How embedding prior constraints improves coding and decoding strategies in a neural distributed architecture for depth perception