AI RESEARCH

VGS-Decoding: Visual Grounding Score Guided Decoding for Hallucination Mitigation in Medical VLMs

arXiv CS.LG

ArXi:2603.20314v1 Announce Type: cross Medical Vision-Language Models (VLMs) often hallucinate by generating responses based on language priors rather than visual evidence, posing risks in clinical applications. We propose Visual Grounding Score Guided Decoding (VGS-Decoding), a