notesum.ai

Published at November 12

GaussianAnything: Interactive Point Cloud Latent Diffusion for 3D Generation

cs.CV

cs.AI

cs.GR

Released Date: November 12, 2024

Authors: Yushi Lan¹, Shangchen Zhou¹, Zhaoyang Lyu², Fangzhou Hong¹, Shuai Yang³, Bo Dai², Xingang Pan¹, Chen Change Loy¹

Aff.: ¹S-Lab, Nanyang Technological University, Singapore; ²Shanghai Artificial Intelligence Laboratory; ³Peking University

Arxiv: http://arxiv.org/abs/2411.08033v1

Refer to caption

Method	FID $\downarrow$	KID(%) $\downarrow$	MUSIQ $\uparrow$	P-FID $\downarrow$	P-KID(%) $\downarrow$	COV(%) $\uparrow$	MMD(‰) $\downarrow$
OpenLRM	38.41	1.87	45.46	35.74	12.60	39.33	29.08
Splatter-Image	48.80	3.65	30.33	19.72	7.03	37.66	30.69
One-2-3-45 (V=12)	88.39	6.34	59.02	72.40	30.83	33.33	35.09
CRM (V=6)	45.53	1.93	64.10	35.21	13.19	38.83	28.91
Lara (V=4)	43.74	1.95	39.37	32.37	12.44	39.33	28.84
LGM (V=4)	19.93	0.55	54.78	40.17	19.45	50.83	22.06
Shape-E	138.53	11.95	31.51	20.98	7.41	61.33	19.17
LN3Diff	29.08	0.89	50.39	27.17	10.02	55.17	19.94
Ours	24.21	0.76	65.17	8.72	3.22	59.50	15.48