Optimizing 3D Gaussian Splatting via Point Cloud Upsampling

2026-05-29

AI Summary

This work addresses the limitation of 3D Gaussian Splatting (3DGS) reconstruction quality, which is highly dependent on the geometric completeness of the initial point cloud—particularly in regions with rich geometric detail or missing texture. The authors systematically evaluate multiple point cloud upsampling strategies, including interpolation, spline surfaces, moving least squares, and Voronoi-based generation, and further propose a depth-guided point lifting method to enhance geometric consistency. Experimental results demonstrate that surface-based reconstruction excels in organic, detailed scenes, interpolation performs well on piecewise-smooth structures, and the depth-guided strategy substantially improves geometry representation in textureless regions. This study provides the first practical guideline for selecting initialization methods in 3DGS, achieving notable quality improvements on the Mip-NeRF360 and Replica datasets.

Abstract

3D Gaussian Splatting (3DGS) is a technique for creating and rendering 3D scenes, however its performance depends heavily on the quality of initial seed points. To improve 3DGS initialization, this study presents and evaluates several point cloud upsampling approaches: linear interpolation, triangular interpolation, spline-based surface reconstruction, moving least squares surface fitting, and Voronoi-based point generation. Additionally, this research introduces a depth-guided point lifting method that leverages depth maps to maintain geometric consistency with Structure-from-Motion (SfM) reconstructions. Through extensive experiments on the Mip-NeRF360 and Replica datasets, the proposed methods demonstrate improvements in reconstruction quality across diverse scene types. Results indicate that different upsampling strategies excel in different scenarios: surface reconstruction methods perform better with organic, detailed scenes, while simpler interpolation approaches are more suited for scenes dominated by piecewise-smooth geometries. In comparison, the depth-guided approach shows promise for adding geometry-aware points across the entire scene, importantly in texture-less regions. These findings, which provide preliminary practical guidelines for selecting appropriate upsampling methods based on scene characteristics and computational constraints, advances the understanding of how point cloud initialization affects 3DGS quality.

Problem

• 3D Gaussian Splatting
• point cloud upsampling
• initialization quality
• 3D reconstruction
• seed points

Innovation

• 3D Gaussian Splatting
• point cloud upsampling
• depth-guided point lifting
• surface reconstruction
• geometric consistency