Towards Annotation-Free Validation of MLLMs: A Vision-Language Logical Consistency Metric
arXiv:2605.06201v1 Announce Type: new
Abstract: Dominant accuracy evaluation might reward unwarranted guessing of Large Language Models, and it might not be applicable to novel tasks for model validation without ground-truth (gt) annotation. Based on …