notesum.ai

Published at December 5

DiCoDe: Diffusion-Compressed Deep Tokens for Autoregressive Video Generation with Language Models

cs.CV

Released Date: December 5, 2024

Authors: Yizhuo Li¹, Yuying Ge², Yixiao Ge², Ping Luo¹, Ying Shan²

Aff.: ¹The University of Hong Kong; ²ARC Lab, Tencent

Arxiv: http://arxiv.org/pdf/2412.04446v1

Refer to caption

Model	CLIPSIM $\uparrow$	FVD $\downarrow$
CogVideo [17]	0.2631	1294
ModelScopeT2V [42]	0.2930	550
Show-1 [56]	0.3072	538
VideoPoet [21]	0.3049	213
Loong [45]	0.2903	274
DiCoDe-Llama3.2-1B	0.2950	367