Transfer classification for distinct manifestations with shared information
Qi, Lu1; Yin, Peijie2; Huang, Xiayuan2; Chen, Ken3; Qiao, Hong1; Qi, L
2016
会议名称12th World Congress on Intelligent Control and Automation (WCICA)
会议录名称PROCEEDINGS OF THE 2016 12TH WORLD CONGRESS ON INTELLIGENT CONTROL AND AUTOMATION (WCICA)
会议日期JUN 12-15, 2016
会议地点Guilin, PEOPLES R CHINA
摘要An object often has many distinct manifestations in computer vision, which brings a great challenge to utilizing more comprehensive information. Inspired by some biological researches about edge sensitivity and global structure priority, our key insight is to establish unified transfer classification network withshared contour information. Combining two convolutional networks with three cascaded filters, we build a unified kernel SVM classifier based on shared contour features. Two convolutional networks are usedfor acquiring the contour information of objects exactly. Obtained by three cascaded filters, sharededge features are used by a unified kernels SVM classifier. Our transfer classification network(TCN) is trained and tested with distinct manifestations including real photos(imagenet dataset or cifar-10 dataset) and cartoon abstracts. The model is able to extract robust contour features and achieve considerable transfer recognition accuracy(40% relative improvement to some popular convolutional models).
关键词Visual-cortex
收录类别EI
文献类型会议论文
条目标识符http://ir.ia.ac.cn/handle/173211/12827
专题复杂系统管理与控制国家重点实验室_机器人理论与应用
通讯作者Qi, L
作者单位1.Institution of Automation, Chinese Academy of Sciences
2.Academy of Mathematics and Systems Sciences, Chinese Academy of Sciences
3.Tsinghua University
推荐引用方式
GB/T 7714
Qi, Lu,Yin, Peijie,Huang, Xiayuan,et al. Transfer classification for distinct manifestations with shared information[C],2016.
条目包含的文件 下载所有文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
Transfer classificat(1351KB)会议论文 开放获取CC BY-NC-SA浏览 下载
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[Qi, Lu]的文章
[Yin, Peijie]的文章
[Huang, Xiayuan]的文章
百度学术
百度学术中相似的文章
[Qi, Lu]的文章
[Yin, Peijie]的文章
[Huang, Xiayuan]的文章
必应学术
必应学术中相似的文章
[Qi, Lu]的文章
[Yin, Peijie]的文章
[Huang, Xiayuan]的文章
相关权益政策
暂无数据
收藏/分享
文件名: Transfer classification for distinct manifestations.pdf
格式: Adobe PDF
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。