notesum.ai

Published at December 6

MixedGaussianAvatar: Realistically and Geometrically Accurate Head Avatar via Mixed 2D-3D Gaussian Splatting

cs.CV

Released Date: December 6, 2024

Authors: Peng Chen¹, Xiaobao Wei, Qingpo Wuwu, Xinyi Wang, Xingyu Xiao, Ming Lu

Aff.: ¹Institute of Software, Chinese Academy of Sciences

Arxiv: http://arxiv.org/pdf/2412.04955v1

Refer to caption

Method	dataset	L2 $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
PointAvatar (Zheng et al. 2023)	NeRSemble	0.0020	25.9	0.887	0.096
INSTA (Zielonka, Bolkart, and Thies 2022)		0.0018	26.8	0.895	0.117
FlashAvatar (Xiang et al. 2024)		0.0016	29.3	0.932	0.075
Gaussian Head Avatar (Xu et al. 2024)		0.0015	30.8	0.945	0.078
GaussianAvatars (Qian et al. 2024)		0.0013	31.4	0.941	0.064
(Ours)		0.0011	31.8	0.953	0.067
NHA (Grassal et al. 2022)	INSTA	0.0024	27.0	0.942	0.043
IMAvatar (Zheng et al. 2022)		0.0021	27.9	0.943	0.061
INSTA (Zielonka, Bolkart, and Thies 2022)		0.0017	28.6	0.944	0.047
FlashAvatar (Xiang et al. 2024)		0.0017	29.2	0.949	0.040
Gaussian Head Avatar (Xu et al. 2024)		0.0016	29.7	0.958	0.043
GaussianAvatars (Qian et al. 2024)		0.0014	29.1	0.953	0.046
(Ours)		0.0012	30.4	0.962	0.039