报告会时间：5.7（周五）2:00pm
报告会地点：1教109党员之家
报告主题：周晓巍教授：Human Motion Capture from RGB Videos
王高昂教授：Towards the Challenges in Multi-Object Tracking
欢迎广大师生参会学习交流！
Human Motion Capture from RGB Videos
Human motion capture (MoCap) is the basis of many human-centered applications such as human-computer interaction, AR/VR, sports analysis, healthcare, etc. Existing commercial MoCap systems are often expensive and only usable in well-controlled environments. How to make MoCap systems more lightweight and widely applicable is a long-standing problem in computer vision and graphics communities. In this talk, I will introduce our recent efforts towards this goal, which aim to recover 3D human motion with only RGB videos as input. Specifically, I will discuss how to estimate 3D human poses from multi-view or single-view images, recover high-quality human motion from unsynchronized internet videos, and synthesize free-view videos (bullet time) of dynamic humans from sparse multi-view videos.
周晓巍，浙江大学“百人计划”研究员、博士生导师。2008年本科毕业于浙江大学，2013年博士毕业于香港科技大，2014至2017年在宾夕法尼亚大学 GRASP 机器人实验室从事博士后研究。2017年入选国家级青年项目并加入浙江大学。研究方向主要为计算机视觉及其在混合现实、机器人等领域的应用，在3D目标检测、姿态估计、运动捕捉、图像匹配等方面取得了一系列成果，相关工作曾多次获得CVPR及ICCV等顶级会议口头报告，曾获得CVPR18 3DHUMANS Workshop Best Poster Award、CVPR19 Best Paper Finalists、“陆增镛CAD&CG高科技奖”一等奖。担任CVPR21和ICCV21领域主席。更多信息请见个人主页：xzhou.me
Towards the Challenges in Multi-Object Tracking
Multi-object tracking (MOT) has drawn great attention in recent years. This technique is critically needed in many tasks, such as traffic flow analysis, human behavior prediction, autonomous driving assistance, etc. MOT has achieved great improvement over the past few years. However, some challenges remain, such as sensitiveness to occlusion, instability for multi-modality fusion, and non-robustness to cross-view perception. In this talk, I will present our recent works towards the challenges in MOT. Specifically, I will introduce how to exploit motion patterns in long-term tracking, how to deal with association errors, and how to combine geometry constraints in the cross-view setting.
王高昂，浙江大学国际联合学院研究员，UIUC兼职助理教授，博士生导师。2013年本科毕业于复旦大学，2015年硕士毕业于威斯康星麦迪逊分校，2019年博士毕业于华盛顿大学电子与计算机工程，之后工作于旷视北美研究院和Wyze Labs，2020年9月加入浙江大学。主要研究方向有计算机视觉，机器学习，图像和视频处理，具体还包括多目标跟踪，姿态估计，主动学习等。在IJCAI，CVAUI担任程序委员会成员，在许多国际著名期刊及会议担任审稿人角色。自2017年起，连续三年参加了英伟达（Nvidia）组织的智慧城市挑战赛，总计在5个赛道中取得4次第一1次第二的名次。2021年依托国际会议ICMR与华盛顿大学联合组织举办雷达目标检测挑战赛ROD2021。2020年荣获“创新嘉兴精英引领计划”领军人才项目。更多信息见个人主页：https://person.zju.edu.cn/gaoangwang
