Geometric-priors Guided 3D Reconstruction / Visual Contents Generation

Lecture 1 (short tutorial) — Geometric-priors guided 3D reconstruction. When humans perceive the world, we easily grasp holistic geometric structures — wireframes, room layouts, planes — and these priors greatly facilitate 3D scene reconstruction. This tutorial introduces geometric-primitive extraction methods, representative works leveraging geometric structure for 3D reconstruction, and different geometric-prior representations.

Lecture 2 (talk) — Geometric-prior Guided Visual Contents Generation. Traditional image/video generation methods often suffer from multi-view inconsistency and unrealistic artifacts; regularization in geometry greatly alleviates these issues. This talk shares work on geometric-prior guided visual-contents generation and 3D shape generation for objects and scenes.

Speaker

Shenghua Gao is an Associate Professor at the School of Computing and Data Science, The University of Hong Kong. His research covers 3D reconstruction, image generation, and video understanding, with 120+ papers and 14,800+ citations (H-index 51). He serves as area chair for NeurIPS/CVPR/ICCV/AAAI.