Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

FlashGS: Efficient 3D Gaussian Splatting for Large-scale and High-resolution Rendering

About

This work introduces FlashGS, an open-source CUDA Python library, designed to facilitate the efficient differentiable rasterization of 3D Gaussian Splatting through algorithmic and kernel-level optimizations. FlashGS is developed based on the observations from a comprehensive analysis of the rendering process to enhance computational efficiency and bring the technique to wide adoption. The paper includes a suite of optimization strategies, encompassing redundancy elimination, efficient pipelining, refined control and scheduling mechanisms, and memory access optimizations, all of which are meticulously integrated to amplify the performance of the rasterization process. An extensive evaluation of FlashGS' performance has been conducted across a diverse spectrum of synthetic and real-world large-scale scenes, encompassing a variety of image resolutions. The empirical findings demonstrate that FlashGS consistently achieves an average 4x acceleration over mobile consumer GPUs, coupled with reduced memory consumption. These results underscore the superior performance and resource optimization capabilities of FlashGS, positioning it as a formidable tool in the domain of 3D rendering.

Guofeng Feng, Siyan Chen, Rong Fu, Zimu Liao, Yi Wang, Tao Liu, Zhilin Pei, Hengjie Li, Xingcheng Zhang, Bo Dai• 2024

Related benchmarks

TaskDatasetResultRank
3D Gaussian Splatting RenderingMip-NeRF 360 1080p 1.0
FPS1.54e+3
64
3D Gaussian Splatting RenderingMip-NeRF 360 4K 1.0
FPS1.12e+3
64
3D Scene RenderingMip-NeRF 360 & nvcampus COLMAP (test)
PSNR27.65
8
Novel View SynthesisRef-NeRF Real Scenes 50 (test)
PSNR22.93
8
Novel View SynthesisNeRF-Casting Shiny Scenes 51 (test)
PSNR27.1
8
3D Scene RenderingMip-NeRF 360 & nvcampus (vs. gsplat baseline) (test)
PSNR49.74
7
3D Scene RenderingMatrixCity (800th frame)
Memory Before Rendering (GB)6.83
4
Real-time Gaussian Splatting RenderingTanks & Temple Truck 1080p
Rendering Latency (Average) (ms)2.22
3
Real-time Gaussian Splatting RenderingTanks & Temple Truck 4k
Average Rendering Time (ms)3.46
3
Real-time Gaussian Splatting RenderingTanks & Temple 1080p (train)
Average Rendering Time (ms)1.93
3
Showing 10 of 18 rows

Other info

Code

Follow for update