DeSplat: Decomposed Gaussian Splatting for Distractor-Free Rendering

1Technical University of Munich, 2Aalto University, 3University of Oulu

T L D R :

We model distractors and multi-view inconsistencies in captured frames with view-specific Gaussians in each camera view. We decompose alpha compositing and explicitly separate occluders and the underlying static 3D scene.

Abstract

Gaussian splatting enables fast novel view synthesis in static 3D environments. However, reconstructing real-world environments remains challenging as distractors or occluders break the multi-view consistency assumption required for accurate 3D reconstruction. Most existing methods rely on external semantic information from pre-trained models, introducing additional computational overhead as pre-processing steps or during optimization. In this work, we propose a novel method, DeSplat, that directly separates distractors and static scene elements purely based on volume rendering of Gaussian primitives. We initialize Gaussians within each camera view for reconstructing the view-specific distractors to separately model the static 3D scene and distractors in the alpha compositing stages. DeSplat yields an explicit scene separation of static elements and distractors, achieving comparable results to prior distractor-free approaches without sacrificing rendering speed. We demonstrate DeSplat's effectiveness on three benchmark data sets for distractor-free novel view synthesis.

Pipeline

We decompose 3DGS to model the static scene and per-view distractors explicitly.

Videos

We visualize the spot, statue, and yoda scenes, highlighting the differences between ours and the Splatfacto.

Qualitative Visuals

Scene decomposition: We visualize the result of optimized distractor Gaussians, the static 3D scene, and the combined alpha-composited images.

Distractor Gaussians Static scene Combined
Distractor Gaussians Static scene Combined
Distractor Gaussians Static scene Combined

Baseline comparisons: We visualize renderings compared to baseline methods.

Splatfacto SpotLessSplats Ours GT
Splatfacto SpotlessSplats Ours GT
Splatfacto SpotlessSplats Ours GT
Splatfacto WildGaussians Ours GT
Splatfacto WildGaussians Ours GT
Splatfacto SpotlessSplats Ours GT

BibTeX

@article{wang2024desplat,
        title={{DeSplat}: {D}ecomposed {G}aussian Splatting for Distractor-Free Rendering}, 
        author={Yihao Wang and Marcus Klasson and Matias Turkulainen and Shuzhe Wang and Juho Kannala and Arno Solin},
        year={2024},
        journal={arXiv preprint arxiv:2411.19756}
      }