Bake It Till You Make It: Ultrafast Spatial Texture-Atlas Splatting
A new spatial texture-atlas splatting approach is pushing 3D scene reconstruction toward practical real-time use. The core idea is to decouple the representation: low-frequency geometry and view-dependent appearance stay on 2D surfels, while high-frequency texture detail is stored in a view-independent spatial hash grid that gets baked into a compact texture atlas.
That shift matters because recent 3D Gaussian Splatting variants can look excellent, but fragment rendering gets expensive fast once you start layering in hash-grid appearance features. Baking the detail into a texture atlas leans harder on GPU texture mapping and cuts down the per-fragment cost, which is exactly where these pipelines tend to bog down.
The method also adds sparsity-focused optimization that penalizes semi-transparency and per-primitive falloff. In practice, that prunes away insignificant surfels and produces sparser reconstructions, which should help both runtime performance and memory pressure. The claimed result is up to a fivefold speedup over 3DGS without giving up visual fidelity.
For game teams, the interesting part is not just the benchmark number but the direction: if this holds up outside controlled scenes, it could make splat-based capture and reconstruction more viable for real-time content, virtual production, and fast environment iteration. The paper claims 4K60 on consumer hardware, which puts it squarely in the range developers care about.
“up to a fivefold speedup over 3DGS”
- what
- A new spatial texture-atlas splatting method bakes high-frequency detail into a compact atlas and prunes weak surfels for faster rendering.
- who
- Neel Kelkar, Simon Niedermayr, Kaloian Petkov, Klaus Engel, and Rüdiger Westermann.
- when
- Submitted to arXiv on 15 July 2026.
- impact
- Claims up to 5x faster rendering than 3D Gaussian Splatting and real-time 4K60 on consumer hardware.
Promising performance gains without losing fidelity
Follow graphics updates
See relevant stories in your personalized news feed.
Discussion