Visual Imitation Learning with Patch Rewards