ScoreMix: A Scalable Augmentation Strategy for Training GANs With Limited Data

doi:10.1109/TPAMI.2022.3231649

CORC > 自动化研究所 > 中国科学院自动化研究所 > 多模态人工智能系统全国重点实验室

	ScoreMix: A Scalable Augmentation Strategy for Training GANs With Limited Data
	Cao, Jie 2,3,4; Luo, Mandi 2,3,4; Yu, Junchi 2,3,4; Yang, Ming-Hsuan 1; He, Ran 2,3,4
刊名	IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
	2023-07-01
卷号	45 期号:7 页码:8920-8935
关键词	Generative adversarial networks image synthesis data augmentation few-shot image-to-image translation
ISSN号	0162-8828
DOI	10.1109/TPAMI.2022.3231649
通讯作者	He, Ran(ran.he@ia.ac.cn)
英文摘要	Generative Adversarial Networks (GANs) typically suffer from overfitting when limited training data is available. To facilitate GAN training, current methods propose to use data-specific augmentation techniques. Despite the effectiveness, it is difficult for these methods to scale to practical applications. In this article, we present ScoreMix, a novel and scalable data augmentation approach for various image synthesis tasks. We first produce augmented samples using the convex combinations of the real samples. Then, we optimize the augmented samples by minimizing the norms of the data scores, i.e., the gradients of the log-density functions. This procedure enforces the augmented samples close to the data manifold. To estimate the scores, we train a deep estimation network with multi-scale score matching. For different image synthesis tasks, we train the score estimation network using different data. We do not require the tuning of the hyperparameters or modifications to the network architecture. The ScoreMix method effectively increases the diversity of data and reduces the overfitting problem. Moreover, it can be easily incorporated into existing GAN models with minor modifications. Experimental results on numerous tasks demonstrate that GAN models equipped with the ScoreMix method achieve significant improvements.
资助项目	National Natural Science Foundation of China[6220073425] ; National Natural Science Foundation of China[U21B2045] ; National Natural Science Foundation of China[U20A20223] ; NSF CAREER[1149783]
WOS研究方向	Computer Science ; Engineering
语种	英语
出版者	IEEE COMPUTER SOC
WOS记录号	WOS:001004665900065
资助机构	National Natural Science Foundation of China ; NSF CAREER
内容类型	期刊论文
源URL	[http://ir.ia.ac.cn/handle/173211/53630]
专题	多模态人工智能系统全国重点实验室
通讯作者	He, Ran
作者单位	1.Univ Calif Merced, Merced, CA 95343 USA 2.Univ Chinese Acad Sci, Beijing 101408, Peoples R China 3.Chinese Acad Sci, Inst Automat, NLPR, Beijing 100045, Peoples R China 4.Chinese Acad Sci, Inst Automat, CRIPAC, Beijing 100045, Peoples R China
推荐引用方式 GB/T 7714	Cao, Jie,Luo, Mandi,Yu, Junchi,et al. ScoreMix: A Scalable Augmentation Strategy for Training GANs With Limited Data[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,2023,45(7):8920-8935.
APA	Cao, Jie,Luo, Mandi,Yu, Junchi,Yang, Ming-Hsuan,&He, Ran.(2023).ScoreMix: A Scalable Augmentation Strategy for Training GANs With Limited Data.IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,45(7),8920-8935.
MLA	Cao, Jie,et al."ScoreMix: A Scalable Augmentation Strategy for Training GANs With Limited Data".IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 45.7(2023):8920-8935.