Complexity aware rendering improves 3D scenes at small budgets
KoalaGS: Complexity-Aware Continuous Level of Detail for Gaussian Splatting
Computer Vision and Pattern Recognition
Summary
It can be hard to show detailed 3D scenes on devices with limited computing power. The authors found that parts of a scene that are simple actually use too many pieces and lose quality when the computer tries to save resources. They created a way to guess which parts of the scene need more detail by looking at how complex the images are before making the 3D model. Their method, KoalaGS, keeps the important parts detailed even when the computer has to work with less power, making smoother and better looking 3D views as resources change.
What this means in practice
- •For game developers: Deliver smooth and high-quality 3D scenes on devices with varying computational power by adjusting detail based on image complexity.
- •For mobile app developers: Enable fast progressive loading of 3D scenes that maintain visual quality even on low-power devices by prioritizing important scene parts.
Authors
Radu Beche, Raoul de Charette
Abstract
Deploying a trained 3D Gaussian Splatting scene under shrinking compute budgets requires one primitive set to remain effective when rendered at smaller prefixes. In this continuous level of detail (LoD) setting, the challenge is to decide which Gaussians should retain capacity as the budget decreases. A pilot study on 3DGS-MCMC reconstructions shows that image complexity, computed from the input views before optimization, predicts where reconstruction error concentrates in image and 3D space. It also reveals that simple regions contain too many Gaussians and degrade as much as complex ones at lower budgets, despite being inherently cheaper to represent. Subsequently, in KoalaGS, during training we use this complexity prior to bias relocation toward poorly reconstructed areas while ensuring smooth degradation across LoDs. A primitive importance score then enables inference at low-capacity, while preserving the scene structure. Tested across three datasets, KoalaGS maintains competitive full-capacity quality while significantly improving quality across the budget spectrum, yielding smoother transitions across LoDs, and enabling faster progressive streaming.