Learning to parse images of articulated bodies