RetinaGS: Scalable Training for Dense Scene Rendering with Billion-Scale 3D Gaussians

TL;DR

RetinaGS introduces a scalable training method for 3D Gaussian splatting models, enabling high-resolution, large-scale scene rendering with over a billion primitives, surpassing previous quality benchmarks.

Contribution

The paper presents RetinaGS, a novel model parallel training approach for 3D Gaussian splatting that scales to billions of primitives and high resolutions, improving reconstruction quality.

Findings

Training with more primitives improves visual quality.

First billion-primitive 3DGS model trained on large datasets.

Scalable training enables high-resolution scene rendering.

Abstract

In this work, we explore the possibility of training high-parameter 3D Gaussian splatting (3DGS) models on large-scale, high-resolution datasets. We design a general model parallel training method for 3DGS, named RetinaGS, which uses a proper rendering equation and can be applied to any scene and arbitrary distribution of Gaussian primitives. It enables us to explore the scaling behavior of 3DGS in terms of primitive numbers and training resolutions that were difficult to explore before and surpass previous state-of-the-art reconstruction quality. We observe a clear positive trend of increasing visual quality when increasing primitive numbers with our method. We also demonstrate the first attempt at training a 3DGS model with more than one billion primitives on the full MatrixCity dataset that attains a promising visual quality.