LOD-GS: Level-of-Detail-Sensitive 3D Gaussian Splatting for Detail Conserved Anti-Aliasing

TL;DR

This paper introduces LOD-GS, a novel level-of-detail-sensitive filtering framework for 3D Gaussian Splatting that adaptively reduces aliasing artifacts by predicting optimal filtering strengths based on sampling rate, camera distance, and focal length.

Contribution

The authors propose a new filtering framework that dynamically predicts filtering strength for Gaussian primitives, incorporating sampling rate sensitivity and a comprehensive dataset for evaluation.

Findings

Achieves state-of-the-art rendering quality with reduced aliasing.

Effectively models appearance variations based on sampling rate.

Demonstrates robustness across public and new datasets.

Abstract

Despite the advancements in quality and efficiency achieved by 3D Gaussian Splatting (3DGS) in 3D scene rendering, aliasing artifacts remain a persistent challenge. Existing approaches primarily rely on low-pass filtering to mitigate aliasing. However, these methods are not sensitive to the sampling rate, often resulting in under-filtering and over-smoothing renderings. To address this limitation, we propose LOD-GS, a Level-of-Detail-sensitive filtering framework for Gaussian Splatting, which dynamically predicts the optimal filtering strength for each 3D Gaussian primitive. Specifically, we introduce a set of basis functions to each Gaussian, which take the sampling rate as input to model appearance variations, enabling sampling-rate-sensitive filtering. These basis function parameters are jointly optimized with the 3D Gaussian in an end-to-end manner. The sampling rate is influenced by both focal length and camera distance. However, existing methods and datasets rely solely on down-sampling to simulate focal length changes for anti-aliasing evaluation, overlooking the impact of camera distance. To enable a more comprehensive assessment, we introduce a new synthetic dataset featuring objects rendered at varying camera distances. Extensive experiments on both public datasets and our newly collected dataset demonstrate that our method achieves SOTA rendering quality while effectively eliminating aliasing. The code and dataset have been open-sourced.