Multimedia-related research

ITRI 2D Talking Head Generation

Goal: Create controllable talking head given few-shot target person data.

Challenges:

• Head-lip motion match with audio
• Achieve realistic results
• Unseen camera pose problem.

3D Talking Avatar

Goal: Create a fully controllable 3D photorealistic avatar.

Challenges: OOD-NVS problem, few input views, controllable by mesh

Robot 3D vision

3D Scene Relighting

Physical / World model

We focuses on developing physical and world models that empower intelligent agents to understand, predict, and interact with their environment. By combining perception, dynamics, and reasoning, we aim to build systems that can simulate real-world scenarios, enabling robust decision-making and generalization in complex tasks.