notesum.ai

Published at December 5

cs.CV

cs.GR

Released Date: December 5, 2024

Authors: Jinbin Bai¹, Wei Chow², Ling Yang³, Xiangtai Li¹, Juncheng Li², Hanwang Zhang¹, Shuicheng Yan¹

Aff.: ¹Skywork AI; ²National University of Singapore; ³Peking University

Methods	L1 $\downarrow$	L2 $\downarrow$	CLIP-I $\uparrow$	DINO $\uparrow$	CLIP-T $\uparrow$
HumanEdit-full
InstructPix2Pix (Brooks et al., 2023)	0.1601	0.0551	0.7716	0.5335	0.2591
MGIE (Fu et al., 2023)	0.1240	0.0535	0.8697	0.7221	0.2661
HIVE SD1.5 (Zhang et al., 2024b)	0.1014	0.0278	0.8526	0.7726	0.2777
MagicBrush (Zhang et al., 2024a)	0.0807	0.0298	0.8915	0.7963	0.2676
HumanEdit-core
InstructPix2Pix (Brooks et al., 2023)	0.1625	0.0570	0.7627	0.5349	0.2533
MGIE (Fu et al., 2023)	0.1294	0.0610	0.8670	0.7359	0.2589
HIVE SD1.5 (Zhang et al., 2024b)	0.1162	0.0373	0.8441	0.7038	0.2563
MagicBrush (Zhang et al., 2024a)	0.0760	0.0283	0.8946	0.8121	0.2619