notesum.ai

Published at May 6

GaussianCube: A Structured and Explicit Radiance Representation for 3D Generative Modeling

NeurIPS

Released Date: May 6, 2024

Authors: Bowen Zhang¹, Yiji Cheng², Jiaolong Yang³, Chunyu Wang³, Feng Zhao¹, Yansong Tang², Dong Chen³, Baining Guo³

Aff.: ¹University of Science and Technology of China; ²Tsinghua University; ³Microsoft Research Asia

Arxiv: https://openreview.net/pdf/31846ee9e75a1424f980048568944a1577ba41e8.pdf

Refer to caption

Representation	Spatially-structured	PSNR $\uparrow$	LPIPS $\downarrow$	SSIM $\uparrow$	Rel. Speed $\uparrow$	Params (M) $\downarrow$
Instant-NGP	✗	33.98	0.0386	0.9809	$1\times$	12.25
Gaussian Splatting	✗	35.32	0.0303	0.9874	$2.58\times$	1.84
Voxels	✓	31.78	0.0676	0.9664	$0.15\times$	67.12
Voxels^∗	✓	30.25	0.0926	0.9541	$0.15\times$	67.12
Triplane	✓	32.61	0.0611	0.9709	$1.05\times$	6.30
Triplane^∗	✓	31.39	0.0759	0.9635	$1.05\times$	6.30
Our GaussianCube	✓	34.94	0.0347	0.9863	$\mathbf{3.33\times}$	0.46