EVENT

Event News

Talk on Digital Humans by Dr. Shanxin Yuan from QMUL

We are pleased to inform you about the upcoming seminar by Dr. Shanxin Yuan titled:"Motion Intelligence for Digital Humans: Rendering, Retargeting, and Generation" Everyone interested is cordially invited to attend!

Title:

Motion Intelligence for Digital Humans: Rendering, Retargeting, and Generation

Speaker's

Dr. Shanxin Yuan (Lecturer in Queen Mary University of London)

Abstract:

Creating expressive, animatable digital humans has long been a grand challenge in computer vision, requiring systems that can understand the full complexity of human appearance, structure, and motion. Recent advances in neural rendering and generative modelling are rapidly closing this gap, making it possible to synthesise photorealistic humans that move, gesture, and emote with striking fidelity.

This talk presents our team's recent research toward this vision, spanning four interconnected fronts. First, I will introduce HandSCS (ECCV 2026), our Gaussian splatting-based frameworks for high-fidelity hand rendering and animation, which leverage structural coordinate spaces to overcome the longstanding challenge of articulated hand synthesis. Next, I will present STaR (ICCV 2025), a seamless spatial-temporal motion retargeting system that transfers full-body motion across characters while enforcing physical penetration and consistency constraints. I will then discuss DanceChat, which uses large language models to bridge music understanding and dance generation, enabling expressive, semantically-controlled human motion. Finally, I will introduce our latest work on self-learning expression deformations for Gaussian avatars, which enables data-efficient modelling of fine-grained facial dynamics without requiring dense supervision.

Together, these works point toward a unified pipeline for generating, controlling, and animating digital humans, from fingertips to full-body movement to facial expression. I will close by discussing the remaining open challenges: achieving long-term temporal consistency, handling the extreme diversity of human appearance, and grounding generative models in the physical constraints of the real world.

Speaker's Biography:

Shanxin Yuan is a Lecturer in Digital Environment at School of Electronic Engineering and Computer Science, Queen Mary University of London, where he leads the Motion AI Lab. He holds a PhD from Imperial College London, where his research focused on hand pose estimation. His research interests are machine learning and computer vision, particularly 3D digital humans and motion intelligence. He serves as an Associate Editor for The Visual Computer journal. He regularly reviews major computer vision conferences (CVPR, ICCV, ECCV, and NeurIPS) and related journals (TPAMI, IJCV and TIP).
For more of Dr. Yuan's information, please check his homepage:
https://shanxinyuan.github.io/