报告题目：视觉模型设计及其应用

报告时间：12月4日下午14：00-17：00

报告地点：D533学术报告厅

报告内容摘要：

互联网和物联网时代催生了海量视频大数据，从这些海量视频数据中有效提取知识迫切需要各种人工智能的技术和手段。因此，如何进行人工智能驱动的视觉计算已经成为当今知识经济时代亟待解决的核心技术问题。本报告主要围绕数据驱动的人工智能学习方法，进行大规模图像/视频数据的视觉特征学习，从目标视觉感知特性、视觉特征表达、深度学习器构建机制、高层语义理解等多维度视角进行了深入剖析，并引入了大规模视觉特征学习所涉及的主要研究问题和技术方法。然后系统地回顾了视觉特征表达和学习领域的不同发展阶段，介绍了近年来我们利用视觉特征学习进行视觉语义分析和理解所做的一系列代表性的研究工作及其实际应用。报告的最后将和大家一起探讨一下涉及视觉特征学习所面临的一些开放性问题和难题。  

个人简介：

李玺，男，博士，教授，博士生导师。浙江省杰出青年科学基金获得者和浙江省151人才培养工程第二层次，获聘中国信息与电子科技发展战略研究中心专家委员会的特聘专家。同时担任中国图象图形学会理事、中国图象图形学会视觉大数据专委会常务委员、浙江省计算机协会计算机视觉专委会和多媒体专委会的副主任委员。另外，担任多个计算机视觉和模式识别方面的国际权威刊物和国际顶级会议的审稿人和程序委员(如IEEE TPAMI、IJCV、IEEE TIP、IEEE TNNLS、IEEE TKDE、CVPR 2019、ICML 2019、ICML 2018、ECCV 2018、NIPS 2017、CVPR 2017、2018、IJCAI 2017、CVPR 2016、ICCV 2015等)。在国内外知名学术会议（如RACV 2016、ICSW 2017、ICDS 2017、IEEE FMT 2018等）上做大会特邀报告。李玺教授的研究方向集中在计算机视觉和机器学习，其在国际权威期刊和国际顶级学术会议文章130篇，Google Scholar引用近3100多次，拥有ESI高被引论文，并担任，担任国际计算机视觉领域顶级会议ICCV 2019的领域主席（Area Chair）, 担任神经网络和学习系统领域顶级期刊《IEEE Transactions on Neural Networks and Learning Systems》的Associate Editor（2019.1—），同时担任神经计算领域知名国际刊物Neurocomputing和Neural Processing Letters的Associate Editor，同时担任国际模式识别大会ICPR 2018的Computer Vision Track的Area Chair。李玺教授获得两项最佳国际会议论文奖(包括ACCV 2010和DICTA 2012)，以及一项ICIP 2015 Top 10% 会议论文奖，一项ACML 2017最佳学生论文奖。另外分别获得两项中国北京市自然科学技术奖(包括一等奖和二等奖)，以及一项中国专利优秀奖。

Biologically-inspired image quality modeling

Photo quality evaluation is a challenging task in multimedia and computer vision fields. Conventional approaches suffer from the following three drawbacks: 1) the deemphasized role of semantic content that is many times more important than low-level visual features in photo aesthetics; 2) the difficulty to optimally fuse low-level and high-level visual cues in photo aesthetics evaluation; and 3) the absence of a sequential viewing path in the existing models, as humans perceive visually salient regions sequentially when viewing a photo.

To address these challenges, we propose a new biologically-inspired descriptor that mimics the way humans sequentially perceive visually/semantically salient regions in a photo. In particular, a weakly supervised learning paradigm is developed to project the local descriptors into a low-dimensional semantic space. Thereafter, each local descriptor can be described by multiple types of visual features, both at low-level and in high- level. Since humans usually perceive only a few salient regions in a photo, a sparsity-constrained ranking algorithm is proposed that seamlessly integrates both the low-level and the high-level visual cues. The top-ranked local descriptors are the discerned visually/semantically prominent ones in a photo. They are sequentially linked into a path that simulates the process of humans actively viewing (we call them actively viewing paths (AVPs)). Afterward, a deep aggregation network is utilized to deeply encode the AVPs. Finally, we learn a probabilistic measure based on such deep AVPs from the training photos that are marked as high quality by multiple users. Experimental results show that: 1) the AVPs are 87.65% consistent with real human gaze shifting paths, as verified by the eye-tracking data; and 2) our photo quality measure outperforms many of its competitors.

Short bio: Luming Zhang received his Ph.D. degree in computer science from Zhejiang University, China. Currently he is a research professor at Zhejiang University. He was professor at Hefei University of Technology and was a senior research fellow at School of Computing, National University of Singapore. His research interests include multimedia analysis, image enhancement, and pattern recognition. He has authored/co-authored more than 100 scientific articles at top venues including IEEE T-IP, T-MM, T-CYB, CVPR, ACM MM, IJCAI, AAAI and KDD. He was the program committee member/organizer of many international conferences, such as MMM, PCM, ACM Multimedia. He served as a Guest editor of 10+ international journals, such as IEEE Trans. Big data, Neurocomputing, Signal Processing, Multimedia tools and applications, Multimedia systems, and JVCI. He is the associate editor of Neurocomputing and KSII Transactions on Internet and Information Systems. As the first author, he received the best paper award at PCM 2015. He is an adjunct investigator at NUSRI.