Deep learning for assistive computer vision