Streaming Captured Towns into a Browser Tab: Level Selection, Rendering and Occlusion for Large Gaussian Splat Scenes on a 2018 Laptop

We describe SHLS, a system that streams Gaussian splat captures of whole towns, from 4M to 33M splats, into a web browser over plain HTTP and draws them with its own WebGPU renderer. It combines a level-of-detail selector that ranks nodes by on-screen error under a splat budget, a byte-dense pinned coarse level that every node shows first, a renderer with a fixed page pool, compute projection and its own radix sort, and a baked table of which tree nodes are hidden from standing positions at two eye heights. On one 2018 laptop with integrated and entry discrete graphics, we compare it with the streamed splat path of PlayCanvas, the engine it grew out of, in four releases and modes, on four captured towns and three camera paths, scored by the same on-screen error metric. In all twelve scenarios SHLS leaves 2.1 to 4.7 times less error-seconds than the best engine configuration for that scenario, and settles at or near full detail where the engine mostly does not. The coarse level reaches full coverage in about half a second where it took 1.6 to 3.2 seconds before the byte-dense layout, and the occlusion table rules out a median 32% to 81% of the nodes frustum culling leaves, depending on the town. SHLS has been the default renderer of a public web application since 16 October 2026. None of its parts is new on its own; we report what the combination measures, and where it falls short.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-07
DOI
https://doi.org/10.5281/zenodo.24039875
Primary Topic
Computer Graphics and Visualization Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Streaming Captured Towns into a Browser Tab: Level Selection, Rendering and Occlusion for Large Gaussian Splat Scenes on a 2018 Laptop

Harsh P
3 links
Zenodo (CERN European Organization for Nuclear Research)
Computer Graphics and Visualization Techniques
preprint

Streaming Captured Towns into a Browser Tab: Level Selection, Rendering and Occlusion for Large Gaussian Splat Scenes on a 2018 Laptop

Harsh P
preprint en

Abstract

We describe SHLS, a system that streams Gaussian splat captures of whole towns, from 4M to 33M splats, into a web browser over plain HTTP and draws them with its own WebGPU renderer. It combines a level-of-detail selector that ranks nodes by on-screen error under a splat budget, a byte-dense pinned coarse level that every node shows first, a renderer with a fixed page pool, compute projection and its own radix sort, and a baked table of which tree nodes are hidden from standing positions at two eye heights. On one 2018 laptop with integrated and entry discrete graphics, we compare it with the streamed splat path of PlayCanvas, the engine it grew out of, in four releases and modes, on four captured towns and three camera paths, scored by the same on-screen error metric. In all twelve scenarios SHLS leaves 2.1 to 4.7 times less error-seconds than the best engine configuration for that scenario, and settles at or near full detail where the engine mostly does not. The coarse level reaches full coverage in about half a second where it took 1.6 to 3.2 seconds before the byte-dense layout, and the occlusion table rules out a median 32% to 81% of the nodes frustum culling leaves, depending on the town. SHLS has been the default renderer of a public web application since 16 October 2026. None of its parts is new on its own; we report what the combination measures, and where it falls short.

Zenodo (CERN European Organization for Nuclear Research)
Computer Graphics and Visualization Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon