I guess the key is our ability to extrapolate, imagine things from a single picture. When a child is shown a proper image of a cat and it sees a cartoon cat, it

I guess the key is our ability to extrapolate, imagine things from a single picture. When a child is shown a proper image of a cat and it sees a cartoon cat, it has done that extrapolation of cat's body contours. Or rather I would say it is some sort of meta data that is learnt out of each experience like the way we model in OO - class and object instances. We somehow are able to abstract the class out of a image even if it is just a single image and I feel it is the meta data that gets refined over time rather than the storing pixels of actual images.