Personal Research

Inspired by image segmentation methods, we propose a joint-temporal action segmentation method that performs multi-action recognition at each human body joint. Our proposed learning objectives incorporate the class distribution of each point to address the continuous label problem and consider that the inter-dependency between joints is crucial..

This study demonstrates a straightforward method obtaining statistical style differences (Δ𝑆𝑡𝑦𝑙𝑒) which in turn effectively replace original style from a motion content to another target style. The strength of transferred style is adjusted using a predefined parameter of our model.