MoST: Motion Style Transformer between Diverse Action Contents

TL;DR

This paper introduces MoST, a novel motion style transformer that effectively disentangles style from content in motion data, enabling high-quality style transfer across diverse actions without heuristic post-processing.

Contribution

The paper presents a new architecture with part-attentive style modulation and Siamese encoders, along with a style disentanglement loss, to improve motion style transfer between different contents.

Findings

Outperforms existing motion style transfer methods

Achieves high-quality style transfer across diverse motion contents

Does not require heuristic post-processing

Abstract

While existing motion style transfer methods are effective between two motions with identical content, their performance significantly diminishes when transferring style between motions with different contents. This challenge lies in the lack of clear separation between content and style of a motion. To tackle this challenge, we propose a novel motion style transformer that effectively disentangles style from content and generates a plausible motion with transferred style from a source motion. Our distinctive approach to achieving the goal of disentanglement is twofold: (1) a new architecture for motion style transformer with `part-attentive style modulator across body parts' and `Siamese encoders that encode style and content features separately'; (2) style disentanglement loss. Our method outperforms existing methods and demonstrates exceptionally high quality, particularly in motion pairs with different contents, without the need for heuristic post-processing. Codes are available at https://github.com/Boeun-Kim/MoST.