notesum.ai

Published at December 10

Faster and Better 3D Splatting via Group Training

cs.CV

Released Date: December 10, 2024

Authors: Chengbo Wang¹, Guozheng Ma², Yifei Xue¹, Yizhen Lao¹

Aff.: ¹Hunan University; ²Nanyang Technological University

Arxiv: http://arxiv.org/pdf/2412.07608v1

Refer to caption

	Grouping Iterations	Mip-NeRF360 [1]				Tanks&Temples [13]				Deep Blending [9]				Blender [20]
	Grouping Iterations	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Time $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Time $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Time $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	Time $\downarrow$
3D GS [12]	–	27.205	0.815	0.2143	–	23.142	0.841	0.183	–	29.405	0.903	0.2425	–	33.33	0.969	0.030	–
3D GS $*$	–	27.445	0.816	0.2155	26.7	23.697	0.849	0.1764	15.0	29.586	0.904	0.2437	23.9	33.772	0.970	0.0306	6.1
Group Training w/ RS	0 $\sim$ 15K	27.621	0.822	0.2074	26.6	23.773	0.851	0.1703	14.6	29.185	0.901	0.2473	21.9	33.959	0.971	0.0284	5.7
Group Training w/ RS	0 $\sim$ 30K	27.537	0.821	0.2157	22.6	23.703	0.848	0.1823	12.3	29.417	0.907	0.2481	18.5	33.877	0.971	0.0295	5.3
Group Training w/ OPS	0 $\sim$ 15K	27.582	0.820	0.2103	22.5	23.842	0.851	0.1726	12.8	29.713	0.906	0.2435	19.3	33.889	0.971	0.0292	5.0
Group Training w/ OPS	0 $\sim$ 30K	27.564	0.820	0.2133	19.6	23.853	0.850	0.1764	11.0	29.746	0.907	0.2450	16.9	33.808	0.970	0.0299	4.8