TurboPLY: A high-performance, C++20 PLY library optimized for Gaussian Splatting

[guangxi-research]

Hi

I’ve been following the project and noticed the focus on performance. I recently developed a library called TurboPLY that might be of interest to this community, especially for handling the massive property sets typical in Gaussian Splatting (SH coefficients, opacities, etc.).

TurboPLY:
High-performance, lightweight PLY file I/O library designed for large-scale datasets with zero-copy memory-mapped I/O

https://github.com/guangxi-research/turboply

I have integrated TurboPly into OpenSplat and implemented both Model::saveWithTurboPly() and Model::loadWithTurboPly().
Here are the benchmark results on PLY I/O under different vertex counts and spherical harmonic coefficients:

Number of Vertices	Spherical Harmonic Coefficients	Model::savePly()(s)	Model::saveWithTurboPly(enable_file_mapping:true)(s)	Model::saveWithTurboPly(enable_file_mapping:false)(s)	File Size (KB)	Format
100,000	0	0.422	0.051	0.048	6642	Binary
	9	0.430	0.054	0.062	10157	Binary
	24	0.438	0.130	0.148	16017	Binary
	45	0.448	0.140	0.158	24221	Binary
500,000	0	1.822	0.181	0.245	33204	Binary
	9	1.948	0.328	0.377	50782	Binary
	24	2.003	0.673	0.567	80080	Binary
	45	2.013	0.726	0.959	121096	Binary
10,000,000	0	3.759	0.388	0.466	66407	Binary
	9	3.758	0.582	0.665	101564	Binary
	24	3.793	0.908	1.023	160158	Binary
	45	3.968	1.388	1.504	242190	Binary

Number of Vertices	Spherical Harmonic Coefficients	Model::loadPly()(s)	Model::loadWithTurboPly(enable_file_mapping:true)(s)	Model::loadWithTurboPly(enable_file_mapping:false)(s)	File Size (KB)	Format
100,000	0	0.376	0.027	0.031	6642	Binary
	9	0.357	0.041	0.046	10157	Binary
	24	0.431	0.063	0.073	16017	Binary
	45	0.444	0.090	0.112	24221	Binary
500,000	0	1.782	0.146	0.186	33204	Binary
	9	1.832	0.214	0.220	50782	Binary
	24	1.856	0.310	0.384	80080	Binary
	45	1.868	0.504	0.597	121096	Binary
10,000,000	0	3.578	0.313	0.313	66407	Binary
	9	3.563	0.657	0.780	101564	Binary
	24	3.607	0.941	1.138	160158	Binary
	45	3.642	1.138	1.138	242190	Binary