PolyRoof: Precision Roof Polygonization in Urban Residential Building with Graph Neural Networks

2025-03-13

AI Summary

To address low geometric accuracy, frequent topological errors, and poor computational efficiency in automatic roof polygon extraction from complex urban residential buildings, this paper proposes an attention-enhanced Graph Neural Network (GNN) framework for end-to-end joint point-line prediction and multi-task optimization. The method integrates positional regression loss with area segmentation loss and incorporates an attention mechanism to explicitly model structural relationships among roof components, thereby significantly improving geometric fidelity and topological consistency. Evaluated on dense, irregular rooftop scenes, it achieves a point localization error of 1.33 pixels, a line distance error of 14.39 pixels, and a roof reconstruction score of 91.99%, outperforming baseline methods including Re:PolyWorld. The core contributions are (i) an attention-driven graph-structured representation of roof topology and (ii) a unified point-line co-optimization framework. This work provides a scalable, automated solution for high-fidelity urban 3D modeling.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Graph-based Machine Learning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

The growing demand for detailed building roof data has driven the development of automated extraction methods to overcome the inefficiencies of traditional approaches, particularly in handling complex variations in building geometries. Re:PolyWorld, which integrates point detection with graph neural networks, presents a promising solution for reconstructing high-detail building roof vector data. This study enhances Re:PolyWorld's performance on complex urban residential structures by incorporating attention-based backbones and additional area segmentation loss. Despite dataset limitations, our experiments demonstrated improvements in point position accuracy (1.33 pixels) and line distance accuracy (14.39 pixels), along with a notable increase in the reconstruction score to 91.99%. These findings highlight the potential of advanced neural network architectures in addressing the challenges of complex urban residential geometries.

Problem

• Automated extraction of detailed building roof data
• Handling complex variations in urban residential geometries
• Improving accuracy in roof polygon reconstruction

Innovation

• Graph neural networks for roof polygonization
• Attention-based backbones enhance performance
• Area segmentation loss improves accuracy