When you train a video model for motion, only some clips actually matter. New from NVIDIA Research and recipient of an Outstanding Paper Honorable Mention at #

When you train a video model for motion, only some clips actually matter. New from NVIDIA Research and recipient of an Outstanding Paper Honorable Mention at #ICML2026: MOTIVE shows exactly which training clips improve motion. MOTIVE re-weights training signals toward moving regions and away from …