Multimodal 2D/3D/4D Avatar Generation and Editing

PhD Thesis Proposal Defence


Title: "Multimodal 2D/3D/4D Avatar Generation and Editing"

by

Mr. Hongyu LIU


Abstract:

The creation and animation of high-fidelity digital avatars represent a 
critical frontier in modern Computer Vision and Computer Graphics. As the 
demand for immersive experiences in Virtual Reality (VR), digital 
communication, and the Metaverse accelerates, the ability to synthesize 
controllable and photorealistic human representations has become a central 
objective of AI-Generated Content (AIGC). This research addresses the 
multifaceted challenges inherent in avatar synthesis, specifically focusing 
on the complex interplay between geometric accuracy, appearance fidelity, and 
motion controllability across diverse modalities. Driven by recent 
breakthroughs in deep learning, particularly in neural rendering and 
generative diffusion models, the field has witnessed a paradigm shift from 
traditional reconstruction pipelines to data-driven generative approaches. 
This thesis presents a comprehensive survey and taxonomy of modern techniques 
for 2D, 3D, and 4D avatar generation and editing. We systematically 
categorize existing literature based on input modalities and underlying 
neural architectures, critically evaluating their performance in terms of 
rendering quality and temporal coherence. Furthermore, this work identifies 
key bottlenecks in current 4D dynamic modeling and proposes potential 
directions for future research, aiming to establish a robust theoretical 
foundation for the next generation of virtual humans.


Date:                   Wednesday, 10 December 2025

Time:                   4:00pm - 6:00pm

Venue:                  Room 2129C
                        Lift 19

Committee Members:      Dr. Qifeng Chen (Supervisor)
                        Prof. Pedro Sander (Chairperson)
                        Dr. Long Chen