中国计算机学会计算机视觉专业组

走进高校系列报告会

CCF－CV Series Lectures

北京工业大学·北京（第8期）

2016年1月11日（星期一）13:30-18:10

北京工业大学人文楼312

报告会主题

计算机视觉前沿技术及应用

程 序

13:00 签到

13:30 报告会开始

特邀讲者：颜水成 博士，360首席科学家、人工智能研究院院长,新加坡国立大学副教授

演讲题目：Deep Learning for Face and Human Analytics

特邀讲者：王晓刚 博士,香港中文大学副教授

演讲题目：DeepID: Deep Learning for Face Recognition

特邀讲者：王涛 博士，中国计算机学会（CCF）理事，计算机视觉专委副主任，爱奇艺公司首席科学家

演讲题目： 计算机视觉和虚拟现实在视频应用中的挑战

特邀讲者：张勇东 博士，中国科学院计算技术研究所研究员

演讲题目：大规模网络多媒体内容分析与网络空间安全

特邀讲者：韩军伟 博士，西北工业大学教授

演讲题目：视觉显著性计算及应用

        执行主席：毋立芳 博士，北京工业大学教授

中国计算机学会计算机视觉专业组副秘书长、委员

参加人员：视觉领域专业人士、研究生、媒体、其他有兴趣者

报名方式：马玉琨，15836120020

Email：dip_ee@emails.bjut.edu.cn(请于1月8日前将参会回执回复至该邮箱，邮件主题请注明“CCF－CV北京工业大学报告会回执”)

参加方式：免费参加，敬请光临。

参会回执

     特邀讲者 颜水成

Dr. Yan Shuicheng is chief scientist of Qihoo/360, director of 360AI institute, and also the Dean’s Chair Associate Professor at NationalUniversity of Singapore. Dr. Yan’s research areas include machine learning,computer vision and multimedia, and he has authored/co-authored hundreds oftechnical papers over a wide range of research topics, with Google Scholarcitation over 22,000 times and H-index 61. He is ISI Highly-cited Researcher2014 and 2015, and IAPR Fellow 2014. He received the Best Paper Awards fromTMM’15 (honourable-mention), ACM MM’13 (both Best Paper and Best Student Paper),ACM MM’12 (Best Demo), PCM’11, ACM MM’10, ICME’10 and ICIMCS’09, the runner-upprize of ILSVRC’13, the winner prize of ILSVRC’14 detection task without extradata, the winner prizes of the classification task in PASCAL VOC 2010-2012, thewinner prize of the segmentation task in PASCAL VOC 2012, the honourablemention prize of the detection task in PASCAL VOC’10, 2010 TCSVT Best AssociateEditor (BAE) Award, 2010 Young Faculty Research Award, 2011 Singapore YoungScientist Award, and 2012 NUS Young Researcher Award.

报告摘要：This talk introduces the research on how challenge-aware deeplearning solutions boost performance on face and human body analytics. Forface, blur-aware face detection and occlusion-aware face alignment shall beintroduced. For human body, scale-aware human detection and context-aware humanparsing shall be introduced. Particularly, state-of-the-art performances areachieved for face alignment, human detection and human parsing tasks on popularbenchmark datasets.

    特邀讲者 王晓刚

Dr WangXiaogang received his Bachelor degree in Electrical Engineering and InformationScience from the Special Class of Gifted Young at the University of Science andTechnology of China in 2001, M. Phil. degree in Information Engineering fromthe Chinese University of Hong Kong in 2004, and PhD degree in Computer Sciencefrom Massachusetts Institute of Technology in 2009. He is an associateprofessor in the Department of Electronic Engineering at the Chinese Universityof Hong Kong since August 2009. He received the Outstanding Young Researcher inAutomatic Human Behaviour Analysis Award in 2011, Hong Kong RGC Early CareerAward in 2012, and Young Researcher Award of the Chinese University of HongKong. He is the associate editor of the Image and Visual Computing Journal. Hewas the area chair of ICCV 2011, ICCV 2015, ECCV 2014 and ACCV 2014. Hisresearch interests include computer vision, deep learning, crowd videosurveillance, object detection, and face recognition.

报告摘要：In this talk, I will present our works on deeplearning for face recognition. With a novel deep model and a moderate trainingset with 400,000 face images, 99.47% accuracy has been achieved on LFW, themost challenging and extensively studied face recognition dataset. Deeplearning provides a powerful tool to separate intra-personal and inter-personalvariations, whose distributions are complex and highly nonlinear, throughhierarchical feature transforms. It is essential to learn effective face representationsby using two supervisory signals simultaneously, i.e. the face identificationand verification signals. Some people understand the success of deep learningas using a complex model with many parameters to fit a dataset. To clarify suchmisunderstanding, we further investigate face recognition process in deep nets,what information is encoded in neurons, and how robust they are to datacorruptions. We discovered several interesting properties of deep nets,including sparseness, selectiveness and robustness.

In Multi-View Perception, a hybrid deep model is proposed tosimultaneously accomplish the tasks of face recognition, pose estimation, andface reconstruction. It employs deterministic and random neurons to encodeidentity and pose information respectively. Given a face image taken in anarbitrary view, it can untangle the identity and view features, and in themeanwhile the full spectrum of multi-view images of the same identity can bereconstructed. It is also capable to interpolate and predict images underviewpoints that are unobserved in the training data.

特邀讲者 王涛

王涛，中国计算机学会（CCF）理事，计算机视觉专委副主任，爱奇艺公司首席科学家。主要从事计算机视觉、模式识别、多媒体分析、虚拟现实、数据挖掘等相关技术的研究。在IJCV、ACM MMSJ、CVPR、ACM Multimedia等国际期刊和会议上发表论文六十余篇，申请二十多项专利，翻译《软件优化技术》著作一本。

报告摘要：互联网是一个强大的工具，更好的连接了人和服务，也催生了互联网+与智能视频的融合。未来，智能分析和虚拟现实技术面临很多挑战和机遇。互联网+智能视频快速发展的同时，人类对虚拟现实技术正呈现出前所未有的需求和创造力。通过video in闪植、video out随视购等智能视频分析技术可以创造全新的广告营销方式。虚拟现实全景技术通过技术创新和内容创意相结合，可以为视频行业带来更多机会，创造崭新的视觉体验。

特邀讲者张勇东

张勇东，中国科学院计算技术研究所研究员、博士生导师，中国科学院大学岗位教授，2015年度国家杰出青年科学基金获得者，IEEE高级会员，中国计算机学会多媒体专业委员会委员，担任多个知名国际期刊编委，主要从事网络多媒体内容分析与处理技术的研究，作为项目负责人，已承担国家973、863、国家自然科学基金、国家信息安全242计划等二十余项国家级科研课题，在IEEE T-IP、T-MM、T-CSVT等多媒体领域著名学术期刊和会议上发表论文100余篇，拥有授权发明专利40余项。相关研究成果已经在国家网络多媒体监管领域取得大规模应用，作为第一完成人，获得2014年度北京市科学技术奖一等奖和2012年度中国计算机学会科学技术奖。

报告摘要：随着网络多媒体数据爆发式增长，大量有害多媒体内容通过互联网无序传播，危及社会稳定与国家安全，严重影响网络多媒体产业健康发展，加强网络多媒体内容审核成为迫切需求，大规模网络多媒体内容分析技术将在其中发挥重要作用。此报告将重点介绍讲者近年来在面向网络内容安全的大规模多媒体内容分析技术上取得的研究进展，包括以内容为中心的分析技术和以用户为中心的分析技术。

    特邀讲者 韩军伟 

韩军伟，西北工业大学教授，博士生导师，自动化学院副院长，信息融合技术教育部重点实验室副主任。主要研究方向是多媒体信息处理和脑成像分析。在IEEE汇刊和领域顶级的国际会议如CVPR、ICCV、ACMMM、MICCAI等发表学术论文40余篇，获得国际会议ACM Multimedia 2010和MICCAI 2011最佳学生论文提名，担任IEEETrans. on Human-Machine Systems, Neurocomputing等六个国际期刊编委/客座编委。获得国家自然科学基金委优秀青年基金和欧盟玛丽居里国际人才引进基金，入选教育部新世纪优秀人才支持计划和陕西省青年科技新星计划。

报告摘要：模拟人类视觉注意机制，视觉显著性计算能够自动估计图像和视频中包含的重要内容，从而为诸多多媒体应用提供便利，是目前计算机视觉领域的一个研究热点。本报告首先将简单介绍视觉注意机制的工作原理和研究进展；其次详细介绍视觉显著性计算技术的基本原理、实现方法、关键技术、难点问题、最新进展，以及我们在这一研究方向上的一些创新工作；最后展示显著性分析在图像/视频检索、压缩、传输、摘要、分类、监控、人机交互等方向的一些应用。

会场路线图

扫描二维码，关注专委公众号