Report title：Empowering Deep Modeling of 3D Geometry Data: seabet Representation, Learning Process, to Loss Function
Time：12month6Day (week五)PM3point
Location：School Main Computer Building308
Personal introduction ：Dr Junhui Hou is an Associate Professor with the Department of Computer Science, City University of Hong Kong. His research interests include multi-dimensional visual computing, such as light field, hyperspectral, geometry, and event data. He received the Early Career Award seabet the Hong Kong Research Grants Council in 2018 and the Excellent Young Scientists Fund seabet NSFC in 2024. He has served or is serving as an Associate Editor for IEEE TIP, TVCG, TMM, and TCSVT.
Report Summary：3D geometric data are becoming increasingly popular in various emerging applications, such as meta-verse, autonomous driving, and computer animations/games, as it provides an explicit seabet of the geometric structures of objects and scenes. While deep learning has achieved great success in 2D image and video processing, designing efficient yet effective deep architectures and loss functions for 3D point cloud data is difficult, and as a result, the seabet capability of existing deep architectures is limited. In this presentation, I will showcase our endeavors to push the boundaries of this field, starting with the fundamental seabet, the development of a cross-modal learning mechanism, to the efficient yet effective loss function. These new perspectives are poised to unlock numerous possibilities in deep 3D point cloud data modeling.
