智能感知与计算研究中心
中国科学院自动化研究所   联系我们    English
 
    学术讲座

Going deeper and wider for semantic segmentation

智能感知与计算系列讲座
Lecture Series in Intelligent Perception and Computing 

    目(TITLE):Going deeper and wider for semantic segmentation

人(SPEAKER: Dr. Fisher Yu, Princeton University

(CHAIR)Dr. Wei Wang

    (TIME)June 1, 2016 (Wednesday), 14:30PM

    (VENUE)Meeting Room, 16 Floor, Intelligent building

报告摘要(ABSTRACT):
Semantic segmentation and many other computer vision tasks require predicting an output for each pixel of an input image. Recently, with the success of deep learning methods on image classification, a lot of works have studied how to use the network designed for image classification as basis to solve the segmentation problem. In this talk, I argue that in order to design better model for image segmentation, we have to change the basis to make it more suitable for segmentation. Towards this goal, I introduce dilated convolution as an additional building block for semantic segmentation and talk about its properties. We use dilated convolution to build a deeper network that has much wider receptive field than the original image classification network. Our model has competitive performance on different datasets and it achieves the state-of-the-art results on Cityscape, CamVid and KITTI datasets even without structured prediction. Further, I will introduce my on-going project, LSUN, which aims to make it possible to explore weakly annotated data to help solve the image segmentation problem. In LSUN project, we have collected orders of magnitude more labeled and unlabeled images than any previous datasets, opening up new avenues for computer vision research.

报告人简介(BIOGRAPHY):
Fisher Yu is a fifth-year Ph.D. student in Princeton University, advised by Prof. Jianxiong Xiao and Prof. Thomas Funkhouser. He is also working with Vladlen Koltun, director of Visual Computing Lab in Intel, on image semantic segmentation. Before that, he received BSE and MSE from University of Michigan. His research is in computer vision and he is interested in studying deep learning networks and exploring big data. He contributes to ShapeNet project, which is used for large-scale 3D shape analysis. Together with the other members at Princeton Vision Group, he organized LSUN Challenge Workshop in CVPR 2015 and 2016.

友情链接
 
中科院自动化研究所 智能感知与计算研究中心
中国科学院自动化研究所  事业单位   京ICP备14019135号-3