I think there's another level of representation that needs to happen for NNs to learn the appearance of an object in a similar way that humans do. I've seen a l

I think there's another level of representation that needs to happen for NNs to learn the appearance of an object in a similar way that humans do. I've seen a lot of research recently on compressed representations, I think applying that to object recognition might yield some interesting results.