MVGSR: Multi-View Consistency Gaussian Splatting for Robust Surface Reconstruction

TL;DR

MVGSR tackles robust surface reconstruction with 3D Gaussian Splatting in dynamic scenes by enforcing multi-view consistency. It combines a distractor masking scheme based on cross-view feature consistency, a multi-view contribution-based pruning mechanism to curb floating artifacts, and a multi-view consistency loss to align geometry and color across views. The approach yields competitive geometric accuracy and rendering fidelity on robust datasets (DTU-Robust, TnT-Robust) while maintaining lightweight resource usage. Together, these components enable high-fidelity surface reconstruction in non-static environments with distractors, advancing fast, view-consistent Gaussian-based scene representations.

Abstract

3D Gaussian Splatting (3DGS) has gained significant attention for its high-quality rendering capabilities, ultra-fast training, and inference speeds. However, when we apply 3DGS to surface reconstruction tasks, especially in environments with dynamic objects and distractors, the method suffers from floating artifacts and color errors due to inconsistency from different viewpoints. To address this challenge, we propose Multi-View Consistency Gaussian Splatting for the domain of Robust Surface Reconstruction (\textbf{MVGSR}), which takes advantage of lightweight Gaussian models and a {heuristics-guided distractor masking} strategy for robust surface reconstruction in non-static environments. Compared to existing methods that rely on MLPs for distractor segmentation strategies, our approach separates distractors from static scene elements by comparing multi-view feature consistency, allowing us to obtain precise distractor masks early in training. Furthermore, we introduce a pruning measure based on multi-view contributions to reset transmittance, effectively reducing floating artifacts. Finally, a multi-view consistency loss is applied to achieve high-quality performance in surface reconstruction tasks. Experimental results demonstrate that MVGSR achieves competitive geometric accuracy and rendering fidelity compared to the state-of-the-art surface reconstruction algorithms. More information is available on our project page (https://mvgsr.github.io).

Figures (14)

• Figure 1: MVGSR Performance Across Datasets: RobustNeRFsabour2023robustnerf (line 1), On-the-GoRen2024NeRFonthego (line 2), and TnT tnt (lines 3-4). When input images contain distractors, MVGSR generates distractor masks through multi-view feature contrast, effectively preventing gradient leakage and thereby achieving high-quality surface reconstruction with photorealistic rendering fidelity
• Figure 2: Surface reconstruction in scenes with distractors.
• Figure 3: The detailed architecture of our MVGSR Framework. Images with distractors are fed to the system that makes use of multi-view consistency Gaussian Splatting algorithm to achieve robust surface reconstruction for non-static environments.
• Figure 4: Example of distractors of DTU-Robust and TnT-Robust dataset.
• Figure 5: Reference image (a), nearest image after 7000 iteration (b), rendered image (c), feature similarity map between reference and nearest viewpoints (d), and feature similarity map between reference and rendered images (e). The rendered image can remove some distracting objects, preventing interference with the features of the reference image.