VG-CoT: Towards Trustworthy Visual Reasoning via Grounded Chain-of-Thought

Byeonggeuk Lim, Kyeonghyun Kim, JungMin Yun +1

The advancement of Large Vision-Language Models (LVLMs) requires precise local region-based reasoning that faithfully grounds the model's logic in actual visual evidence. However,…