报告时间：2023年12 月15 日 星期五 上午09：30—11：00
会议地点：电航楼219
报告摘要：
Point clouds are often sparse and incomplete in real-world scenarios. The prevailing methods for point cloud completion typically rely on encoding the partial points and then decoding complete points from a global feature vector, which might lose the existing patterns and elaborate structures. Furthermore, the complexity of different incomplete parts presents challenges for capturing nonlocal features and leads to uneven distribution. To address these issues, we propose WalkFormer, a novel approach to predict complete point clouds through a partial deformation process. Our method first samples locally dominant points based on the feature similarity. Since these points always maintain representative information of surrounding structures, they are suitable to be selected as the starting points for multiple guided walks. Moreover, we design a topology-aware Route Transformer module to exploit and aggregate the walk information in multi-step for predicting structure deformation. These guided walks facilitate the learning of long-range dependencies and encourage the points to form the missing part. This talk gives the introduction about this innovative point cloud completion framework.
报告人简介：
Yushi Li received the Ph.D. from The Hong Kong Polytechnic University, in 2021. He is currently an Assistant Professor at the Department of Intelligent Science, Xi'an Jiaotong-Liverpool University. His main research interests include computer vision, computer graphics, graph learning, and multimodal-based learning. He has published around 20 papers in several prestigious journals and conferences such IEEE TIP, IEEE TVCG, WACV, and ICONIP. He serves as the reviewer of TVCG, CAD/CG and WWW. He also serves as the PC member of WWW 2024 (Industry track) and co-chair of the ICVR 2023 special session.
信息科学技术学院
2023年12月12日
