StableGS: A Floater-Free Framework for 3D Gaussian Splatting

2025-03-24

AI Summary

3D Gaussian Splatting (3DGS) suffers from suboptimal local minima due to coupled optimization of opacity and color parameters, leading to floating artifacts and geometric instability. To address this, we propose a dual-opacity Gaussian model that explicitly decouples geometry—modeled via depth-dependent opacity—from surface appearance—governed by color-dependent opacity. We further introduce a cross-view depth consistency constraint and, for the first time, integrate monocular depth priors from DUSt3R to improve reconstruction robustness in texture-poor regions. Our method is fully compatible with standard 3DGS training and requires no additional supervision. Extensive experiments on multiple public benchmarks demonstrate that our approach significantly suppresses floating artifacts, improves geometric accuracy by 12.6%, and achieves average gains of 1.8 dB in PSNR and 0.023 in SSIM—outperforming all existing state-of-the-art methods.

Technology Category

• Computer Vision: 3D Computer Vision
• Search and Optimization: Mixed Discrete/Continuous Search
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Data transparency and provenance

Abstract

Recent years have witnessed remarkable success of 3D Gaussian Splatting (3DGS) in novel view synthesis, surpassing prior differentiable rendering methods in both quality and efficiency. However, its training process suffers from coupled opacity-color optimization that frequently converges to local minima, producing floater artifacts that degrade visual fidelity. We present StableGS, a framework that eliminates floaters through cross-view depth consistency constraints while introducing a dual-opacity GS model to decouple geometry and material properties of translucent objects. To further enhance reconstruction quality in weakly-textured regions, we integrate DUSt3R depth estimation, significantly improving geometric stability. Our method fundamentally addresses 3DGS training instabilities, outperforming existing state-of-the-art methods across open-source datasets.

Problem

• Eliminates floater artifacts in 3D Gaussian Splatting
• Decouples geometry and material properties of translucent objects
• Enhances reconstruction in weakly-textured regions using depth estimation

Innovation

• Cross-view depth consistency eliminates floater artifacts
• Dual-opacity model decouples geometry and material properties
• DUSt3R depth estimation enhances weakly-textured regions