AuraTracer智迹闻
中文

EVENT DOSSIER

STyMo: Fast and Controllable Few-Shot Motion Style Transfer

2026-09-07 12:00 Models 🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
1mentions
SummaryAI generated

STyMo is a small-sample motion style transfer method that can be trained in just a few seconds, with training taking only one to two minutes. This method breaks the style into static components that capture time-invariant poses and dynamic time components that capture inter-frame movements, allowing for adjustment of pose intensity, temporal exaggeration, and body region styles during operation. The system introduces an interpretable mechanism to enable iterative creative workflows and adds a plasticity gate to automatically prevent artifacts caused by non-uniform motion. The study demonstrates diverse style effects ranging from subtle emotional changes to exaggerated character prototypes, and publishes the processed paired datasets to facilitate further research.

Related eventsRELATED EVENTS
Key entitiesKEY ENTITIES
STyMo

SignalsSIGNALS

Keyword heat
  • STyMo1

All reports (1)SOURCES

A arXiv cs.CV en 2026-09-07 12:00

STyMo: Fast and Controllable Few-Shot Motion Style Transfer

STyMo 提出一种仅需数秒配对数据即可学习运动风格、训练耗时仅一至两分钟的少样本方法。该方法将风格分解为捕捉时间不变姿态的静态分量和捕捉帧间动态的时间分量,支持运行时调整姿态强度、时间夸张度及身体区域风格。此外,系统引入可解释性机制以允许迭代创作工作流,并增设可塑性门控自动防止非分布运动产生的伪影。研究展示了从细微情感变化到夸张角色原型的多样化风格效果,并发布了处理后的配对数据集以促进后续研究。