Siyu Zhu
Research Fellow and Full Professor, Institute of Artificial Intelligence Innovation and Industry, Fudan University
Siyu Zhu is a researcher, long-time full professor and doctoral supervisor at the Institute of Artificial Intelligence Innovation and Industry, Fudan University, with research interests in video and 3D generative modeling. He has published more than 60 papers in international conferences and journals such as CVPR, ICCV, ECCV, and PAMI. During his Ph.D., he co-founded Altizure, a 3D vision company, which was later acquired by Apple Inc. and served as the Director of AliCloud Artificial Intelligence Lab from 2017 to 2023. He has served as ICCV/AAAI domain chair/program committee member, and won the Distinguished Engineer Award from the Chinese Computer Society.
Topic
Digital human generation modeling: frontiers and applications
In recent years, significant progress has been made in diffusion and autoregressive based visual generation models. However, existing visual generation results still have obvious limitations in the representation of real-world physical attributes such as appearance, geometry and motion. In addition, existing methods usually rely on predefined control conditions and lack effective interaction mechanisms with the external environment. Therefore, this talk will explore the use of 3D models as the basis of physical representations for portrait video generation to better portray real physical properties and provide end-to-end interaction capabilities. By combining 3D reconstruction with visually generated models, it is expected to enhance the flexibility and controllability of gesture, motion, and environment interaction while preserving the fine visual quality of portraits, exploring new research and application space for portrait video generation.
