报告时间：2023年7月14日下午14:00
报告题目：Learning with Limited Supervision for 3D Scene Understanding
报告地点：浙江大学紫金港校区蒙民伟楼402室
报告人：林国省（新加坡南洋理工）
 
报告内容介绍：
Weakly supervised point cloud segmentation aims to learn a segmentation model using only a few labelled points in 3D scenes, which greatly reduces the efforts for obtaining dense annotations on 3D point clouds. Existing methods remain challenging to accurately segment 3D point clouds since limited annotated data may lead to insufficient guidance on unlabeled data. In this talk, I will present our recent weakly supervised methods for semantic segmentation and instance segmentation based on consistency learning and label propagation. I will also talk about our recent method for weakly supervised learning on 3D point cloud sequences (4D point clouds). Self-supervised learning is another important learning scheme to learn from unlabelled data. I will introduce our self-supervised learning methods for scene flow estimation on 3D point clouds.
 
 
报告人简介：
Guosheng Lin is an Assistant Professor at the School of Computer Science and Engineering, Nanyang Technological University. He received his PhD from the University of Adelaide in 2014. His research interests generally lie in deep learning, visual understanding, and content generation. He has published over 100 research articles in top-tier research venues. He is named in the world's top 2% of scientists List.  He serves as an area chair or senior program committee member for CVPR, ACCV, IJCAI and AAAI.  He is an Associate Editor for TCSVT.