PolyGNN: Polyhedron-based Graph Neural Network for 3D Building Reconstruction from Point Clouds
Zhaiyu Chen, Yilei Shi, Liangliang Nan, Zhitong Xiong, Xiao Xiang Zhu
TL;DR
PolyGNN targets efficient reconstruction of compact, watertight 3D building models from point clouds by modeling space with polyhedral primitives and learning their assembly. It replaces continuous occupancy with a piecewise planar occupancy learned via a skeleton-based sampling of polyhedra queries conditioned on a latent shape code. A graph neural network performs end-to-end node classification over polyhedra and their adjacencies to assemble the final model. The method is trained and evaluated on a large synthetic dataset with ground-truth polyhedral labels, with transferability analyses to other cities and real-world data, showing improvements in scalability and reconstruction quality. Code and data are released to support reproducibility.
Abstract
We present PolyGNN, a polyhedron-based graph neural network for 3D building reconstruction from point clouds. PolyGNN learns to assemble primitives obtained by polyhedral decomposition via graph node classification, achieving a watertight and compact reconstruction. To effectively represent arbitrary-shaped polyhedra in the neural network, we propose a skeleton-based sampling strategy to generate polyhedron-wise queries. These queries are then incorporated with inter-polyhedron adjacency to enhance the classification. PolyGNN is end-to-end optimizable and is designed to accommodate variable-size input points, polyhedra, and queries with an index-driven batching technique. To address the abstraction gap between existing city-building models and the underlying instances, and provide a fair evaluation of the proposed method, we develop our method on a large-scale synthetic dataset with well-defined ground truths of polyhedral labels. We further conduct a transferability analysis across cities and on real-world point clouds. Both qualitative and quantitative results demonstrate the effectiveness of our method, particularly its efficiency for large-scale reconstructions. The source code and data are available at https://github.com/chenzhaiyu/polygnn.