notesum.ai

Published at December 5

RMD: A Simple Baseline for More General Human Motion Generation via Training-free Retrieval-Augmented Motion Diffuse

cs.CV

cs.AI

cs.GR

Released Date: December 5, 2024

Authors: Zhouyingcheng Liao¹, Mingyuan Zhang², Wenjia Wang¹, Lei Yang³, Taku Komura¹

Aff.: ¹The University of Hong Kong; ²Nanyang Technological University; ³SenseTime Research

Arxiv: http://arxiv.org/pdf/2412.04343v1

[Uncaptioned image]

Datasets	Methods	R Precision $\uparrow$			FID $\downarrow$	MM Dist $\downarrow$	Diversity $\rightarrow$	MultiModality $\uparrow$
Datasets	Methods	Top 1	Top 2	Top 3	FID $\downarrow$	MM Dist $\downarrow$	Diversity $\rightarrow$	MultiModality $\uparrow$
Human ML3D	MDM [29]	${0.455}^{\pm{.006}}$	${0.645}^{\pm{.007}}$	${0.749}^{\pm{.006}}$	${0.489}^{\pm{.047}}$	${3.330}^{\pm{.025}}$	$\mathbf{{9.920}}^{\pm{.083}}$	${2.290}^{\pm{.070}}$
	T2M-GPT [31]	${0.492}^{\pm{.003}}$	${0.679}^{\pm{.002}}$	${0.775}^{\pm{.002}}$	${0.141}^{\pm{.005}}$	${3.121}^{\pm{.009}}$	$\underline{{9.761}}^{\pm{.081}}$	${1.831}^{\pm{.048}}$
	ReMoDiffuse [33]	${0.510}^{\pm{.005}}$	${0.698}^{\pm{.006}}$	${0.795}^{\pm{.004}}$	$\underline{{0.103}}^{\pm{.004}}$	${2.974}^{\pm{.016}}$	${9.018}^{\pm{.075}}$	${1.795}^{\pm{.043}}$
	FineMoGen [34]	${0.504}^{\pm{.002}}$	${0.690}^{\pm{.002}}$	${0.784}^{\pm{.002}}$	${0.151}^{\pm{.008}}$	${2.998}^{\pm{.008}}$	${9.263}^{\pm{.094}}$	$\mathbf{{2.696}}^{\pm{.079}}$
	MoMask [9]	$\underline{{0.521}}^{\pm{.002}}$	$\underline{{0.713}}^{\pm{.002}}$	$\underline{{0.807}}^{\pm{.002}}$	$\mathbf{{0.045}}^{\pm{.002}}$	${2.958}^{\pm{.008}}$	-	${1.241}^{\pm{.040}}$
	MotionDiffuse [32]	${0.515}^{\pm{.003}}$	${0.708}^{\pm{.003}}$	${0.806}^{\pm{.002}}$	${0.141}^{\pm{.007}}$	$\underline{{2.919}}^{\pm{.008}}$	${9.485}^{\pm{.093}}$	$\underline{{2.669}}^{\pm{.087}}$
	Ours	$\mathbf{{0.524}}^{\pm{.002}}$	$\mathbf{{0.715}}^{\pm{.002}}$	$\mathbf{{0.811}}^{\pm{.001}}$	${0.111}^{\pm{.005}}$	$\mathbf{{2.879}}^{\pm{.006}}$	${9.527}^{\pm{.090}}$	${2.604}^{\pm{.084}}$