Learning to Be a Depth Camera for Close-Range Human Capture and Interaction