Learning Representation For Scene Understanding: Epitomes, Crfs, And Cnns