I am a first-year Ph.D. student in Electrical Engineering at the City University of Hong Kong (CityU), supervised by Professor Haoliang Li. Before CityU, I received my master's degree (2021-2024) at the Institute of Computing Technology, Chinese Academy of Sciences (ICT, CAS) supervised by Professor Shihong Xia, researching Face Forgery Detection and Talking Face Generation.
My research interests include:
Applications: AI for Digital Human, AI for Security
I entered the City University of Hong Kong as a Ph.D. student.
2024/05
Completed my master's theis defense.
2024/04
Invited to serve as a Reviewer for ACM MM 2024.
2024/03
Invited to serve as a PC member for WWW 2024.
2023/07
One first-authored paper got accepted by MTAP.
2023/06
One co-authored paper got accepted.
Publications
MyPortrait: Morphable Prior-Guided Personalized Portrait Generation Bo Ding, Zhenfeng Fan, Shuang Yang, and Shihong Xia
Preprint TL;DR: We present Myportrait, a method for learning personalized neural portraits. By incorporating personalized prior from a monocular video and morphable prior from 3DMM, our framework can generate realistic portraits with personalized details.
Mining Collaborative Spatio-Temporal Clues for Face Forgery Detection Bo Ding, Zhenfeng Fan, Zejun Zhao, and Shihong Xia
Multimedia Tools and Applications (JCR Q2, CCF-C), 2023 Paper /
Chinese Blog TL;DR: We propose a multi-branch spatio-temporal difference network for face forgery detection by capturing complementary low-level spatio-temporal features in videos, which can enhance the generalization ability of the model.
Deepfake Detection Based on Incremental Learning
Zejun Zhao, Zhenfeng Fan, Bo Ding, and Shihong Xia
Frontiers of Data and Computing (In Chinese), 2023 Paper TL;DR: We design a face forgery detection system based on incremental learning, which reduces the training cost of the model when introducing new forgery samples.
GT-GAN: A General Transductive Zero-Shot Learning Method Based on GAN
Junhao Dong, Bo Xiao, Bo Ding, and Haoyv Wang
IEEE Access (JCR Q2), 2020 Paper TL;DR: We propose a general transductive ZSL method based on GANs, called GT-GAN, which introduces unlabeled unseen samples for training, to improve the performance of the generator.