Enables continuous Level of Detail (LoD) for 3D Gaussian Splatting without the typical trade-off in full-capacity rendering quality.
arXiv · March 20, 2026 · 2603.19234
The Takeaway
Practitioners can now use a single model to serve different hardware (mobile to workstation) by simply rendering a prefix of the Gaussians, providing smooth speed-quality trade-offs previously impossible without discrete, multi-model management.
From the abstract
The ability to render scenes at adjustable fidelity from a single model, known as level of detail (LoD), is crucial for practical deployment of 3D Gaussian Splatting (3DGS). Existing discrete LoD methods expose only a limited set of operating points, while concurrent continuous LoD approaches enable smoother scaling but often suffer noticeable quality degradation at full capacity, making LoD a costly design decision. We introduce Matryoshka Gaussian Splatting (MGS), a training framework that ena