notesum.ai

Published at May 10

Expressive Gaussian Human Avatars from Monocular RGB Video

NeurIPS

Released Date: May 10, 2024

Authors: Hezhen Hu¹, Zhiwen Fan¹, Tianhao Walter Wu, Yihan Xi¹, Seoyoung Lee¹, Georgios Pavlakos¹, Zhangyang Wang¹

Aff.: ¹University of Texas at Austin

Arxiv: https://openreview.net/pdf/d3eeb6925f27576c641205c98103704bbe8e32eb.pdf

Refer to caption

Method	N-GS	Full			Hand			Face
Method	N-GS	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
Controlled setting: XHumans dataset
3DGS [20] + SMPLX	19,458	28.88	0.9609	44.93	25.28	0.9189	91.37	25.91	0.9087	101.04
GART [23] + SMPLX	89,571	27.73	0.9553	50.32	25.42	0.9151	99.53	25.86	0.9013	105.06
GauHuman [12] + SMPLX	17,134	29.16	0.9623	41.16	25.69	0.9225	88.16	26.27	0.9124	93.35
EVA	19,993	29.67	0.9632	33.05	26.27	0.9279	72.95	26.56	0.9157	72.30
Real-world setting: UPB dataset
3DGS [20] + SMPLX	21,008	25.31	0.9469	90.80	24.89	0.9425	66.19	24.57	0.9072	136.53
GART [23] + SMPLX	90,676	26.20	0.9511	78.90	25.25	0.9411	61.44	26.62	0.9253	93.28
GauHuman [12] + SMPLX	12,372	25.17	0.9455	84.87	24.67	0.9418	67.61	24.33	0.9035	113.13
EVA	20,829	26.78	0.9519	65.07	27.00	0.9524	45.90	26.85	0.9298	65.90