Style translation filter to change attribute of motion

Akihiko Yamaguchi, Shiori Sato, Kentaro Takemura, Jun Takamatsu, Tsukasa Ogasawara

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we propose a style translation filter that changes the attribute (style) of the motion coming from the actors' ages, genders, and so on. Using this filter, we can diversify the motions. Specifically, this filter is modeled by the Gaussian process regression that estimates the difference of pose (joint angles) between a neutral motion and the motion of a target attribute. In learning this filter, a key technique is to find pairs of corresponding posed from the sample motions. We solve this problem by employing the Multifactor Gaussian Process Model (MGPM) proposed by Wang et al. [1]. In the experiments, we constructed multiple style translation filters from several attributes of walking motions, such as genders, ages, and emotions. The obtained filters were applied to some kinds of testing motions, such as walking, jumping, kicking, and dancing. The acquired motions were verified by a questionnaire study; the most of their attributes were changed to the filters' target attributes.

Original languageEnglish
Title of host publication2012 12th IEEE-RAS International Conference on Humanoid Robots, Humanoids 2012
Pages660-665
Number of pages6
DOIs
Publication statusPublished - 2012 Dec 1
Externally publishedYes
Event2012 12th IEEE-RAS International Conference on Humanoid Robots, Humanoids 2012 - Osaka, Japan
Duration: 2012 Nov 292012 Dec 1

Publication series

NameIEEE-RAS International Conference on Humanoid Robots
ISSN (Print)2164-0572
ISSN (Electronic)2164-0580

Other

Other2012 12th IEEE-RAS International Conference on Humanoid Robots, Humanoids 2012
CountryJapan
CityOsaka
Period12/11/2912/12/1

ASJC Scopus subject areas

  • Artificial Intelligence
  • Computer Vision and Pattern Recognition
  • Hardware and Architecture
  • Human-Computer Interaction
  • Electrical and Electronic Engineering

Fingerprint Dive into the research topics of 'Style translation filter to change attribute of motion'. Together they form a unique fingerprint.

Cite this