notesum.ai

Published at October 20

FrameBridge: Improving Image-to-Video Generation with Bridge Models

cs.MA

cs.AI

cs.CL

cs.LG

Released Date: October 20, 2024

Authors: Yuji Wang¹, Zehua Chen¹, Xiaoyu Chen¹, Jun Zhu¹, Jianfei Chen¹

Aff.: ¹Tsinghua University

Arxiv: https://arxiv.org/abs/2410.15371v1

Refer to caption

Method	UCF-101			MSR-VTT
Method	FVD $\downarrow$	IS $\uparrow$	PIC $\uparrow$	FVD $\downarrow$	CLIPSIM $\uparrow$	PIC $\uparrow$
DynamiCrafter (Reproduce)	503	28.85	0.5506	265	0.2240	0.5093
DynamiCrafter^∗	429	–	0.6078	234	–	0.5803
SEINE	461	22.32	0.6665	176	0.2250	0.4217
FrameBridge (Ours, w/o SAF)	433	28.61	0.5989	228	0.2246	0.5559
FrameBridge (Ours, w/ SAF)	283	42.85	0.7101	83	0.2249	0.7185