1 paper · 1 filter
Yanan Wang, Tingsong Li, Kaixun Jiang +3
Vision-Language Models (VLMs) can generate rich video captions, yet often misidentify which person performs an action or which limb is involved, particularly across camera cuts. Im…