Speaker: 南亮亮副教授

Inviter: 贾晓红研究员

Title: Polygonal modeling and semantic understanding of 3D urban data

Language: Chinese

Time & Venue: 2026.09.17 15:30-16:00 南楼N212

Abstract: This talk presents a structured approach to reconstructing and understanding 3D urban environments from point clouds. We first briefly review PolyFit, a method that reconstructs lightweight, watertight polygonal surfaces by selecting optimal planar faces through binary optimization. To handle incomplete input data, we then present a parametric point cloud completion method that infers high-level planar primitives rather than individual points, enabling robust reconstruction in challenging conditions. Finally, we introduce SUM Parts, the first large-scale benchmark for part-level semantic segmentation of urban textured meshes, along with an efficient annotation tool. Together, these works provide a unified pipeline from sparse or noisy point clouds to compact, semantically labeled 3D models, advancing applications in urban modeling, simulation, and digital twins.