SolidGS: Consolidating Gaussian Surfel Splatting for
Sparse-View Surface Reconstruction

Zhuowen Shen 1,2   Yuan Liu 3,4   Zhang Chen 2   Zhong Li 2   Jiepeng Wang 5   Yongqing Liang 1  
Zhengming Yu 1   Jingdong Zhang 1   Yi Xu 2   Scott Schaefer 1   Xin Li 1   Wenping Wang 1  
1Texas A&M University   2OPPO US Research   3Nanyang Technological University  
4Hong Kong University of Science and Technology   5University of Hong Kong  
Arxiv 2024

Abstract


We present SolidGS, which reconstructs a consolidated Gaussian field from sparse inputs. Given only three input views, our approach enables high-precision and detailed mesh extraction, and high-quality novel view synthesis, achieved within just three minutes.

Gaussian splatting has achieved impressive improvements for both novel-view synthesis and surface reconstruction from multi-view images. However, current methods still struggle to reconstruct high-quality surfaces from only sparse view input images using Gaussian splatting. In this paper, we propose a novel method called SolidGS to address this problem. We observed that the reconstructed geometry can be severely inconsistent across multi-views, due to the property of Gaussian function in geometry rendering. This motivates us to consolidate all Gaussians by adopting a more solid kernel function, which effectively improves the surface reconstruction quality. With the additional help of geometrical regularization and monocular normal estimation, our method achieves superior performance on the sparse view surface reconstruction than all the Gaussian splatting methods and neural field methods on the widely used DTU, Tanks-and-Temples, and LLFF datasets.


Main Idea

We propose SolidGS, a novel representation that consolidates the opacity of Gaussians by introducing a shared, learnable solidness factor, enabling multi-view consistent geometry rendering. We also introduce a new framework with geometric constraints to train our SolidGS representation, which consists of geometric priors and regularizations.


Reconstruction results on DTU


Our model's reconstruction results and comparisons on the DTU dataset.

We show our mesh, rendered normal, and Novel View Synthesis (NVS) results on the DTU dataset.

Reconstruction results on Tanks-and-Temples (TNT)


Our model's reconstruction results and comparisons on the TNT dataset with 3 input views. Our model reconstructs the geometry with higher completeness and more details.

We show our mesh, rendered depth, rendered normal, and Novel View Synthesis (NVS) results on the DTU dataset.



Check out our paper for more details.