Comment by loa_in_

I think it's because we classify single images which lose 3d information. We expect them to work with single image but not even humans do that (without stereoscopic vision) if we can help it.