Semi-Automatic View-Based Segmentation of Gaussian Splat Scenes
Loading...
Date
2026
Journal Title
Journal ISSN
Volume Title
Publisher
The Eurographics Association
Abstract
Gaussian Splatting (GS) has become a widely utilized method for the visualization of highly detailed 3D scenes, capturing small details, surface material information, light interactions, and complex surface shapes. A byproduct of the way GS are generated leaves a large amount of noise around the captured objects and surfaces, making the initial captures unusable without extensive post-processing, often performed manually. Furthermore, selecting and isolating parts of GS reconstructions can be challenging for asset creation. In this paper, we present our initial work on a human-in-the-loop view-based GS segmentation pipeline. We test our system on additional GS scenes and demonstrate that it consistently reduces background noisy splats and can be used to create GS assets. Anonymized code for the prototype: https://github.com/Bisgaardo/Gaussian-Splatting-Segmentation-Project
Description
CCS Concepts: Computing methodologies → Image segmentation; Machine learning
@inproceedings{10.2312:egp.20261013,
booktitle = {Eurographics 2026 - Posters},
editor = {Gerrits, Tim and Teschner, Matthias},
title = {{Semi-Automatic View-Based Segmentation of Gaussian Splat Scenes}},
author = {Bisgaard, Mathias and Møller, Frederik and Nielsen, Jonas Moody and Mørch, Katrine and Baran, Samuel and Gaarsdal, Jesper and Nikolov, Ivan and Madsen, Claus },
year = {2026},
publisher = {The Eurographics Association},
ISSN = {1017-4656},
ISBN = {978-3-03868-300-1},
DOI = {10.2312/egp.20261013}
}