I am not sure that's true. It would learn to not focus in texture which is what current nns do.