首页 > 科学研究 > 学术看板 > 正文

类别标签的语义信息在视觉识别模型中的应用

供稿:    责任编辑:安果    时间:2018-05-02    阅读:次

88125b大爆奖手机版 www.plumbing101.net 报告题目:类别标签的语义信息在视觉识别模型中的应用

The semantic information of class labels and its applications in visual recognition

报告时间:2018年5月3日10:00

报告地点:望江西五教101副楼

主讲人:刘凌峤博士

刘凌峤博士现为阿德莱德大学讲师(Lecturer)。他在澳大利亚国立大学取得博士学位。其主要研究兴趣为计算机视觉与机器学习。他在计算机视觉和机器学习顶级会议(如CVPR, ICCV, NIPS, ECCV)以及期刊(如TPAMI, IJCV)上发表论文共30余篇。他于2016年获得由澳大利亚研究理事会颁发的Discovery Early Career Researcher奖(澳大利亚优秀青年基金)。

Dr Lingqiao Liu is a Lecturer in University of Adelaide. He obtained his PhD from the Australian National University. His main research interests are machine learning and computer vision. He has published more than 30 papers on the top conference/journals of machine learning and computer vision, including CVPR, ICCV, NIPS, ECCV, TPAMI, IJCV. He has been awarded the prestigious DECRA fellowship in 2016.

个人主页:

https://sites.google.com/site/lingqiaoliu83/


报告摘要:

在一般计算机识别系统里。类别标签仅作为一种区别类与类的标记。而实际上,类别标签也包含了丰富的语义信息。这些语义信息可以被用来帮助识别系统取得更好的性能。本次报告将介绍应用该思路的两个识别系统案例。其中一个用类别标签来提取图像与标签吻合的视觉信息并用来帮助图像搜索。另一个利用类别标签的语义信息来解决视觉关系检测问题。

Abstract

In traditional visual recognition, class labels are merely used as tokens to distinguish different categories. However, in many visual recognition problems, the semantic information beneath the class label can be invaluable for improving the recognition performance. In this talk, I will introduce two of our recent works which leverage semantic information from the class label to guide visual recognition. Specifically, one uses semantic information to extract the visual patterns which are consistent with the class label and uses this scheme for image retrieval. The other uses label semantic information for solving the visual relationship detection problem.


欢迎各位光临!