Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

WebSplatter: Enabling Cross-Device Efficient Gaussian Splatting in Web Browsers via WebGPU

Yudong Han, Chao Xu, Xiaodan Ye, Weichen Bi, Zilong Dong, Yun Ma

TL;DR

WebSplatter tackles the challenge of real-time 3D Gaussian Splatting in web browsers across heterogeneous devices by building a fully GPU-driven WebGPU pipeline. It introduces a wait-free hierarchical radix sort to guarantee deterministic, deadlock-free execution and an opacity-aware pre-processing that prunes splats before rasterization, reducing overdraw. The three-stage pipeline—pre-processing, sorting, and rasterization—achieves 1.2x to 4.5x speedups over state-of-the-art web viewers and lowers peak VRAM usage, enabling stable rendering on memory-constrained devices. This work advances cross-platform browser-based neural rendering workflows by delivering high fidelity and interactive performance for large-scale Gaussian-splat scenes.

Abstract

We present WebSplatter, an end-to-end GPU rendering pipeline for the heterogeneous web ecosystem. Unlike naive ports, WebSplatter introduces a wait-free hierarchical radix sort that circumvents the lack of global atomics in WebGPU, ensuring deterministic execution across diverse hardware. Furthermore, we propose an opacity-aware geometry culling stage that dynamically prunes splats before rasterization, significantly reducing overdraw and peak memory footprint. Evaluation demonstrates that WebSplatter consistently achieves 1.2$\times$ to 4.5$\times$ speedups over state-of-the-art web viewers.

WebSplatter: Enabling Cross-Device Efficient Gaussian Splatting in Web Browsers via WebGPU

TL;DR

WebSplatter tackles the challenge of real-time 3D Gaussian Splatting in web browsers across heterogeneous devices by building a fully GPU-driven WebGPU pipeline. It introduces a wait-free hierarchical radix sort to guarantee deterministic, deadlock-free execution and an opacity-aware pre-processing that prunes splats before rasterization, reducing overdraw. The three-stage pipeline—pre-processing, sorting, and rasterization—achieves 1.2x to 4.5x speedups over state-of-the-art web viewers and lowers peak VRAM usage, enabling stable rendering on memory-constrained devices. This work advances cross-platform browser-based neural rendering workflows by delivering high fidelity and interactive performance for large-scale Gaussian-splat scenes.

Abstract

We present WebSplatter, an end-to-end GPU rendering pipeline for the heterogeneous web ecosystem. Unlike naive ports, WebSplatter introduces a wait-free hierarchical radix sort that circumvents the lack of global atomics in WebGPU, ensuring deterministic execution across diverse hardware. Furthermore, we propose an opacity-aware geometry culling stage that dynamically prunes splats before rasterization, significantly reducing overdraw and peak memory footprint. Evaluation demonstrates that WebSplatter consistently achieves 1.2 to 4.5 speedups over state-of-the-art web viewers.
Paper Structure (24 sections, 9 equations, 3 figures, 5 tables)

This paper contains 24 sections, 9 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background
  3. 3D Gaussian Splatting
  4. Web Graphics Rendering API
  5. Design
  6. Overview
  7. Pre-processing Stage
  8. Sort Stage
  9. The Challenge of Sorting on WebGPU
  10. A Wait-Free Radix Sort
  11. Rasterization Stage
  12. Evaluation
  13. Experimental Setup
  14. Cross-Platform Performance
  15. Scalability with Scene Complexity.
  16. ...and 9 more sections

Figures (3)

  • Figure 1: System overview of WebSplatter. The architecture follows a hybrid compute-render pipeline: blue stages represent compute passes, and gray stages represent render passes.
  • Figure 2: Scalability comparison of WebSplatter against baselines across three devices. Each bar group shows the median frame time (ms) for a specific scene. Lower is better. The KeKsBoTer/web-splat baseline crashes when rendering the truck scene on the Redmi K70 Pro, and its result is omitted.
  • Figure 3: Visual correctness verification on the garden scene. Left: Output rendered by the official native CUDA implementation. Right: Output rendered by WebSplatter in Google Chrome.