In computer vision, end-to-end learning is currently limited to supervised learning for a specific task, leading to a narrow view of the learning process. This approach divides vision into low-level, mid-level, and high-level modules, which corresponds to the psychologist's notion of sensation, perception, and cognition.