Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Clean-GS: Semantic Mask-Guided Pruning for 3D Gaussian Splatting

Subhankar Mishra

TL;DR

Clean-GS addresses the problem of pervasive floaters in 3D Gaussian Splatting by introducing a semantic-mask-guided pruning pipeline. It combines three stages—whitelist filtering, depth-buffered color validation, and neighbor-based outlier removal—operating with sparse semantic supervision (as few as three masks) to isolate target objects. The approach achieves 60-80% model compression on monument datasets while preserving rendering quality, enabling web deployment and AR/VR applications; processing runs in minutes on commodity multi-core CPUs. The work demonstrates that semantic guidance can outperform global importance-based pruning for object isolation in 3DGS and provides open-source code to facilitate adoption in heritage, visualization, and real-time rendering workflows.

Abstract

3D Gaussian Splatting produces high-quality scene reconstructions but generates hundreds of thousands of spurious Gaussians (floaters) scattered throughout the environment. These artifacts obscure objects of interest and inflate model sizes, hindering deployment in bandwidth-constrained applications. We present Clean-GS, a method for removing background clutter and floaters from 3DGS reconstructions using sparse semantic masks. Our approach combines whitelist-based spatial filtering with color-guided validation and outlier removal to achieve 60-80\% model compression while preserving object quality. Unlike existing 3DGS pruning methods that rely on global importance metrics, Clean-GS uses semantic information from as few as 3 segmentation masks (1\% of views) to identify and remove Gaussians not belonging to the target object. Our multi-stage approach consisting of (1) whitelist filtering via projection to masked regions, (2) depth-buffered color validation, and (3) neighbor-based outlier removal isolates monuments and objects from complex outdoor scenes. Experiments on Tanks and Temples show that Clean-GS reduces file sizes from 125MB to 47MB while maintaining rendering quality, making 3DGS models practical for web deployment and AR/VR applications. Our code is available at https://github.com/smlab-niser/clean-gs

Clean-GS: Semantic Mask-Guided Pruning for 3D Gaussian Splatting

TL;DR

Clean-GS addresses the problem of pervasive floaters in 3D Gaussian Splatting by introducing a semantic-mask-guided pruning pipeline. It combines three stages—whitelist filtering, depth-buffered color validation, and neighbor-based outlier removal—operating with sparse semantic supervision (as few as three masks) to isolate target objects. The approach achieves 60-80% model compression on monument datasets while preserving rendering quality, enabling web deployment and AR/VR applications; processing runs in minutes on commodity multi-core CPUs. The work demonstrates that semantic guidance can outperform global importance-based pruning for object isolation in 3DGS and provides open-source code to facilitate adoption in heritage, visualization, and real-time rendering workflows.

Abstract

3D Gaussian Splatting produces high-quality scene reconstructions but generates hundreds of thousands of spurious Gaussians (floaters) scattered throughout the environment. These artifacts obscure objects of interest and inflate model sizes, hindering deployment in bandwidth-constrained applications. We present Clean-GS, a method for removing background clutter and floaters from 3DGS reconstructions using sparse semantic masks. Our approach combines whitelist-based spatial filtering with color-guided validation and outlier removal to achieve 60-80\% model compression while preserving object quality. Unlike existing 3DGS pruning methods that rely on global importance metrics, Clean-GS uses semantic information from as few as 3 segmentation masks (1\% of views) to identify and remove Gaussians not belonging to the target object. Our multi-stage approach consisting of (1) whitelist filtering via projection to masked regions, (2) depth-buffered color validation, and (3) neighbor-based outlier removal isolates monuments and objects from complex outdoor scenes. Experiments on Tanks and Temples show that Clean-GS reduces file sizes from 125MB to 47MB while maintaining rendering quality, making 3DGS models practical for web deployment and AR/VR applications. Our code is available at https://github.com/smlab-niser/clean-gs
Paper Structure (40 sections, 10 equations, 3 figures, 2 tables)

This paper contains 40 sections, 10 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. 3D Gaussian Splatting
  4. Neural Scene Compression
  5. 3DGS Pruning and Compression
  6. Semantic 3D Reconstruction
  7. Object Isolation and Background Removal
  8. Point Cloud Outlier Removal
  9. Method
  10. Stage 1: Whitelist Filtering
  11. Projection.
  12. Whitelist Construction.
  13. Stage 2: Color Validation
  14. Depth Buffering.
  15. Color Matching.
  16. ...and 25 more sections

Figures (3)

  • Figure 1: 3D Gaussian Splatting reconstructions contain massive numbers of floaters: artifacts scattered throughout the scene that obscure objects of interest. Clean-GS removes environmental floaters and background elements, producing clean isolated reconstructions. Temple: 525K→198K Gaussians (62% reduction, 125MB→47MB).
  • Figure 2: Rendered comparison across five viewpoints: front, front alternate, side, back, back alternate. Clean-GS removes environmental floaters and background while preserving temple structure.
  • Figure 3: All 3 semantic masks used for Temple isolation (1% of 302 views). Left: Training images. Right: Mask overlays.