GST-Bench: Evaluating Global Spatial Awareness in Video VLMs
August 5, 2026
GST-Bench is a VQA benchmark for global spatial intelligence in video, utilizing 6,790 minutes of synthetic video. It tests a model's ability to perform spatial inference from novel viewpoints and map egocentric observations to global top-down images.
HOW THIS AFFECTS YOU
●
researcherThis provides a more rigorous test for embodied AI spatial reasoning than existing local-viewpoint benchmarks.