期刊论文、会议论文

Zheng, Kexin

英文

Proceedings of SPIE - The International Society for Optical Engineering

0277-786X

2017

10420

9th International Conference on Digital Image Processing (ICDIP)

Proceedings of SPIE

MAY 19-22, 2017

Hong Kong, PEOPLES R CHINA

[Zheng, Kexin;Lv, Shaohe;Ma, Fang;Chen, Fei;Dou, Yong] Natl Univ Def Technol, Natl Lab Parallel & Distributed Proc, 109 Deya Rd, Changsha 410073, Hunan, Peoples R China.^[Jin, Chi] Univ South China, Sch Comp Sci & Technol, Changsha 410012, Hunan, Peoples R China.

Int Assoc Comp Sci & Informat Technol, Sichuan Province Comp Federat, SW Jiaotong Univ

Falco, CM Jiang, X

1000 20TH ST, PO BOX 10, BELLINGHAM, WA 98227-0010 USA

SPIE-INT SOC OPTICAL ENGINEERING

978-1-5106-1305-8; 978-1-5106-1304-1

10.1117/12.2281747

本校为其他机构

Image annotation is a task of assigning semantic labels to an image. Recently, deep neural networks with visual attention have been utilized successfully in many computer vision tasks. In this paper, we show that conventional attention mechanism is easily misled by the salient class, i.e., the attended region always contains part of the image area describing the content of salient class at different attention iterations. To this end, we propose a novel attention shaping mechanism, which aims to maximize the non-overlapping area between consecutive attention processes by taking into account the...