GausSurf: Geometry-Guided 3D Gaussian Splatting for Surface Reconstruction

Jiepeng Wang 1   Yuan Liu 2,3   Peng Wang 1   Cheng Lin 1   Junhui Hou 4   Xin Li 5   Taku Komura 1   Wenping Wang 5  
1The University of Hong Kong   2Hong Kong University of Science and Technology   3Nanyang Technological University   
4City University of Hong Kong   5Texas A&M University  

Abstract


3D Gaussian Splatting has achieved impressive performance in novel view synthesis with real-time rendering capabilities. However, reconstructing high-quality surfaces with fine details using 3D Gaussians remains a challenging task. In this work, we introduce GausSurf, a novel approach to high-quality surface reconstruction by employing geometry guidance from multi-view consistency in texture-rich areas and normal priors in texture-less areas of a scene. We observe that a scene can be mainly divided into two primary regions: 1) texture-rich and 2) texture-less areas. To enforce multi-view consistency at texture-rich areas, we enhance the reconstruction quality by incorporating a traditional patch-match based Multi-View Stereo (MVS) approach to guide the geometry optimization in an iterative scheme. This scheme allows for mutual reinforcement between the optimization of Gaussians and patch-match refinement, which significantly improves the reconstruction results and accelerates the training process. Meanwhile, for the texture-less areas, we leverage normal priors from a pre-trained normal estimation model to guide optimization. Extensive experiments on the DTU and Tanks and Temples datasets demonstrate that our method surpasses state-of-the-art methods in terms of reconstruction quality and computation time.

Introduction

In this paper, we present a new 3D Gaussian Surface based method, GausSurf, for efficient and high-quality multiview surface reconstruction with geometric gauidance (1) from patch-match and normal priors. For texture-rich areas, we utilize multi-view consistency constraints to guide the optimization process. For texture-less regions, we incorporate normal priors from a pretrained model to provide supplementary supervision signals. By effectively integrating these geometric priors, our method achieves both high-quality and efficient surface reconstruction.

Reconstruction results on DTU


In this part, we show our model's reconstruction results and comparisons on the DTU dataset.

Our reconstruction results on all the scenes:

Reconstruction results on TnT and Mip-NeRF360


In this part, we show our model's reconstruction results and comparisons on the Tanks and Temples dataset and Mip-NeRF360 dataset.

Our reconstruction results on Mip-NeRF360, each frame shows the rendered RGB, depth, normal from Gaussians and rendered mesh:

Our reconstruction results on TnT:

Citation

@article{wang2024GausSurf,
    title={GausSurf: Geometry-Guided 3D Gaussian Splatting for Surface Reconstruction}, 
    author={Wang, Jiepeng and Liu, Yuan and Wang, Peng and Lin, Cheng and Hou, Junhui and Li, Xin and Komura, Taku and Wang, Wenping},
    journal={arXiv preprint arXiv:2411.19454},
    year={2024}
}
This page is Zotero translator friendly. Page last updated