JUDO Framework Enhances Industrial Anomaly QA with Domain Knowledge

Researchers have introduced JUDO (Juxtaposed Domain-Oriented Multimodal Reasoner), a framework designed to improve large multimodal models (LMMs) for industrial anomaly detection and question answering. JUDO addresses the lack of domain-specific knowledge in LMMs by incorporating visual and textual reasoning. It segments defect regions through visual comparison of query and normal images, and uses supervised fine-tuning (SFT) and reinforcement learning (GRPO) to enhance domain understanding. The framework aims to generate more accurate responses in complex industrial scenarios.