notesum.ai

Published at December 6

Pushing Rendering Boundaries: Hard Gaussian Splatting

cs.CV

Released Date: December 6, 2024

Authors: Qingshan Xu¹, Jiequan Cui, Xuanyu Yi, Yuxuan Wang, Yuan Zhou, Yew-Soon Ong², Hanwang Zhang

Aff.: ¹Nanyang Technological University; ²A*STAR, Singapore

Arxiv: http://arxiv.org/pdf/2412.04826v1

[Uncaptioned image]

Dataset	Mip-NeRF360			Tanks&Temples			Deep Blending
Method $\,\|\,$ Metric	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
Plenoxels [9]	23.08	0.626	0.463	-	-	-	23.06	0.795	0.510
INGP [25]	25.59	0.699	0.331	-	-	-	23.62	0.797	0.423
Mip-NeRF360 [2]	27.69	0.792	0.237	-	-	-	29.40	0.901	0.245
3DGS [14]	27.71	0.825	0.202	24.90	0.850	0.178	29.46	0.899	0.247
Pixel-GS [53]	27.87	0.835	0.176	25.18	0.860	0.157	28.88	0.892	0.251
Mip-Splatting [49]	27.92	0.838	0.175	25.02	0.857	0.166	29.25	0.903	0.240
Scaffold-GS [22]	27.98	0.825	0.208	25.74	0.861	0.170	30.23	0.907	0.245
Effi-HGS (Ours)	28.14	0.831	0.179	26.02	0.870	0.148	30.33	0.908	0.238
HGS (Ours)	28.30	0.837	0.166	26.36	0.880	0.126	30.40	0.908	0.225