新闻公告
学术报告
最新通知
文件规定
就业信息
首页 >> 新闻公告 >> 学术报告 >> 12月14日(周四)上午10:40澳大利亚阿德莱德大学沈春华教授学术报告--Deep Learning for Dense Per-Pixel Prediction and Vision-to-Language Problems
12月14日(周四)上午10:40澳大利亚阿德莱德大学沈春华教授学术报告--Deep Learning for Dense Per-Pixel Prediction and Vision-to-Language Problems
2017-12-12 13:00

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

报告题目:Deep Learning for Dense Per-Pixel Prediction and Vision-to-Language Problems

人:沈春华 教授

    位:澳大利亚 阿德莱德大学

    间:20171214日周四上午10:40

    点:中心楼二楼教育部重点实验室会议室

邀请人/主持人:孙长银 教授

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

                                                    

欢迎各位老师和研究生参加!

 

报告摘要:

Dense per-pixel prediction provides an estimate for each pixel given an image, offering much richer information than conventional sparse prediction models. Thus the Computer Vision community have been increasingly shifting the research focus to per-pixel prediction. In the first part of my talk, I will introduce my recent work on deep structured methods for per-pixel prediction that combine deep learning and graphical models such as conditional random fields. I show how to improve depth estimation from single images and semantic segmentation with the use of contextual information in the context of deep structured learning.

 

In deep learning, the trend towards increasingly deep neural networks has been driven by a general observation that increasing depth increases the performance of a neural network. Recently, however, evidence has been amassing that simply increasing depth may not be the best way to increase performance, particularly given other limitations.  As a result, a new, shallower, architecture of residual networks is proposed, which significantly outperforms much deeper models on classification as well as semantic segmentation by a large margin.

 

Recent advances in computer vision and natural language processing (NLP) have led to new interesting applications. Two popular ones are automatically generating natural captions for images/video and answering questions relevant to a given image (i.e., visual question answering or VQA).  In the second part of my talk, I will describe several recent work from my group that take advantage of state-of-the-art computer vision and NLP techniques to produce promising results on both tasks of image captioning and VQA.

 

报告人简介:

沈春华博士现任澳大利亚阿德莱德大学(澳大利亚8所研究型大学之一)计算机科学学院终身正教授。曾在南京大学(强化部本科及电子系硕士),澳大利亚国立大学(硕士)学习,并在阿德莱德大学获得计算机视觉方向的博士学位。2011加盟阿德莱德大学之前,他在澳大利亚国家信息通讯技术研究院堪培拉实验室的计算机视觉组工作近6年,先后担任研究员、高级研究员以及终生Lab Staff。他同时担任澳大利亚机器人视觉研究中心(Australian Centre for Robotic Vision, http://roboticvision.org/)项目负责人(机器人视觉中的机器学习)和Adelaide分部的执行主任,是该中心13名首席研究员(Chief Investigator)之一。

 

20181月起,他同时担任Australian Cyber Security CRC 的项目负责人,该项目得到澳大利亚联邦政府75000万澳币的资助。2012年到2016年沈春华教授被澳大利亚科研局(Australian Research Council)授予ARC Future Fellowship

 

沈春华教授在阿德莱德大学的团队目前主要从事统计机器学习以及计算机视觉领域的研究工作。他在重要国际学术期刊和会议发表论文193篇。 其中在计算机视觉、机器学习的顶级会议 NIPS, ICML, CVPR, ICCV, ECCV发表论文62篇。他担任IEEE Transactions on Neural Networks and Learning Systems 副主编。多次担任重要国际学术会议(ICCV, CVPR, ECCV)程序委员。

学术报告   more
关于本站 | 联系我们 | 登录 ©2010 东南大学自动化学院