NoiseGS: Boosting 3D Gaussian Splatting with Positional Noise for Large-Scale Scene Rendering

Simple item page
dc.contributor.authorKweon, Minseongen_US
dc.contributor.authorCheng, Kaien_US
dc.contributor.authorChen, Xuejinen_US
dc.contributor.authorPark, Jinsunen_US
dc.contributor.editorCeylan, Duyguen_US
dc.contributor.editorLi, Tzu-Maoen_US
dc.date.accessioned2025-05-09T09:36:11Z
dc.date.available2025-05-09T09:36:11Z
dc.date.issued2025
dc.description.abstract3D Gaussian Splatting (3DGS) efficiently renders 3D spaces by adaptively densifying anisotropic Gaussians from initial points. However, in complex scenes such as city-scale environments, large Gaussians often overlap with high-frequency regions rich in edges and fine details. In these areas, conflicting per-pixel gradient directions cause gradient cancellation, reducing the overall gradient magnitude and potentially causing Gaussians to remain trapped in suboptimal positions even after densification. To address this, we propose NoiseGS, a novel approach that integrates randomized noise injection into 3DGS, guiding suboptimal Gaussians selected for densification toward more optimal positions. In addition, to mitigate the instability caused by oversized Gaussians, we introduce an ℓp-penalization on the scale of Gaussians. Our method integrates seamlessly with existing heuristicbased optimization and demonstrates strong generalization in reconstructing complex scenes such as MatrixCity and Building.en_US
dc.description.sectionheadersShort Paper 3
dc.description.seriesinformationEurographics 2025 - Short Papers
dc.identifier.doi10.2312/egs.20251042
dc.identifier.isbn978-3-03868-268-4
dc.identifier.issn1017-4656
dc.identifier.pages4 pages
dc.identifier.urihttps://doi.org/10.2312/egs.20251042
dc.identifier.urihttps://diglib.eg.org/handle/10.2312/egs20251042
dc.publisherThe Eurographics Associationen_US
dc.rightsAttribution 4.0 International License
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/
dc.subjectCCS Concepts: Computing methodologies → Point-based models; Computer vision; Machine learning approaches
dc.subjectComputing methodologies → Point
dc.subjectbased models
dc.subjectComputer vision
dc.subjectMachine learning approaches
dc.titleNoiseGS: Boosting 3D Gaussian Splatting with Positional Noise for Large-Scale Scene Renderingen_US
Files
Original bundle
Now showing 1 - 4 of 4
Thumbnail Image
Name:
egs20251042.pdf
Size:
15.74 MB
Format:
Adobe Portable Document Format
Download
No Thumbnail Available
Name:
short1025_mm1.mp4
Size:
11.72 MB
Format:
Video MP4
Download
No Thumbnail Available
Name:
short1025_mm2.mp4
Size:
74.53 MB
Format:
Video MP4
Download
Thumbnail Image
Name:
short1025_supp.pdf
Size:
1.32 MB
Format:
Adobe Portable Document Format
Download
Collections
EG 2025 - Short Papers