32×32 bits ought to be enough (to recognize) anybody

Impressive how much information about faces can be conveyed with relatively few bits (in this case, 32*32). (Justification for the machine learning community spending so many years on datasets like MNIST?).

steveicon