Research

Co-designing optics, sensors, and algorithms.

Computational imaging leverages the joint design of optics, sensors, and algorithms to overcome the limits of conventional imaging systems. We work across optics, vision, graphics, machine learning, optimization, and hardware to build next-generation systems that capture and reconstruct what conventional cameras cannot.

Vision & Graphics

Vision foundation models & generative models

Vision foundation models, video diffusion transformers, and 4D scene representations that learn the geometry, appearance, and dynamics of the visual world — from animatable portrait avatars to controllable text-to-4D generation and full 3D scene reconstruction.

Physics-Informed Vision

Scientific imaging

Imaging at the frontier of science — combining novel sensors, physics-informed machine learning, and neural fields to reveal new insights about our world and the cosmos. We work across radio, optical, and time-resolved sensing (ALMA, EHT, JWST, SPADs) to study black holes, exoplanets, protoplanetary disks, atmospheric particles, and dark matter — building algorithms for inverse problems, differentiable physics, and dynamic inference at the limits of what can be seen.

Orbital polarimetric tomography of a flare near Sagittarius A*

Light Transport

Neural inverse rendering

Recovering geometry, materials, and the path of light itself from images and transients — using differentiable rendering, neural radiance and transient fields, and physically-based models of light propagating through scenes.

Photon-level Sensing

Single-photon imaging

Capturing scenes one photon at a time, with picosecond timing. We design algorithms and optical systems for single-photon avalanche diodes and coherent receivers — enabling high-resolution depth, ultra-wideband sensing, and full-wavefield lidar.

Active Imaging

Active 3D imaging & structured light

Active 3D capture with coded illumination — jointly designing patterns, the camera, and the inverse reconstruction so that depth is dense, accurate, fast, and robust to global light transport.

Hardware

Computational cameras & sensors

Custom CMOS image sensors and unconventional camera designs — co-designed with downstream algorithms — including programmable pixel-wise exposure, dual-tap pipelines, event-based sensing, snapshot hyperspectral, and superoscillatory imaging.