In our research, we view visual computing as a closed loop: analysis methods (i.e. computer vision) extract rich scene models from visual data (e.g. images, 3D shapes), and synthesis methods (i.e. computer graphics) convert those models back into observable visual data. The two sides of this loop are mutually reinforcing: analysis of large-scale visual data helps to build/learn better synthesis methods (e.g. generative adversarial networks), and the outputs of those synthesis models help to build/learn better analysis methods (e.g. synthetic training data for computer vision systems).

Finally, visual computing systems are designed to be used by people: through visualization of data, interaction with / manipulation of visual scene models, etc. We design visual computing systems that maximize their users' ability to realize their creative visions, to understand complex datasets and phenomena, and to make informed predictions about future events.