Generation of Appearance-Driven Low-Poly Textured Building Models with Differentiable Rendering
Building models, characterized by their expansive scale and intricate assembly of components, often possess inherently complex geometric forms. Efficiently accessing such models on low-resource hardware such as netbooks and mobile devices remains a significant technical challenge. To address this issue, this paper proposes a novel appearance-driven building mesh simplification approach, building mesh simplification with differential rendering (BMS-DR), which is designed to yield visually consistent low-polygonal building meshes with enhanced textures. Upon receiving a building mesh model as input, the proposed method proceeds in a two-stage process: initially, an approximate building envelope is constructed through techniques including three-dimensional (3D) alpha-shaped hull construction, edge collapse, and UV unwrapping. Then, the geometry and texture of this envelope are refined through the application of differentiable rendering, forming the final low-polygon count model. For extensive evaluation, this paper introduces a novel Blinn–Phong function-based visual error metric and presents a data set comprising 134 distinct building mesh models exhibiting a variety of architectural styles. The experimental results demonstrated that BMS-DR performed well across all building models featuring apertures and recesses, consistently outperforming state-of-the-art methods in multiple visual quality metrics.
Authors
- Weiya Chen (ORCID: https://orcid.org/0000-0001-6505-7559)
- Yizhe Zhang (ORCID: https://orcid.org/0009-0001-5336-8103)
- Lang Chen (ORCID: https://orcid.org/0000-0002-9517-1210)
- Lieyun Ding (ORCID: https://orcid.org/0000-0002-9873-3776)
- Lin Wan (ORCID: https://orcid.org/0000-0001-5037-2986)
Institutions
- Huazhong University of Science and Technology (CN)
Publication Details
- Journal
- Journal of Computing in Civil Engineering
- Published
- 2025-05-19
- DOI
- https://doi.org/10.1061/jccee5.cpeng-6282
- Primary Topic
- Computer Graphics and Visualization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00