20180711-20 郑银强：Data-driven Optimal Camera Response Selection and Design for ...

报告嘉宾：郑银强（日本国立信息学研究所）

报告时间：2018年07月11日（星期三）晚上20:00（北京时间）

报告题目：Data-driven Optimal Camera Response Selection and Design for Spectral Reconstruction

主持人：付莹（北京理工大学）

报告人简介：

Yinqiang Zheng received his Bachelor degree from the Department of Automation, Tianjin University, Tianjin, China, in 2006, Master degree of engineering from Shanghai Jiao Tong University, Shanghai, China, in 2009, and Doctoral degree of engineering from the Department of Mechanical and Control Engineering, Tokyo Institute of Technology, Tokyo, Japan, in 2013. He is currently an assistant professor in the National Institute of Informatics, Japan. His research interests include 3D reconstruction, spectral imaging, biomedical imaging and mathematical optimization.

个人主页：

https://researchmap.jp/yinqiangzheng/

报告摘要：

Spectral construction from a multichannel image, in particular, from an RGB image, is promising in avoiding the cost and resolution issues of existing spectral devices. In this talk, I will first introduce our deep neural network based methods  for spectral reconstruction, and evaluate the effect of camera response onto spectral reconstruction accuracy. Later, I will show how to automatically select the optimal camera response from a candidate pool by introducing a camera response selection layer. Finally, the response selection layer will be replaced by a response design layer, which enables to design even better camera response for the purpose of spectral reconstruction. The algorithmically designed camera response has been realized  by using film filters and a prototype camera system constructed, which has drawn the curtain of data-driven and task-oriented optimal hardware design. 

参考文献：

[1] Shijie Nie, Ling Gu, Yinqiang Zheng, Antony Lam, Nobutaka Ono and Imari Sato, Deeply Learned Filter Response Functions for Hyperspectral Reconstruction, IEEE Conference on Computer Vision and Pattern Recognition (CVPR2018), pp. 4767-4776, Salk Lake City, Utah, USA, June, 2018.
[2] Boaz Arad, Ohad Ben-Shahar, Filter selection for hyper-spectral estimation. IEEE International Conference on Computer Vision (ICCV2017), pp.3172–3180, Venice, Italy, October, 2017.
[3] Yan Jia, Yinqiang Zheng, Lin Gu, Art Subpa-Asa, Antony Lam, Yoichi Sato and Imari Sato, From RGB to Spectrum for Natural Scenes via Manifold-based Mapping, IEEE International Conference on Computer Vision (ICCV2017), pp.4715-4723, Venice, Italy, October, 2017.

18-20期VALSE在线学术报告参与方式：

长按或扫描下方二维码，关注”VALSE“微信公众号（valse_wechat），后台回复”20期“，获取直播地址。

特别鸣谢本次Webinar主要组织者：

VOOC责任委员：付莹（北京理工大学）

VODB协调理事：左旺孟（哈尔滨工业大学）

活动参与方式：

1、VALSE Webinar活动依托在线直播平台进行，活动时讲者会上传PPT或共享屏幕，听众可以看到Slides，听到讲者的语音，并通过聊天功能与讲者交互；

2、为参加活动，请关注VALSE微信公众号：valse_wechat 或加入VALSE QQ群（目前A、B、C、D、E、F、G群已满，除讲者等嘉宾外，只能申请加入VALSE H群，群号：701662399）；

*注：申请加入VALSE QQ群时需验证姓名、单位和身份，缺一不可。入群后，请实名，姓名身份单位。身份：学校及科研单位人员T；企业研发I；博士D；硕士M。

3、在活动开始前5分钟左右，讲者会开启直播，听众点击直播链接即可参加活动，支持安装Windows系统的电脑、MAC电脑、手机等设备；

4、活动过程中，请不要说无关话语，以免影响活动正常进行；

5、活动过程中，如出现听不到或看不到视频等问题，建议退出再重新进入，一般都能解决问题；

6、建议务必在速度较快的网络上参加活动，优先采用有线网络连接；

7、VALSE微信公众号会在每周一推送上一周Webinar报告的总结及视频（经讲者允许后），每周四发布下一周Webinar报告的通知及直播链接。

[slides]