Towards Zero-Shot Anomaly Detection and Reasoning with Multimodal Large Language Models
·11640 words·55 mins
Introduces a specialist visual assistant, Anomaly-OV, leveraging an anomaly expert and visual token selection mechanism to improve zero-shot anomaly detection and reasoning, establishing new datasets and benchmarks in the domain.