时间:2020年10月31日上午10点
地点:创新港5号巨构新港报告厅
迈向可理解的计算机视觉
Issues Towards Comprehensible Computer Vision
In the past decades, computer vision has become the hottest area in artificial intelligence due to it reaches similar or even better results in some typical tasks, such as objects recognition, than human being. However, computer vision is still a far from the goal of automatic understanding scene. To understand scene means the machine should not only know what about an object’s categorization, but also why / how / … / about an object and also their relationship in real world.
In this talk, I will briefly review the history of computer vision, and discuss its tendencies. Some key issues are listed as open problems for next decades. Understandable / explainable will be a crucial feature for open world vision system. Meanwhile, mobility, non-uniform imaging, exploration are all key problems for understandable / explainable computer vision. I will share my points on these relevant problems in this talk. Finally, I will also report some of our preliminary works on these topics.
在过去的几十年里,计算机视觉由于在一些典型的任务中,如物体识别,比人类取得了相似甚至更好的结果,成为人工智能领域中最热门的领域。然而,计算机视觉离自动理解场景的目标还很遥远。理解场景意味着机器不仅要知道一个对象的分类是什么,还应该知道为什么/如何/…/关于一个对象以及它们在现实世界中的关系。
在这篇演讲中,我将简要回顾计算机视觉的历史,并讨论其发展趋势。一些关键问题被列为未来几十年的公开问题。可理解/可解释性将是开放世界视觉系统的一个重要特征。同时,机动性、非均匀成像、探索等都是可理解/可解释计算机视觉的关键问题。在这次演讲中,我将谈谈我对这些相关问题的看法。最后,我将汇报一些我们在这些主题上的基本工作。
报告人简介:
陈熙霖
中科院计算技术研究所
Institute of Computing Technology, Chinese Academy of Sciences
陈熙霖博士,ACM / CCF / IAPR / IEEE Fellow,中科院计算技术研究所研究员。其主要研究领域为计算机视觉、模式识别、多媒体技术以及多模式人机接口。曾主持国家自然科学基金重大、重点项目、973计划课题、863计划项目等的研究,曾任IEEE Trans. on Image Processing和IEEE Trans. on Multimedia的 AE,目前担任Journal of Visual Communication and Image Representation的Senior AE、计算机学报和模式识别与人工智能的副主编。担任过FG2013 / 2018、ChinaMM 2018 / 2019和PRCV 2019 / 2020大会主席,并多次担任CVPR和ICCV等的领域主席。陈熙霖博士在国内外重要刊物和会议上发表论文300多篇,先后获得国家自然科学二等奖一项,国家科技进步二等奖四项。