His research interests include computer vision, machine learning and multimedia, aiming at endowing machines with the capability to perceive, understand, reconstruct, and interact with the 3D visual world, with the following focuses recently:

Data-driven models for extracting hierarchical 3D semantics, inferring semantical/geometrical relationships, and rendering high-fidelity 2D/3D contents, based on multi-modal signals (including 2D/3D vision, language, etc) and beyond.