notesum.ai

Published at October 24

cs.CV

cs.AI

cs.CL

cs.GR

cs.LG

Released Date: October 24, 2024

Authors: Jialu Li¹, Yuanzhen Li¹, Neal Wadhwa¹, Yael Pritch¹, David E. Jacobs¹, Michael Rubinstein¹, Mohit Bansal², Nataniel Ruiz¹

Aff.: ¹Google; ²The University of North Carolina at Chapel Hill

Methods	Environment Consistency			Character Consistency			Semantic Alignment
Methods	CLIP-I^E $\uparrow$	DINO^E $\uparrow$	DreamSim^E $\downarrow$	CLIP-I^C $\uparrow$	DINO^C $\uparrow$	DreamSim^C $\downarrow$	CLIP-T $\uparrow$
IP-Adapter (Ye et al., 2023)	0.470	0.381	0.595	0.366	0.139	0.832	0.168
IP-Adapter-Instruct (Rowles et al., 2024)	0.334	0.151	0.832	0.246	0.124	0.872	0.098
StoryDiffusion (Zhou et al., 2024b)	0.528	0.257	0.733	0.629	0.464	0.545	0.242
Ours	0.563	0.322	0.675	0.676	0.470	0.488	0.242