Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SF-Recon: Simplification-Free Lightweight Building Reconstruction via 3D Gaussian Splatting

Zihan Li, Tengfei Wang, Wentian Gan, Hao Zhan, Xin Wang, Zongqian Zhan

TL;DR

SF-Recon presents a direct pipeline to reconstruct lightweight building surfaces from multi-view imagery by extending 3D Gaussian Splatting with normal-gradient-guided optimization and multi-view edge-consistency pruning, followed by a depth-constrained Delaunay triangulation to produce compact, structurally faithful meshes. The method explicitly encodes building structure (rooflines, wall boundaries) in an unsupervised manner and leverages training-depth maps for robust meshing, achieving substantially fewer faces/vertices while maintaining accuracy on the proposed SF dataset. Key contributions include introducing the building-focused Gaussian framework, a multi-view pruning strategy, and a depth-constrained meshing approach that together enable fast, robust lightweight building reconstruction suitable for digital twins and geospatial analytics.

Abstract

Lightweight building surface models are crucial for digital city, navigation, and fast geospatial analytics, yet conventional multi-view geometry pipelines remain cumbersome and quality-sensitive due to their reliance on dense reconstruction, meshing, and subsequent simplification. This work presents SF-Recon, a method that directly reconstructs lightweight building surfaces from multi-view images without post-hoc mesh simplification. We first train an initial 3D Gaussian Splatting (3DGS) field to obtain a view-consistent representation. Building structure is then distilled by a normal-gradient-guided Gaussian optimization that selects primitives aligned with roof and wall boundaries, followed by multi-view edge-consistency pruning to enhance structural sharpness and suppress non-structural artifacts without external supervision. Finally, a multi-view depth-constrained Delaunay triangulation converts the structured Gaussian field into a lightweight, structurally faithful building mesh. Based on a proposed SF dataset, the experimental results demonstrate that our SF-Recon can directly reconstruct lightweight building models from multi-view imagery, achieving substantially fewer faces and vertices while maintaining computational efficiency. Website:https://lzh282140127-cell.github.io/SF-Recon-project/

SF-Recon: Simplification-Free Lightweight Building Reconstruction via 3D Gaussian Splatting

TL;DR

SF-Recon presents a direct pipeline to reconstruct lightweight building surfaces from multi-view imagery by extending 3D Gaussian Splatting with normal-gradient-guided optimization and multi-view edge-consistency pruning, followed by a depth-constrained Delaunay triangulation to produce compact, structurally faithful meshes. The method explicitly encodes building structure (rooflines, wall boundaries) in an unsupervised manner and leverages training-depth maps for robust meshing, achieving substantially fewer faces/vertices while maintaining accuracy on the proposed SF dataset. Key contributions include introducing the building-focused Gaussian framework, a multi-view pruning strategy, and a depth-constrained meshing approach that together enable fast, robust lightweight building reconstruction suitable for digital twins and geospatial analytics.

Abstract

Lightweight building surface models are crucial for digital city, navigation, and fast geospatial analytics, yet conventional multi-view geometry pipelines remain cumbersome and quality-sensitive due to their reliance on dense reconstruction, meshing, and subsequent simplification. This work presents SF-Recon, a method that directly reconstructs lightweight building surfaces from multi-view images without post-hoc mesh simplification. We first train an initial 3D Gaussian Splatting (3DGS) field to obtain a view-consistent representation. Building structure is then distilled by a normal-gradient-guided Gaussian optimization that selects primitives aligned with roof and wall boundaries, followed by multi-view edge-consistency pruning to enhance structural sharpness and suppress non-structural artifacts without external supervision. Finally, a multi-view depth-constrained Delaunay triangulation converts the structured Gaussian field into a lightweight, structurally faithful building mesh. Based on a proposed SF dataset, the experimental results demonstrate that our SF-Recon can directly reconstruct lightweight building models from multi-view imagery, achieving substantially fewer faces and vertices while maintaining computational efficiency. Website:https://lzh282140127-cell.github.io/SF-Recon-project/

Paper Structure

This paper contains 17 sections, 20 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Surface Reconstruction based 3D Gaussian splatting
  4. Lightweight Mesh Reconstruction
  5. Method
  6. Gaussian Splatting for Building Framework
  7. Normal-Gradient-Guided Gaussian Optimization
  8. Multi-View Edge-Consistency Pruning Strategy
  9. Depth-Constrained Delaunay Triangulation Reconstruction
  10. Depth-Constrained 3D--2D Visibility Validation
  11. Visibility-Driven Delaunay Graph-Cut for Surface Extraction
  12. Experimental Datasets
  13. Experiments
  14. Comparisons with Other Methods
  15. Ablation Studies
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Samples of Lightweight Reconstruction Methods, #F denotes the number of triangular faces.
  • Figure 2: The pipeline of SF-Recon
  • Figure 3: The sample of SF dataset. (a) Camera trajectory around the building, (b) Textured 3D model of the building, (c) Ground truth mesh of the building.
  • Figure 4: Qualitative comparison on SF dataset. #F denotes the number of triangular faces.
  • Figure 5: Qualitative comparison under different resolutions on SF dataset.