DeGauss: Dynamic-Static Decomposition with Gaussian Splatting for Distractor-free 3D Reconstruction
Abstract
Reconstructing clean, distractor-free 3D scenes from real-world captures remains a significant challenge, particularly in highly dynamic and cluttered settings such as egocentric videos. To tackle this problem, we introduce DeGauss, a simple and robust self-supervised framework for dynamic scene reconstruction based on a decoupled dynamic-static Gaussian Splatting design. DeGauss models dynamic elements with foreground Gaussians and static content with background Gaussians, using a probabilistic mask to coordinate their composition and enable independent yet complementary optimization. DeGauss generalizes robustly across a wide range of real-world scenarios, from casual image collections to long, dynamic egocentric videos, without relying on complex heuristics or extensive supervision. Experiments on benchmarks including NeRF-on-the-go, ADT, AEA, Hot3D, and EPIC-Fields demonstrate that DeGauss consistently outperforms existing methods, establishing a strong baseline for generalizable, distractor-free 3D reconstructionin highly dynamic, interaction-rich environments
Video
Pipeline
Distractor-free Scene Reconstruction
We present an on-the-fly rendering comparison of reconstructed scenes using 3D-GS, SpotlessSplats, and our approach on the Nerf on-the-go and Aria Everyday Activities datasets. Our method achieves high quality rendering quality with smooth depth geometry.
Nerf on-the-go Dataset1
Drag the slider to compare different reconstruction methods
Aria Datasets23
Drag the slider to compare different reconstruction methods on Aria datasets
Results on Epic-Field Dataset4
Unlike NeuralDiff, where the dynamic foreground tends to dominate the reconstruction, our method leverages a decoupled two-branch design to achieve a flexible and accurate dynamic-static decomposition, yielding a static reconstruction rich in fine details. (left to right: Input, Composed Render, Foreground Render, Background Render, -/Compoistion mask)
Dynamic Scene Modeling
Our method can also be used for high-quality and efficient dynamic scene representation with the decoupled foreground-background branches. We show the results on the Neu3D dataset5 and HyperNerf dataset6. (left to right: -/Input, Composed Render, Foreground Render, -/Background Render, Compoistion mask)
Related Links
Welcome to check these interesting works for robust 3d reconstruction and efficent dynamic scene modeling.
SpotlessSplats: Ignoring Distractors in 3D Gaussian Splatting
WildGaussians: 3D Gaussian Splatting in the Wild
DeSplat: Decomposed Gaussian Splatting for Distractor-Free Rendering
HybridGS: Decoupling Transients and Statics with 2D and 3D Gaussian Splatting
NeuralDiff: Segmenting 3D objects that move in egocentric videos
BibTeX
@misc{wang2025degaussdynamicstaticdecompositiongaussian,
title={DeGauss: Dynamic-Static Decomposition with Gaussian Splatting for Distractor-free 3D Reconstruction},
author={Rui Wang and Quentin Lohmeyer and Mirko Meboldt and Siyu Tang},
year={2025},
eprint={2503.13176},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2503.13176},
}