GS-Net: Generalizable Plug-and-Play 3D Gaussian Splatting Module

TL;DR

GS-Net is a novel plug-and-play module that enhances 3D Gaussian Splatting by improving generalization across scenes and densifying Gaussian ellipsoids from sparse point clouds, leading to better rendering quality.

Contribution

We introduce GS-Net, the first generalizable, plug-and-play 3DGS module that densifies Gaussian ellipsoids for improved cross-scene performance.

Findings

GS-Net improves PSNR by over 2 dB on conventional viewpoints.

GS-Net enhances rendering quality on novel viewpoints.

Extensive experiments validate GS-Net's robustness and effectiveness.

Abstract

3D Gaussian Splatting (3DGS) integrates the strengths of primitive-based representations and volumetric rendering techniques, enabling real-time, high-quality rendering. However, 3DGS models typically overfit to single-scene training and are highly sensitive to the initialization of Gaussian ellipsoids, heuristically derived from Structure from Motion (SfM) point clouds, which limits both generalization and practicality. To address these limitations, we propose GS-Net, a generalizable, plug-and-play 3DGS module that densifies Gaussian ellipsoids from sparse SfM point clouds, enhancing geometric structure representation. To the best of our knowledge, GS-Net is the first plug-and-play 3DGS module with cross-scene generalization capabilities. Additionally, we introduce the CARLA-NVS dataset, which incorporates additional camera viewpoints to thoroughly evaluate reconstruction and rendering quality. Extensive experiments demonstrate that applying GS-Net to 3DGS yields a PSNR improvement of 2.08 dB for conventional viewpoints and 1.86 dB for novel viewpoints, confirming the method's effectiveness and robustness.