Image captioning via hierarchical attention mechanism and policy gradient optimization

doi:10.1016/j.sigpro.2019.107329

CASIA OpenIR > 精密感知与控制研究中心

	Image captioning via hierarchical attention mechanism and policy gradient optimization
	Yan, Shiyang 1; Xie, Yuan2,3,5,7 ; Wu, Fangyu 4,6; Smith, Jeremy S.4; Lu, Wenjin 6; Zhang, Bailing 2,3
发表期刊	SIGNAL PROCESSING
ISSN	0165-1684
	2020-02-01
卷号	167 页码:12
通讯作者	Yan, Shiyang(shiyang.yan@qub.ac.uk)
摘要	Automatically generating the descriptions of an image, i.e., image captioning, is an important and fundamental topic in artificial intelligence, which bridges the gap between computer vision and natural language processing. Based on the successful deep learning models, especially the CNN model and Long Short Term Memories (LSTMs) with attention mechanism, we propose a hierarchical attention model by utilizing both of the global CNN features and the local object features for more effective feature representation and reasoning in image captioning. The generative adversarial network (GAN), together with a reinforcement learning (RL) algorithm, is applied to solve the exposure bias problem in RNN-based supervised training for language problems. In addition, through the automatic measurement of the consistency between the generated caption and the image content by the discriminator in the GAN framework and RL optimization, we make the finally generated sentences more accurate and natural. Comprehensive experiments show the improved performance of the hierarchical attention mechanism and the effectiveness of our RL-based optimization method. Our model achieves state-of-the-art results on several important metrics in the MSCOCO dataset, using only greedy inference. (C) 2019 Elsevier B.V. All rights reserved.
关键词	Image captioning Hierarchical attention mechanism Generative adversarial network Reinforcement learning Policy gradient
DOI	10.1016/j.sigpro.2019.107329
关键词[WOS]	NETWORKS
收录类别	SCI
语种	英语
WOS研究方向	Engineering
WOS类目	Engineering, Electrical & Electronic
WOS记录号	WOS:000497600200030
出版者	ELSEVIER
引用统计	被引频次：24[WOS] [WOS记录] [WOS相关记录]
文献类型	期刊论文
条目标识符	http://ir.ia.ac.cn/handle/173211/29388
专题	精密感知与控制研究中心
通讯作者	Yan, Shiyang
作者单位	1.Queens Univ Belfast, Sch Elect Elect Engn & Comp Sci, Belfast, Antrim, North Ireland 2.Inst Adv Artificial Intelligence Nanjing, Nanjing, Jiangsu, Peoples R China 3.Horizon Robot, Beijing, Peoples R China 4.Univ Liverpool, Elect Engn & Elect, Liverpool, Merseyside, England 5.Chinese Acad Sci, Inst Automat, Beijing, Peoples R China 6.Xian Jiaotong Liverpool Univ, Dept Comp Sci & Software Engn, Suzhou, Peoples R China 7.East China Normal Univ, Sch Comp Sci & Software Engn, Shanghai, Peoples R China
推荐引用方式 GB/T 7714	Yan, Shiyang,Xie, Yuan,Wu, Fangyu,et al. Image captioning via hierarchical attention mechanism and policy gradient optimization[J]. SIGNAL PROCESSING,2020,167:12.
APA	Yan, Shiyang,Xie, Yuan,Wu, Fangyu,Smith, Jeremy S.,Lu, Wenjin,&Zhang, Bailing.(2020).Image captioning via hierarchical attention mechanism and policy gradient optimization.SIGNAL PROCESSING,167,12.
MLA	Yan, Shiyang,et al."Image captioning via hierarchical attention mechanism and policy gradient optimization".SIGNAL PROCESSING 167(2020):12.