学 术

分享到微信 ×
打开微信“扫一扫”
即可将网页分享至朋友圈
学术沙龙:Deep Learning Person Re-Identification
文:教师发展中心 来源:党委教师工作部、人力资源部(教师发展中心) 时间:2017-09-18 4065

  本次学术沙龙活动教师发展中心特别邀请来自英国伦敦玛丽女王大学的朱霞天博士,与我校师生分享他用深度学习研究行人再验证的心得。具体安排如下,欢迎感兴趣的师生参加:

  一、主 题:Deep Learning Person Re-Identification

  二、主讲人:朱霞天 博士 (英国伦敦玛丽女王大学) 

  三、时 间:2017年9月20日(周三)10:00

  四、地 点:清水河校区宾诺咖啡

  五、主持人:计算机学院 叶茂 教授

  六、交流内容:

  Existing person re-identification (re-id) methods rely mostly on either localised or global feature representation alone in single scale. This ignores the joint mutual complementary effects between global and local appearance, lacks the potentially useful explicit information of other different scales, and loses the chance of mining the implicit correlated complementary advantages across scales. In this talk, we will introduce two newly developed person re-id approaches to addressing these limitations.

  In this first work, we show the advantages of jointly learning local and global features in a Convolutional Neural Network (CNN) by aiming to discover correlated local and global features in different context. Specifically, we formulate a method for joint learning of local and global feature selection losses designed to optimise person re-id when using only generic matching metrics such as the L2 distance. We design a novel CNN architecture for Jointly Learning Multi-Loss (JLML).

  In the second work, we demonstrate the benefits of learning multi-scale person appearance features using CNN by aiming to jointly learn discriminative scale-specific features and maximise multi-scale feature fusion selections in image pyramid inputs. Specifically, we formulate a novel Deep Pyramid Feature Learning (DPFL) CNN architecture for multi-scale appearance feature fusion optimised simultaneously by concurrent per-scale re-id losses and interactive cross-scale consensus regularisation in a closed-loop design.

  Extensive comparative evaluations demonstrate the advantages of the JLML and DPFL for person re-id over a wide range of state-of-the-art re-id methods on a number of benchmarking datasets.

  七、主讲人简介

  Xiatian Zhu is a computer vision researcher in the Intelligence Science and System Lab, School of Data and Computer Science, Sun Yat-sen University. He received the Ph.D. degree (2015) in computer vision from Queen Mary University of London (QMUL), the M.Eng. (2009) and B.Eng. (2006) degrees from University of Electronic Science and Technology of China. He won The Sullivan Doctoral Thesis Prize (2016), an annual award representing the best doctoral thesis submitted to a UK University in the field of computer or natural vision. Dr. Zhu has published a number of international top-tier journal and conference papers including IEEE Transactions Patten Analysis and Machine Intelligence (TPAMI), International Journal of Computer Vision (IJCV), Artificial Intelligence (AI), IEEE Transactions Neural Network and Learning System (TNNLS), IEEE Transactions Image Processing (TIP), Patten Recognition (PR), CVPR, ICCV, ECCV, IJCAI, AAAI, ICDM, WACV, BMVC, ICIP. He has also won many competitive research supporting grants including 2014 ECCV Student Travel Grant, 2013 IEEE PAMI-TC ICCV Student Travel Grant, 2013 BMVA International Conference Bursary, 2013 QMUL Post-graduate Research Fund.

  八、主办单位:人力资源部教师发展中心

    承办单位:计算机科学与工程学院、机器人研究中心

 

                     人力资源部教师发展中心

                       2017年9月18日


编辑:罗莎  / 审核:罗莎  / 发布:林坤

"