Representation Learning for Vision