Skip to content

← ArchivePaper2020

Motion Retargetting based on Dilated Convolutions and Skeleton-Specific Loss Functions

SangBin Kim, Inbum Park, Seongsu Kwon, JungHyun Han

CGFAcademic12 citesRetargeting

Unsupervised temporal dilated convolution network retargets motion across humanoids of different skeleton proportions while preserving high-frequency detail.

Abstract

Motion retargetting refers to the process of adapting the motion of a source character to a target. This paper presents a motion retargetting model based on temporal dilated convolutions. In an unsupervised manner, the model generates realistic motions for various humanoid characters. The retargetted motions not only preserve the high‐frequency detail of the input motions but also produce natural and stable trajectories despite the skeleton size differences between the source and target. Extensive experiments are made using a 3D character motion dataset and a motion capture dataset. Both qualitative and quantitative comparisons against prior methods demonstrate the effectiveness and robustness of our method.

How to read this

Category
Method: a deep motion retargeting network
Contributions
  • An unsupervised motion retargeting model built on temporal dilated convolutions
  • Retargeting across humanoids of differing skeleton proportions while preserving high-frequency motion detail
  • Natural, stable trajectories despite source-target skeleton size differences
Context
Sits in the deep-learning retargeting line alongside Aberman et al.'s Skeleton-Aware Networks, swapping graph-style architectures for temporal dilated convolutions over the motion sequence.Builds on: Skeleton-Aware Networks for Deep Motion Retargeting
Correctness
Evaluated qualitatively and quantitatively on a 3D character motion dataset and a motion capture dataset against prior methods; as with most learned retargeters, generalization is bounded by the training distribution of skeletons and motions, so out-of-distribution proportions warrant caution.
Clarity
Accessible at the concept level; a first pass conveys the dilated-convolution idea, a second pass is needed for the skeleton-specific loss formulation.
How to read it
First pass for the unsupervised setup and why dilated convolutions capture temporal detail; do a second pass on the skeleton-specific loss functions and the comparison tables if you care about how proportion differences are handled.

Builds on

Built upon by

Nothing yet.

Related work

Keywords

This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →