Learning Universal Representations Across Tasks and Domains

Talk by Wei-Hong LI

Oct 26, 2023 Thursday


A longstanding goal in computer vision research is to produce broad and general-purpose systems that work well on a broad range of vision problems and are capable of learning concepts only from few labelled samples. In contrast, existing models are limited to work only in specific tasks or domains (datasets), e.g., a semantic segmentation model for indoor images (Silberman et al., 2012). In addition, they are data inefficient and require large labelled dataset for each task or domain. While there has been works proposed for domain/task-agnostic representations by either loss balancing strategies or architecture design, it remains a challenging problem on optimizing such universal representation networks. This talk focuses on addressing the challenges of learning universal representations that generalize well over multiple tasks (e.g. segmentation, depth estimation) or various visual domains (e.g. image object classification, image action classification), and can be transferred and adopted to previously unseen tasks/domains in a data-efficient manner. In addition, the talk also shows that these representations can be learned from partial supervision and can be improved by the proposed 3D-aware regularization.


Oct 26, 2023 Thursday



RmE1-101, GZ Campus


628 334 1826 (PW: 234567)

Speaker Bio:

Dr. Wei-Hong Li

Research Associate (Postdoc), VICO Group led by Hakan Bilen

School of Informatics, University of Edinburgh

Wei-Hong Li is a research associate (postdoc) within the VICO Group led by Hakan Bilen in the School of Informatics at the University of Edinburgh. Prior to postdoc, he completed his PhD in the same group, supervised by Hakan Bilen and Timothy Hospedales. His research interests are in computer vision and machine learning, with a focus on multi-task/domain learning and learning visual models from limited human supervision. He was invited to give talks at VGG, Sun Yat-sen University et al. His MTPSL paper was listed in the CVPR 2022 Best Paper Nominees. Before Edinburgh, he did his master and bachelor at Sun Yat-sen University, working with Wei-Shi Zheng who thankfully introduced him to computer vision. During the master program, he was lucky to visit Queen Mary University of London to work with Shaogang Gong.