We recover large-scale static and dynamic environments from images, video and LiDAR. Neural and Gaussian representations, SLAM, real-time rendering and scene generation provide foundations for world models with spatial memory, dynamic prediction and interaction.

We combine images, video, speech, motion, expression and affective cues to reconstruct, generate, animate and edit people, heads and clothing—toward digital humans that can express, communicate and respond naturally.

We study surface and point-cloud processing, rigid and non-rigid registration, geometric representations, sparse optimization and acceleration methods—turning mathematical structure into scalable tools for reconstruction, generation, simulation and design.

We connect computation and physical optics through freeform imaging and non-imaging optics, light-field control, computational imaging, caustic and shadow design, and emerging displays—treating light as a medium that can be designed and optimized.