GP-GS: Gaussian Processes Densification for 3D Gaussian Splatting

TL;DR

GP-GS introduces a Gaussian Process-based densification method to enhance 3D Gaussian Splatting, significantly reducing artefacts and improving rendering quality by learning local mappings from 2D pixels to 3D attributes.

Contribution

The paper presents a novel GP-based densification framework that formulates 3D point cloud refinement as a continuous regression problem, improving upon existing 3DGS methods.

Findings

Achieves up to 1.12 dB PSNR improvement over baselines.

Effectively reduces artefacts and noise in 3D reconstructions.

Enhances rendering fidelity on synthetic and real-world data.

Abstract

3D Gaussian Splatting (3DGS) enables photorealistic rendering but suffers from artefacts due to sparse Structure-from-Motion (SfM) initialisation. To address this limitation, we propose GP-GS, a Gaussian Process (GP) based densification framework for 3DGS optimisation. GP-GS formulates point cloud densification as a continuous regression problem, where a GP learns a local mapping from 2D pixel coordinates to 3D position and colour attributes. An adaptive neighbourhood-based sampling strategy generates candidate pixels for inference, while GP-predicted uncertainty is used to filter unreliable predictions, reducing noise and preserving geometric structure. Extensive experiments on synthetic and real-world benchmarks demonstrate that GP-GS consistently improves reconstruction quality and rendering fidelity, achieving up to 1.12 dB PSNR improvement over strong baselines.