Speaker：Xiaoguang Han（The Chinese University of Hong Kong, Shenzhen）

Time：2022-12-29, 10:30

Location：Tencent Meeting ID：581-4735-3770（No Password）

Abstract:

Explicit geometric representations, such as meshes, have long been the dominant representation in 3D modeling. In recent years, with the rise of 3D deep learning, implicit representations have been found to be well suited for learning-based geometric modeling frameworks and have achieved very good performance in numerous application domains. Should we continue to follow explicit expressions or develop implicit expressions? It has become a lively topic of discussion among researchers. This presentation will introduce two case studies and show how to effectively combine implicit and explicit representations to exploit the advantages of each: 1) geometric contour line deformation guided by implicit representations for efficient image instance segmentation (CVPR 2022 - SharpContour); 2) 3D mesh deformation guided by implicit representations for real-time interactive 3D character modeling (UIST 2021 - SimpModeling).