Pixel-level masks that precisely outline objects across video frames, enabling dense visual grounding beyond bounding boxes.