FilterGS: Traversal-Free Parallel Filtering and Adaptive Shrinking for Large-Scale LoD 3D Gaussian Splatting

Abstract

3D Gaussian Splatting has revolutionized neural rendering with real-time performance. However, scaling this approach to large scenes using Level-of-Detail methods faces critical challenges: inefficient serial traversal consuming over 60\% of rendering time, and redundant Gaussian-tile pairs that incur unnecessary processing overhead. To address these limitations, we introduce FilterGS, featuring a parallel filtering mechanism with two complementary filters that select Gaussian elements efficiently without tree traversal. Additionally, we propose a novel GTC metric that quantifies the redundancy of Gaussian-tile key-value pairs. Based on this metric, we introduce a scene-adaptive Gaussian shrinking strategy that effectively reduces redundant pairs. Extensive experiments demonstrate that FilterGS achieves state-of-the-art rendering speeds while maintaining competitive visual quality across multiple large-scale datasets. Project page: https://github.com/xenon-w/FilterGS

Figures (7)

• Figure 1: FilterGS enables efficient rendering of large-scale 3D Gaussian Splatting LoD trees. By integrating parallel filtering and Gaussian shrinking strategies, our method achieves an average rendering speed of nearly 300 FPS across 6 scenes. It significantly outperforms representative state-of-the-art approaches while maintaining competitive reconstruction quality.
• Figure 2: The framework of FilterGS. (a) Given a set of images and SfM points for a large-scale scene, we first train a LoD-GS tree model. (Sec. \ref{['chap:3.1']}) (b) To quantify redundancy of scene-wide Gaussian-tile pairs, we perform a pre-rendering pass over all training views. While following the standard 3DGS pipeline, this stage computes the GTC metric $\bar{G}$ and derives a shrinking threshold $\tau=f(\bar{G})$. (Sec. \ref{['chap:3.3']}) (c) In the formal rendering stage, the pre-computed scaling factor $\tau$ and the tree model are processed by two specialized filters to efficiently select Gaussians for rendering. These Gaussians are adaptively scaled by $\tau$ during AABB formation, significantly reducing redundant key-value pairs before final sorting and $\alpha$-blending. (Sec. \ref{['chap:3.4']})
• Figure 3: Comparison of serial and parallel filtering strategies. LoG starts from level=0 and progressively applies R&C rules to filter internal nodes with oversized 2D radii and child nodes, retaining only leaves or sufficiently small internal nodes. In contrast, FilterGS simultaneously applies the R&L Filter across all levels to preserve nodes with small 2D radii and all leaf nodes, while introducing an Ancestor Filter to exclude nodes whose ancestor paths contain qualified internal nodes. This yields a key advantage: while LoG’s serial filtering requires steps proportional to tree depth (more levels, more steps), FilterGS completes parallel filtering in just two steps (No more levels, no matter levels).
• Figure 4: Schematic Diagram of the Progressive Relationship Between the Four Core Formulas (Eqa. \ref{['eqa:KPC']}, \ref{['eqa:GTC']}, \ref{['eqa:GTC-view']}, \ref{['eqa:define-tau']}) of Sec. \ref{['chap:3.3']}.
• Figure 5: Comparison of Different Shrinking Strategy on College and Modern-Buildinggauuscene. The best results are highlighted in bold. $N_P$ represents the total number of generated key-value pairs. (a) to (d) show the rendered detail: (a) vanilla 3DGS ($3\sigma$ principle); (b) 3DGS with our shrinking strategy ($\lambda_G=0.2$); (c) FilterGS with the $3\sigma$ principle; (d) FilterGS with $\lambda_G=0.2$.