We develop hardware and software together as an end-to-end system optimized for sensing and computational reconstruction. With this principle, we are focusing on four topics: scene representation, extreme vision, inverse graphics, and nano visual computing. We apply our methodology to practical and scientific applications in computer graphics, vision, imaging, display, and robotics.

Scene Representation

Visual world is described as a combination of material, geometry, and illumination. We are interested in developing neural/analytic computational representations for these components. We use these new visual representations as building blocks for our other research topics of inverse graphics, extreme vision, and nano visual computing. 

Exreme Vision

Common cameras capture RGB images, replicating human vision. However, there are many visual information that human cannot faithfully capture, including polarization, spectrum, phase, speed of light, contents over scattering medium such as fog, and high-dynamic-range intensity. We develop end-to-end computational imaging systems, to surpass human vision capability.

Inverse Graphics

Modern computer-graphics rendering can give you photo-realistic imagery generated from scene information: material, geometry, and illumination. We are interested in inverting this process. Given visual observations of a scene, we aim to estimate scene representations that best explain the observations.

Nano Visual Computing

We are inspired by the recent advances in fabricating nano-scale structures. We apply nano technology, specifically metasurfaces, to computer vision, graphics, and AI. This unique combination enables never-seen-before applications and holds potential for revolutionizing conventional visual-computing research developed with bulky refractive optical elements.