20191211-30 「见微知著」——细粒度视觉识别

报告时间：2019年12月11日（星期三）晚上20:00（北京时间）

主题：「见微知著」——细粒度视觉识别

报告主持人：魏秀参（旷视）

报告嘉宾：王雷（University of Wollongong）

报告题目：Learning Kernel-matrix-based Representation for Fine-grained Image Recognition

报告嘉宾：崔崟（Google）

报告题目：Measuring Dataset Granularity

Panel议题：

1. 细粒度视觉识别和通用图像识别到底有何不同？

2. 细粒度视觉识别领域目前的热点是什么？还有哪些有价值的研究方向？

3. 细粒度视觉识别目前的瓶颈是什么？未来的挑战有哪些？

4. 细粒度视觉识别有哪些重要实际应用？有哪些重要场景？

5. 除了细粒度视觉识别，还有哪些有价值的细粒度图像分析方向？

Panel嘉宾：

王雷（University of Wollongong）、崔崟（Google）、李培华（大连理工大学）、王旗龙（天津大学）

*欢迎大家在下方留言提出主题相关问题，主持人和panel嘉宾会从中选择若干热度高的问题加入panel议题！

报告嘉宾：王雷（University of Wollongong）

报告时间：2019年12月11日（星期三）晚上20:00（北京时间）

报告题目：Learning Kernel-matrix-based Representation for Fine-grained Image Recognition

报告人简介：

Lei Wang received his PhD degree from Nanyang Technological University, Singapore. He is now Associate Professor at School of Computing and Information Technology of University of Wollongong, Australia. His research interests include machine learning, pattern recognition, and computer vision. Lei Wang has published 150+ peer-reviewed papers, including those in highly regarded journals and conferences such as IEEE TPAMI, IJCV, CVPR, ICCV and ECCV, etc. He was awarded the Early Career Researcher Award by Australian Academy of Science and Australian Research Council. He served as the General Co-Chair of DICTA 2014, Program Co-Chair of VCIP2019, and Area Chair of ICIP2019 and is on the Technical Program Committees of 20+ international conferences and workshops. Lei Wang is a senior member of IEEE.

个人主页：

https://sites.google.com/view/lei-hs-wang

报告摘要：

Fine-grained image recognition calls for powerful feature representation to characterise the subtle difference across image classes. During the past several years, covariance matrix has been proposed as an effective feature representation for this purpose. This talk will report our recent work on learning kernel-matrix-based feature representation to achieve better fine-grained image recognition.

The first part of this talk provides a brief overview of covariance-based feature representation. The second part moves beyond covariance matrix to develop kernel matrix as a feature representation. It not only mitigates the high dimensionality and small sample issue in covariance estimation but also has the advantage of modelling nonlinear feature relationship. The last part of this talk embeds the kernel-matrix-based feature representation into deep neural networks to learn it in an end-to-end manner. Experimental study on multiple fine-grained image recognition benchmark datasets demonstrates the efficacy and advantage of the proposed methods.

参考文献：

【1】 R. Wang, H. Guo, L. S. Davis and Q. Dai, "Covariance discriminative learning: A natural and efficient approach to image set classification," 2012 IEEE Conference on Computer Vision and Pattern Recognition, Providence, RI, 2012, pp. 2496-2503.

【2】 S. Jayasumana, R. Hartley, M. Salzmann, H. Li and M. Harandi, "Kernel Methods on the Riemannian Manifold of Symmetric Positive Definite Matrices," 2013 IEEE Conference on Computer Vision and Pattern Recognition, Portland, OR, 2013, pp. 73-80.

【3】 T. Lin, A. RoyChowdhury and S. Maji, "Bilinear CNN Models for Fine-Grained Visual Recognition," 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, 2015, pp. 1449-1457.

【4】 P. Li, J. Xie, Q. Wang and W. Zuo, "Is Second-Order Information Helpful for Large-Scale Visual Recognition?," 2017 IEEE International Conference on Computer Vision (ICCV), Venice, 2017, pp. 2089-2097.

【5】 L. Wang, J. Zhang, L. Zhou, C. Tang and W. Li, "Beyond Covariance: Feature Representation with Nonlinear Kernel Matrices," 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, 2015, pp. 4570-4578.

【6】 M. Engin, L. Wang, L. Zhou, X. Liu “DeepKSPD: Learning Kernel-matrix-based SPD Representation for Fine-grained Image Recognition.” The European Conference on Computer Vision (ECCV), Munich, 2018, pp. 629—645.

【7】 Second- and Higher- order Representations in Computer Vision, tutorial at ICCV2019.

【8】 SPD Representations Methods Archive.

报告嘉宾：崔崟（Google）

报告时间：2019年12月11日（星期三）晚上20:30（北京时间）

报告题目：Measuring Dataset Granularity

报告人简介：

Yin Cui is a Research Scientist at Google. Before joining Google, he received Ph.D. from Cornell University, advised by Professor Serge Belongie. His research interests are computer vision and deep learning, with a focus on learning from fine-grained and long-tailed visual data. He has been co-organizing COCO Visual Recognition Workshops and Fine-Grained Visual Categorization Workshops at major computer vision conferences.

个人主页：

https://ycui.me/

报告摘要：

Despite the increasing visibility of fine-grained recognition in our field, "fine-grained" has thus far lacked a precise definition. In this talk, I will introduce our recent effort to pursue a framework for measuring dataset granularity building upon clustering theory. We argue that dataset granularity should depend not only on the data samples and their labels, but also on the distance function we choose. We propose an axiomatic framework to capture desired properties for a dataset granularity measure and provide examples of measures that satisfy these properties. We assess each measure via experiments on datasets with hierarchical labels of varying granularity. When measuring granularity in commonly used datasets with our granularity measure, we find that certain datasets that are widely considered fine-grained in fact contain subsets of considerable sizes with substantially higher granularity than datasets generally considered coarse-grained. We also investigated interplay between dataset granularity with a variety of factors and find that fine-grained datasets are more difficult to learn from, more difficult to transfer to, more difficult to perform few-shot learning with, and more vulnerable to adversarial attacks.

Panel嘉宾：李培华（大连理工大学）

嘉宾简介：

李培华，教授，博士生导师，现任职于大连理工大学信息与通信工程学院. 2003年获哈尔滨工业大学计算机应用技术专业博士学位，之后在法国INRIA从事博士后研究. 2011年入选教育部新世纪优秀人才支持计划, 获得过黑龙江省青年科技奖和全国优秀博士学位论文提名奖等荣誉. 所指导的团队获CVPR 2018大规模细粒度物种识别挑战赛冠军(1/59), 2015年获阿里巴巴大规模图像搜索竞赛第2名(2/843), 2008年获虹膜识别国际评测NICE第4名(4/97). 研究内容包括深度学习、计算机视觉和模式识别，主要研究兴趣为图像/视频识别、目标检测和语义分割. 在国际期刊如IIEE TPAMI/TIP/TCSVT及国际会议ICCV/CVPR/ECCV/NeurIPS发表几十篇论文. 作为项目负责人，主持包括国家自然科学基金、教育部科学技术重点项目和企业合作项目等十几项。

个人主页：

http://www.peihuali.org/

Panel嘉宾：王旗龙（天津大学）

嘉宾简介：

王旗龙：博士，天津大学智能与计算学部助理教授，主要研究方向是深度学习，概率分布建模和视频图像分析。目前发表学术论文30余篇，大多数发表在国际顶级会议CVPR/ICCV/ECCV/NIPS/IJCAI以及IEEE T-PAMI/IEEE T-IP/IEEE T-CSVT等国际权威期刊。曾获得2015年阿里巴巴大规模图像检索大赛第二名(2/853)、ICIP2015 Best 10% paper。CVPR, ICCV, ECCV, IJCAI, AAAI等会议以及IEEE T-IP, T-NNLS等期刊审稿人。入选2018年博士后创新人才计划，获得国家自然科学基金青年基金以及博士后基金等资助。

个人主页：

https://csqlwang.github.io/homepage/

主持人：魏秀参（旷视）

主持人简介：

魏秀参，博士，旷视南京研究院院长，南京大学学生创业导师，VALSE执行AC。主要研究领域为计算机视觉和机器学习，在相关领域顶级期刊如IEEE TPAMI、IEEE TIP、IEEE TNNLS、IEEE TKDE、Machine Learning Journal等及顶级会议如CVPR、ICCV、IJCAI、ICDM、ACCV等发表论文二十余篇，并带队获得iNaturalist、Apparent Personality Analysis等计算机视觉领域国际权威赛事共3项世界冠军。分别在重要国际会议PRICAI 2018和ICME 2019组织题为"Fine-Grained Image Analysis"的tutorial。著有《解析深度学习--卷积神经网络原理与视觉实践》一书。曾获CVPR 2017最佳审稿人、南京大学博士生校长特别奖学金等荣誉，担任ICCV、CVPR、ECCV、NIPS、IJCAI、AAAI等国际会议PC member。

19-30期VALSE在线学术报告参与方式：

长按或扫描下方二维码，关注“VALSE”微信公众号（valse_wechat），后台回复“30期”，获取直播地址。

特别鸣谢本次Webinar主要组织者：

主办AC：魏秀参（旷视）

责任AC：曹汛（南京大学）

VALSE Webinar改版说明：

自2019年1月起，VALSE Webinar改革活动形式，由过去每次一个讲者的方式改为两种可能的形式：

1）Webinar专题研讨：每次活动有一个研讨主题，先邀请两位主题相关的优秀讲者做专题报告（每人30分钟），随后邀请额外的2~3位嘉宾共同就研讨主题进行讨论（30分钟）。

2）Webinar特邀报告：每次活动邀请一位资深专家主讲，就其在自己熟悉领域的科研工作进行系统深入的介绍，报告时间50分钟，主持人与主讲人互动10分钟，自由问答10分钟。

活动参与方式：

1、VALSE Webinar活动依托在线直播平台进行，活动时讲者会上传PPT或共享屏幕，听众可以看到Slides，听到讲者的语音，并通过聊天功能与讲者交互；

2、为参加活动，请关注VALSE微信公众号：valse_wechat 或加入VALSE QQ群（目前A、B、C、D、E、F、G、H、I、J、K群已满，除讲者等嘉宾外，只能申请加入VALSE L群，群号：641069169）；

*注：申请加入VALSE QQ群时需验证姓名、单位和身份，缺一不可。入群后，请实名，姓名身份单位。身份：学校及科研单位人员T；企业研发I；博士D；硕士M。

3、在活动开始前5分钟左右，讲者会开启直播，听众点击直播链接即可参加活动，支持安装Windows系统的电脑、MAC电脑、手机等设备；

4、活动过程中，请不要说无关话语，以免影响活动正常进行；

5、活动过程中，如出现听不到或看不到视频等问题，建议退出再重新进入，一般都能解决问题；

6、建议务必在速度较快的网络上参加活动，优先采用有线网络连接；

7、VALSE微信公众号会在每周四发布下一周Webinar报告的通知及直播链接。

8、Webinar报告的PPT（经讲者允许后），会在VALSE官网每期报告通知的最下方更新[slides]。

9、Webinar报告的视频（经讲者允许后），会更新在VALSE爱奇艺空间，请在爱奇艺关注Valse Webinar进行观看。

王雷[slides]

崔崟[slides]