notesum.ai

Published at November 26

StableAnimator: High-Quality Identity-Preserving Human Image Animation

cs.CV

cs.AI

Released Date: November 26, 2024

Authors: Shuyuan Tu¹, Zhen Xing¹, Xintong Han², Zhi-Qi Cheng³, Qi Dai⁴, Chong Luo⁴, Zuxuan Wu¹

Aff.: ¹Shanghai Key Lab of Intell. Info. Processing, School of CS, Fudan University; ²Huya Inc.; ³Carnegie Mellon University; ⁴Microsoft Research Asia

Arxiv: http://arxiv.org/abs/2411.17697v1

Model	L1 (E-4) $\downarrow$	PSNR [20] $\uparrow$	PSNR* [47] $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	CSIM [12] $\uparrow$	FVD $\downarrow$	Mem $\downarrow$
MRAA [39]	3.21 / 3.62	- / 26.62	18.14 / 17.28	0.672 / 0.692	0.296 / 0.313	0.248 / 0.221	284.82 / 540.35	5.4G
DisCo [47]	3.78 / 3.74	29.03 / 25.23	16.55 / 15.21	0.668 / 0.702	0.292 / 0.302	0.315 / 0.267	292.80 / 544.64	18.7G
MagicAnimate [57]	3.13 / 3.23	29.16 / 27.03	- / 17.11	0.714 / 0.746	0.239 / 0.264	0.462 / 0.338	179.07 / 398.94	20.84G
AnimateAnyone [22]	- / 3.15	29.56 / 27.14	- / 17.14	0.718 / 0.759	0.285 / 0.251	0.457 / 0.316	171.90 / 383.45	11.18G
Champ [66]	2.94 / 3.02	29.91 / 27.78	- / 17.35	0.802 / 0.772	0.231 / 0.234	0.350 / 0.304	160.82 / 373.77	13.2G
Unianimate [50]	2.66 / 2.82	30.77 / 27.46	20.58 / 18.64	0.811 / 0.778	0.231 / 0.253	0.479 / 0.347	148.06 / 394.32	6.11G
MimicMotion [64]	5.85 / 3.55	- / 22.94	14.44 / 13.97	0.601 / 0.733	0.416 / 0.370	0.262 / 0.242	326.57 / 604.13	8.6G
ControlNeXt [34]	6.20 / 2.90	- / 25.28	13.83 / 14.84	0.615 / 0.743	0.416 / 0.262	0.360 / 0.264	326.57 / 389.45	12.23G
StableAnimator	2.87 / 2.71	30.81 / 28.85	20.66 / 18.85	0.801 / 0.784	0.232 / 0.223	0.831 / 0.805	140.62 / 349.94	12.50G