Search papers, labs, and topics across Lattice.
AnomalyVFM addresses the performance gap between VLMs and VFMs in zero-shot anomaly detection by tackling limitations in auxiliary dataset diversity and VFM adaptation. They introduce a three-stage synthetic dataset generation process to increase diversity and employ low-rank feature adapters with a confidence-weighted pixel loss for parameter-efficient adaptation. Experiments demonstrate that AnomalyVFM significantly outperforms existing methods, achieving a 94.1% average image-level AUROC across diverse datasets using RADIO as a backbone.
VFMs like DINOv2 can now beat vision-language models at zero-shot anomaly detection, thanks to a new training scheme that overcomes dataset limitations and adapts VFMs more effectively.
Zero-shot anomaly detection aims to detect and localise abnormal regions in the image without access to any in-domain training images. While recent approaches leverage vision-language models (VLMs), such as CLIP, to transfer high-level concept knowledge, methods based on purely vision foundation models (VFMs), like DINOv2, have lagged behind in performance. We argue that this gap stems from two practical issues: (i) limited diversity in existing auxiliary anomaly detection datasets and (ii) overly shallow VFM adaptation strategies. To address both challenges, we propose AnomalyVFM, a general and effective framework that turns any pretrained VFM into a strong zero-shot anomaly detector. Our approach combines a robust three-stage synthetic dataset generation scheme with a parameter-efficient adaptation mechanism, utilising low-rank feature adapters and a confidence-weighted pixel loss. Together, these components enable modern VFMs to substantially outperform current state-of-the-art methods. More specifically, with RADIO as a backbone, AnomalyVFM achieves an average image-level AUROC of 94.1% across 9 diverse datasets, surpassing previous methods by significant 3.3 percentage points. Project Page: https://maticfuc.github.io/anomaly_vfm/
