好文作者面授招-20150318

报告嘉宾1张姗姗(德国马克斯-普朗克计算机科学研究所)
主持人程明明(南开大学)
报告时间:2015年3月18日20:30(北京时间)

报告题目:Filtered Channel Features for Pedestrian Detection  [Slides]

文章信息:Shanshan Zhang, Rodrigo Benenson, Bernt Schiele. Filtered Channel Features for Pedestrian Detection. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015. (to appear)

报告摘要:This paper starts from the observation that multiple top performing pedestrian detectors can be modelled by using an intermediate layer filtering low-level features in combination with a boosted decision forest. Based on this observation we propose a unifying framework and experimentally explore different filter families. We report extensive results enabling a systematic analysis.
Using filtered channel features we obtain top performance on the challenging Caltech and KITTI datasets, while using only HOG+LUV as low-level features. When adding optical flow features we further improve detection quality and report the best known results on the Caltech dataset, reaching 93% recall at 1 FPPI.

报告人简介:张姗姗,女,德国马克斯-普朗克计算机科学研究所(MPII)计算机视觉组博士后(导师:Bernt Schiele教授)。曾分别于2008年,2011年在同济大学电信学院获得工学学士和硕士学位;2015年2月在德国波恩大学获得计算机博士学位(导师:Armin B. Cremers和Christian Bauckhage教授)。硕士期间,曾赴日本国立情报学研究所(NII)实习,师从Shin’ichi Satoh教授。目前在CVPR,ICPR,IEEE Transactions on Intelligent Transportation Systems, IEEE Transactions on Circuits and Systems for Video Technology等国际会议和国际期刊上发表论文10余篇,并担任IEEE Transactions on Neural Networks and Learning Systems,IEEE Transactions on Intelligent Transportation Systems等国际期刊的审稿人。

报告嘉宾2王兴刚(华中科大学)
主持人左旺孟
报告时间:2015年3月18日20:30(北京时间)

报告题目:Discriminative and Generative Learning for Object Discovery [Slides]

文章信息:

  1. Xinggang Wang, Baoyuan Wang, Xiang Bai, Wenyu Liu, and Zhuowen Tu, Max-Margin Multiple Instance Dictionary Learning, International Conference on Machine Learning (ICML), Atlanta, June, 2013.
  2. Xinggang Wang, Zhengdong Zhang, Yi Ma, Xiang Bai, Wenyu Liu, and Zhuowen Tu, Robust Subspace Discovery via Relaxed Rank Minimization, Neural Computation, Vol. 26, No. 3 April 2014, Pages 611-635.

报告摘要:Object discovery is an interesting topic in computer vision, which detects object in image in a weakly supervised setting. In this talk, I will introduce two methods for this task in a discriminative manner and a generative manner respectively. The discriminative approach is called max-margin multi-instance dictionary learning (MMDL). MMDL learns image codebook contains rich semantic information. Besides, the codes are compact and efficient for building image representation. In the experiments, MMDL shows excellent results on scene image classification task. The generative approach formulates the object discovery problem as a novel rank minimization problem. We relax the rank optimization problem to be a convex program and solve it using ADMM method. Thus, it is very likely to get a global optimal solution. This generative gives good results on object discovery on challenging images.

报告人简介:王兴刚,华中科技大学,电信学院,讲师。主要研究方向为计算机视觉和机器学习。分别于2009年和2014年在华中科技大学获得学士和博士学位。博士期间曾在美国加州大学洛杉矶分校(UCLA)、微软亚洲研究院、天普大学(Temple Univ.)进行访问研究和学习。是2012年微软学者奖的获得者(全亚洲仅有10名获奖者)。在领域内的期刊和会议上发表论文20余篇,其中包括NIPS、ICML、CVPR、Neural Computation等顶级会议和期刊,Google Scholar引用次数为213次。在众多领域内顶级期刊会议中担当审稿人:IEEE Transactions on Cybernetics, Pattern Recognition, Computer Vision and Image Understanding, ICDM, CVPR, ECCV, ICME, ICCV等,并在国际会议ICSPAC 2014中担任组委会成员。

 

(Visited 446 times, 1 visits today)