arxivcs.CV2026-07-31
VFAD: Variational Semantic Prompting Meets Frequency-Adaptive Representation Learning for Zero-Shot Anomaly Detection
Peng Chen, Kaige Li, Wei Wang, Mingbo Yang, Wenqiang Wang, Li Shen, et al.
Zero-shot anomaly detection (ZSAD) aims to detect and localize anomalies in unseen categories without access to target-specific training data. Although recent CLIP-based methods have demonstrated promising generalization through vision-language alignment, they remain limited in c…