Tag
VSTAT is a new benchmark for visual state tracking in videos that reveals perceptual gaps between humans and multimodal LLMs.