4月11日，“上海人工智能实验室学术论坛”将举办第13期“星启Talk”。本期邀请苏黎世联邦理工学院在读博士张四维，带来主题报告Interaction Capture and Synthesis for Humans in 3D Environments。“星启Talk”不定期邀请全球杰出青年学者，分享在人工智能相关领域的代表性工作。

“星启Talk”第13期

【活动详情】

主题：Interaction Capture and Synthesis for Humans in 3D Environments

嘉宾：张四维 苏黎世联邦理工学院在读博士

主持人：潘亮 上海人工智能实验室青年研究员

时间：北京时间4月11日15:00-16:00

参与方式：请扫码观看线上直播

【讲座简介】

Building the 3D digital world demands an in-depth understanding of human behaviors. Fundamental to this understanding is the accurate perception and realistic generation of 3D human pose, shape, and motion. Since human behaviors are inherently linked to the environment, it is crucial to model the interactions between humans and their surroundings. The tasks of perception and generation are interdependent: precise human perception yields high-quality data for the generation of high-fidelity human behaviors, while the generated data in turn provides rich annotations at a reduced cost, mitigating the dependence on expensive manual annotations for perception tasks. In this speech, the speaker will discuss her recent research on environment-aware human motion perception and generation. While existing research on human pose estimation has delivered notable results with RGB input, it remains insufficiently robust against the emerging challenges in this domain, such as extreme body occlusions/truncations, varied viewpoints (for example, egocentric perspectives), noisy inputs, and human-scene interactions. To overcome these challenges, researchersemploy probabilistic modeling to learn the distributions of body pose and motion, implementing flexible conditioning strategies for partial observations and the 3D environment. For the synthesis of realistic human behaviors, various interaction-based representations for human-scene and human-object interactions are introduced. Researchers demonstrate that the synthesized data substantially improves the performance of existing methods across several perception applications.

【本期嘉宾】

张四维 苏黎世联邦理工学院 在读博士

苏黎世联邦理工学院计算机视觉与学习团队（VLG）的博士研究生，师从Siyu Tang教授。于2020年在苏黎世联邦理工学院获得电气工程与信息技术硕士学位，2017年在清华大学自动化专业获得学士学位。研究重点为三维环境中人体姿态和运动的感知与生成以及第一人称视角下的人类理解。

【主持人】

潘亮 上海AI实验室 青年研究员

 上海人工智能实验室学术论坛

“上海人工智能实验室学术论坛”分设“星河Talk”与“星启Talk”两个主题系列活动，将分别邀请全球顶尖教授和杰出青年学者作为嘉宾，线上线下分享学术成果、探讨科技前沿。更多精彩内容，敬请期待。