Zero-shot Industrial Anomaly Detection},
booktitle = {36th British Machine Vision Conference 2025, {BMVC} 2025, Sheffield, UK, November 24-27, 2025},
publisher = {BMVA},
year      = {2025},
url       = {https://bmva-archive.org.uk/bmvc/2025/assets/papers/Paper_922/paper.pdf}
}


Copyright © 2025 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