SAM 3 video segmentation that keeps every object's ID across the clip: one 16-bit label map per frame, plus an optional colored preview.