GaussianPro: 3D Gaussian Splatting with Progressive Propagation

TL;DR

GaussianPro enhances 3D Gaussian Splatting by employing a progressive propagation strategy inspired by multi-view stereo, improving scene densification and rendering quality, especially in texture-less large-scale scenes.

Contribution

The paper introduces GaussianPro, a novel densification method for 3D Gaussian Splatting that uses scene priors and patch matching for better initialization and rendering quality.

Findings

Outperforms 3DGS on large-scale scenes

Achieves 1.15dB PSNR improvement on Waymo dataset

Effectively handles texture-less surfaces in complex scenes

Abstract

The advent of 3D Gaussian Splatting (3DGS) has recently brought about a revolution in the field of neural rendering, facilitating high-quality renderings at real-time speed. However, 3DGS heavily depends on the initialized point cloud produced by Structure-from-Motion (SfM) techniques. When tackling with large-scale scenes that unavoidably contain texture-less surfaces, the SfM techniques always fail to produce enough points in these surfaces and cannot provide good initialization for 3DGS. As a result, 3DGS suffers from difficult optimization and low-quality renderings. In this paper, inspired by classical multi-view stereo (MVS) techniques, we propose GaussianPro, a novel method that applies a progressive propagation strategy to guide the densification of the 3D Gaussians. Compared to the simple split and clone strategies used in 3DGS, our method leverages the priors of the existing reconstructed geometries of the scene and patch matching techniques to produce new Gaussians with accurate positions and orientations. Experiments on both large-scale and small-scale scenes validate the effectiveness of our method, where our method significantly surpasses 3DGS on the Waymo dataset, exhibiting an improvement of 1.15dB in terms of PSNR.