Abstract

3D Gaussian splatting (3DGS) is a popular radiance field method, with manyapplication-specific extensions. Most variants rely on the same core algorithm:depth-sorting of Gaussian splats then rasterizing in primitive order. Thisensures correct alpha compositing, but can cause rendering artifacts due tobuilt-in approximations. Moreover, for a fixed representation, sorted renderingoffers little control over render cost and visual fidelity. For example, andcounter-intuitively, rendering a lower-resolution image is not necessarilyfaster. In this work, we address the above limitations by combining 3D Gaussiansplatting with stochastic rasterization. Concretely, we leverage an unbiasedMonte Carlo estimator of the volume rendering equation. This removes the needfor sorting, and allows for accurate 3D blending of overlapping Gaussians. Thenumber of Monte Carlo samples further imbues 3DGS with a way to trade offcomputation time and quality. We implement our method using OpenGL shaders,enabling efficient rendering on modern GPU hardware. At a reasonable visualquality, our method renders more than four times faster than sortedrasterization.