arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.09999cs.CVcs.GR
收藏

Faster-GS:分析和改进高斯点云优化

Faster-GS: Analyzing and Improving Gaussian Splatting Optimization

Computer Graphics Lab, TU Braunschweig(图计算机图形实验室,图林根大学)

Florian Hahlbohm, Linus Franke, Martin Eisemann, Marcus Magnor

AI总结:

Faster-GS通过整合和优化3DGS的核心策略,实现了训练速度提升五倍的同时保持视觉质量,为高斯点云优化提供了新的高效基准,并拓展至四维场景重建。

摘要:

最近在三维高斯点云(3DGS)方面的进展集中在加速优化的同时保持重建质量上。然而,许多提出的方法将实现层面的改进与根本算法修改混合在一起,或者以牺牲保真度换取性能,导致研究领域碎片化,使得公平比较变得复杂。在本工作中,我们整合并评估了先前3DGS研究中最有效且广泛适用的策略,并补充了几种新的优化方法。我们进一步研究了框架中未被充分探索的方面,包括数值稳定性、高斯截断和梯度近似。所得到的系统Faster-GS提供了一个经过严格优化的算法,我们对其在全面的基准测试中进行了评估。我们的实验表明,Faster-GS在保持视觉质量的同时,训练速度达到传统方法的五倍,建立了3DGS优化的新成本效益和资源效率基准。此外,我们还展示了优化可以应用于四维高斯重建,从而实现高效的非刚性场景优化。

英文摘要:

Recent advances in 3D Gaussian Splatting (3DGS) have focused on accelerating optimization while preserving reconstruction quality. However, many proposed methods entangle implementation-level improvements with fundamental algorithmic modifications or trade performance for fidelity, leading to a fragmented research landscape that complicates fair comparison. In this work, we consolidate and evaluate the most effective and broadly applicable strategies from prior 3DGS research and augment them with several novel optimizations. We further investigate underexplored aspects of the framework, including numerical stability, Gaussian truncation, and gradient approximation. The resulting system, Faster-GS, provides a rigorously optimized algorithm that we evaluate across a comprehensive suite of benchmarks. Our experiments demonstrate that Faster-GS achieves up to 5$\times$ faster training while maintaining visual quality, establishing a new cost-effective and resource efficient baseline for 3DGS optimization. Furthermore, we demonstrate that optimizations can be applied to 4D Gaussian reconstruction, leading to efficient non-rigid scene optimization.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑