Deep learning architectures in computer vision have traditionally relied on feed-forward models, where outputs are ...