About cows, I would say exactly the opposite. We NEVER see 100% visual information about a cow while we recognise it at a variety of distancies, orientation angles and lights. We can recognise a cow just by fragments of its visual information we retain in our brain, and we can also make a mistake with some per cent of probability. But I do not uphold the NN approach either, because it does not target to perform the visible fragments manipulation and interpolation.

About modeling versus sensation. Yes, the brain models the scenes of the world and the whole world as a global scene. And the shortest proof for this statement is modelling scenes of past and future times in compressed pace. We can model a year within minutes or an hour, not another year. We do not sense it. It is imaginable or modelled, if you like.