您当前的位置:

【学术报告】悉尼科技大学郑良博士于12月25日来访实验室

发布时间:2017-12-25

报告题目:Pedestrian Retrieval: State of the Art and Future

时间:12月25日上午10:00~11:30

地点:计算所446会议室


报告摘要:

Person retrieval, also known as re-identification, is a task aiming to retrieve images of a queried identity from a large pedestrian database. Different from generic retrieval problems, person retrieval benefits from large-scale fully annotated training sets. In recent years, deep learning techniques have become a dominating force in this field, and the state of the art has been advanced in a quicker-than-ever speed. In this talk, after a brief introduction of the task, I will first describe the baselines, including the usage of the identification and triplet losses. I will provide a review on how the baseline performance evolves over the past year and point out some useful tricks in baseline construction. Second, I will describe some effective methods that considerably improve the baseline. These methods include pose normalization, part-informed feature learning, and GAN-based domain adaptation/unsupervised learning. Finally, I will comment on some promising future directions in person retrieval, including language/attribute-based query systems and end-to-end person re-identification.


报告人简介:

Dr Liang Zheng is a postdoc researcher in the University of Technology Sydney. Prior to joining UTS, he obtained his B.E and PhD degrees from Tsinghua University. He has published over 20 papers in highly selected venues such as TPAMI, IJCV, CVPR, ECCV, and ICCV. He has made early attempts in large-scale person re-identification by introducing three benchmarking systems, Market-1501, MARS, and PRW. His works are extensively cited by the community and the most highly cited paper receives 200+ citations within two years of publication. Dr Zheng received the Outstanding PhD Thesis from Chinese Association of Artificial Intelligence, and the Early Career R&D Award from D2D CRC, Australia. His research was featured by the MIT Technical Review and selected into the computer science courses in Stanford University and the University of Texas at Austin.



附件下载: