Discovering the physical parts of an articulated object class from multiple videos

  Abstract