I bet it could do even better by processing input video of people moving in particular ways (e.g. bouncing to jiggle fat).