Error rates (green bars) on a standard image classification benchmark have decreased rapidly in the last few years. AlexNet (2012) was the first deep network used on this task. While not the first deep network in computer vision, AlexNet improved image recognition performance overnight, and attracted many computer vision researchers to deep learning. Networks have become by-and-large deeper and deeper since then (red plot). This figure is from a 2018 survey on deep learning by Samira Pouyanfar et al.
This course explores concepts in visual recognition, visual motion analysis, and visual reconstruction through a sample of techniques. There will be lectures, class discussion, and homework.
The class meets in room 2231 of the French Science Center Building.
COMPSCI 527, Duke University, Site based on the fluid 960 grid system