This course explores the construction of virtual environments through the paradigm of World Models, equipping students with advanced techniques in 3D reconstruction and generation. The curriculum comprehensively covers four core pillars: geometric reconstruction (traditional and feed-forward networks), appearance modeling (neural rendering, 3D Gaussian Splatting, and relighting), dynamic simulation from video, and AI-driven behavior modeling. By synthesizing these concepts, students will critically examine and apply three key types of world models: navigable interactive environments, autonomous driving simulators, and world models for embodied robotics.
Unlock more information by signing in with your HKUST account.