notesum.ai

Published at November 29

DisCoRD: Discrete Tokens to Continuous Motion via Rectified Flow Decoding

cs.CV

cs.AI

cs.LG

Released Date: November 29, 2024

Authors: Jungbin Cho¹, Junwan Kim¹, Jisoo Kim¹, Minseo Kim¹, Mingu Kang², Sungeun Hong², Tae-Hyun Oh¹, Youngjae Yu¹

Aff.: ¹Yonsei University; ²Sungkyunkwan University

Arxiv: http://arxiv.org/pdf/2411.19527v1

Refer to caption

Methods	FID $\downarrow$	MPJPE $\downarrow$	sJPE $\downarrow$
HumanML3D
MLD (cont.)	0.017	14.7	0.404
T2M-GPT	${0.089}^{\pm{.001}}$	60.0	0.564
+DisCoRD(Ours)	$\mathbf{{0.031}}^{\pm{.001}}$	71.5	0.488
MMM	${0.097}^{\pm{.001}}$	46.9	0.517
+DisCoRD(Ours)	$\mathbf{{0.020}}^{\pm{.001}}$	56.8	0.429
MoMask	${0.019}^{\pm{001}}$	29.5	0.512
+DisCoRD(Ours)	$\mathbf{{0.011}}^{\pm{000}}$	33.3	0.385
KIT-ML
T2M-GPT	${0.470}^{\pm{.010}}$	46.4	0.526
+DisCoRD(Ours)	$\mathbf{{0.284}}^{\pm{.009}}$	58.7	0.395
MoMask	${0.113}^{\pm{.001}}$	37.5	0.384
+DisCoRD(Ours)	${\textbf{0.103}}^{\pm{.002}}$	33.0	0.359