AGS-Mesh: Adaptive Gaussian Splatting and Meshing with Geometric Priors for Indoor Room Reconstruction Using Smartphones

TL;DR

This work tackles indoor room reconstruction from smartphone data by integrating geometric priors into Gaussian Splatting. It introduces Depth Normal Consistency (DNC) and Adaptive Normal Regularization (ANR) to adaptively filter depth and monocular normal priors during optimization, coupled with a depth-aware TSDF and IsoOctree meshing that yields finer, smoother surfaces. Extensive experiments on MuSHRoom and ScanNet++ show improved mesh quality and novel-view synthesis over baselines, demonstrating robust geometry recovery despite noisy sensor depths and monocular priors. The approach is presented as a practical plug-in for existing Gaussian Splatting frameworks, enabling high-fidelity indoor reconstructions with consumer devices and offering a scalable meshing strategy for detailed surfaces.

Abstract

Geometric priors are often used to enhance 3D reconstruction. With many smartphones featuring low-resolution depth sensors and the prevalence of off-the-shelf monocular geometry estimators, incorporating geometric priors as regularization signals has become common in 3D vision tasks. However, the accuracy of depth estimates from mobile devices is typically poor for highly detailed geometry, and monocular estimators often suffer from poor multi-view consistency and precision. In this work, we propose an approach for joint surface depth and normal refinement of Gaussian Splatting methods for accurate 3D reconstruction of indoor scenes. We develop supervision strategies that adaptively filters low-quality depth and normal estimates by comparing the consistency of the priors during optimization. We mitigate regularization in regions where prior estimates have high uncertainty or ambiguities. Our filtering strategy and optimization design demonstrate significant improvements in both mesh estimation and novel-view synthesis for both 3D and 2D Gaussian Splatting-based methods on challenging indoor room datasets. Furthermore, we explore the use of alternative meshing strategies for finer geometry extraction. We develop a scale-aware meshing strategy inspired by TSDF and octree-based isosurface extraction, which recovers finer details from Gaussian models compared to other commonly used open-source meshing tools. Our code is released in https://xuqianren.github.io/ags_mesh_website/.

Figures (10)

• Figure 1:
• Figure 2:
• Figure 3:
• Figure 5: Pipeline Overview. Our approach leverages geometric consistency between normals derived from raw sensor depths and those predicted by a pretrained model to filter out noisy sensor depth data. Likewise, we compare rendered normals from a Gaussian scene with pseudo-normal estimates to dynamically filter uncertainties in normal supervision during optimization. Our adaptive depth and normal regularization terms assist various Gaussian-based frameworks in accurately reconstructing the underlying scene. Furthermore, we propose an adaptive TSDF and octree-based Marching Cubes meshing strategy enabling the extraction of smoother and more geometrically detailed meshes.
• Figure 6: We demonstrate our method with two Gaussian-based methods DN-Splatter turkulainen2024dn and 2DGS Huang2DGS2024 with qualitative visuals of the reconstructed meshes for the "honka" (top) and "coffee_ room" (bottom) scenes from the MuSHRoom dataset. Geometric priors significantly aid in surface optimization for the Gaussian models.