The SOMA benchmark introduces 1,632 human-verified questions to evaluate how Multimodal LLMs reason about fine-grained body motion from egocentric video. It tests models across body configuration, motion dynamics, and temporal reasoning capabilities.
HOW THIS AFFECTS YOU
●
researcherThis provides a specialized metric for evaluating spatial and temporal reasoning in MLLMs.
●
designerThis is relevant for developing more intuitive wearable AI and AR interfaces.