3D-Aided Dual-Agent GANs for Unconstrained Face Recognition

doi:10.1109/TPAMI.2018.2858819

	3D-Aided Dual-Agent GANs for Unconstrained Face Recognition
	Zhao, Jian 1,2; Xiong, Lin 3; Li, Jianshu 4; Xing, Junliang5 ; Yan, Shuicheng 1,6; Feng, Jiashi 1
刊名	IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
	2019-10-01
卷号	41 期号:10 页码:2380-2394
关键词	Face synthesis unconstrained face recognition 3D face model generative adversarial networks
ISSN号	0162-8828
DOI	10.1109/TPAMI.2018.2858819
通讯作者	Zhao, Jian(zhaojian90@u.nus.edu)
英文摘要	Synthesizing realistic profile faces is beneficial for more efficiently training deep pose-invariant models for large-scale unconstrained face recognition, by augmenting the number of samples with extreme poses and avoiding costly annotation work. However, learning from synthetic faces may not achieve the desired performance due to the discrepancy betwedistributions of the synthetic and real face images. To narrow this gap, we propose a Dual-Agent Generative Adversarial Network (DA-GAN) model, which can improve the realism of a face simulator's output using unlabeled real faces while preserving the identity information during the realism refinement. The dual agents are specially designed for distinguishing real versus fake and identities simultaneously. In particular, we employ an off-the-shelf 3D face model as a simulator to generate profile face images with varying poses. DA-GAN leverages a fully convolutional network as the generator to generate high-resolution images and an auto-encoder as the discriminator with the dual agents. Besides the novel architecture, we make several key modifications to the standard GAN to preserve pose, texture as well as identity, and stabilize the training process: (i) a pose perception loss; (ii) an identity perception loss; (iii) an adversarial loss with a boundary equilibrium regularization term. Experimental results show that DA-GAN not only achieves outstanding perceptual results but also significantly outperforms state-of-the-arts on the large-scale and challenging NIST IJB-A and CFP unconstrained face recognition benchmarks. In addition, the proposed DA-GAN is also a promising new approach for solving generic transfer learning problems more effectively. DA-GAN is the foundation of our winning entry to the NIST IJB-A face recognition competition in which we secured the 1st places on the tracks of verification and identification.
资助项目	China Scholarship Council (CSC)[201503170248] ; National Science Foundation of Chian[61672519] ; National University of Singapore[R-263-000-C08-133] ; National University of Singapore[MOE Tier-I R-263-000-C21-112] ; National University of Singapore[NUS IDS R-263-000-C67-646] ; National University of Singapore[ECRA R-263-000-C87-133]
WOS研究方向	Computer Science ; Engineering
语种	英语
出版者	IEEE COMPUTER SOC
WOS记录号	WOS:000489763000008
资助机构	China Scholarship Council (CSC) ; National Science Foundation of Chian ; National University of Singapore
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/26672]
专题	中国科学院自动化研究所
通讯作者	Zhao, Jian
作者单位	1.Natl Univ Singapore, Dept Elect & Comp Engn, Singapore 119077, Singapore 2.Natl Univ Def Technol, Sch Comp, Changsha 410073, Hunan, Peoples R China 3.Panasonic R&D Ctr Singapore, Core Technol Grp, Learning & Vis, Singapore 469332, Singapore 4.Natl Univ Singapore, Sch Comp, Singapore 119077, Singapore 5.Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100864, Peoples R China 6.Qihoo 360 AI Inst, Beijing 100015, Peoples R China
推荐引用方式 GB/T 7714	Zhao, Jian,Xiong, Lin,Li, Jianshu,et al. 3D-Aided Dual-Agent GANs for Unconstrained Face Recognition[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,2019,41(10):2380-2394.
APA	Zhao, Jian,Xiong, Lin,Li, Jianshu,Xing, Junliang,Yan, Shuicheng,&Feng, Jiashi.(2019).3D-Aided Dual-Agent GANs for Unconstrained Face Recognition.IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,41(10),2380-2394.
MLA	Zhao, Jian,et al."3D-Aided Dual-Agent GANs for Unconstrained Face Recognition".IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 41.10(2019):2380-2394.