FastGS: Training 3D Gaussian Splatting in 100 Seconds

TL;DR

FastGS introduces a new acceleration framework for 3D Gaussian splatting that significantly reduces training time while maintaining quality, by using multi-view consistency for Gaussian importance regulation.

Contribution

It proposes a simple, general densification and pruning strategy based on multi-view consistency, eliminating the need for a budgeting mechanism in 3D Gaussian splatting training.

Findings

Achieves 3.32× training acceleration on Mip-NeRF 360.

Demonstrates 15.45× faster training compared to vanilla 3DGS.

Delivers 2-7× acceleration across various 3D reconstruction tasks.

Abstract

The dominant 3D Gaussian splatting (3DGS) acceleration methods fail to properly regulate the number of Gaussians during training, causing redundant computational time overhead. In this paper, we propose FastGS, a novel, simple, and general acceleration framework that fully considers the importance of each Gaussian based on multi-view consistency, efficiently solving the trade-off between training time and rendering quality. We innovatively design a densification and pruning strategy based on multi-view consistency, dispensing with the budgeting mechanism. Extensive experiments on Mip-NeRF 360, Tanks & Temples, and Deep Blending datasets demonstrate that our method significantly outperforms the state-of-the-art methods in training speed, achieving a 3.32$\times$ training acceleration and comparable rendering quality compared with DashGaussian on the Mip-NeRF 360 dataset and a 15.45$\times$ acceleration compared with vanilla 3DGS on the Deep Blending dataset. We demonstrate that FastGS exhibits strong generality, delivering 2-7$\times$ training acceleration across various tasks, including dynamic scene reconstruction, surface reconstruction, sparse-view reconstruction, large-scale reconstruction, and simultaneous localization and mapping. The project page is available at https://fastgs.github.io/