arxivcs.CVcs.AI2026-07-21
OPD-IAD: From Language Judgment to Industrial Anomaly Detection via On-Policy Self-Distillation
Shuimu Chen, Jing Jin, Nan Su, Hongbo Xu, Zebang Cheng, Wenming Yang, et al.
Large vision-language models (LVLMs) have recently shown strong potential for industrial anomaly detection (IAD) by providing image-level anomaly judgments and interpretable defect reasoning. However, current LVLM-based IAD methods still struggle to produce precise pixel-level an…