A Comparative Study of CNN- and Transformer-Based Visual Style Transfer

doi:10.1007/s11390-022-2140-7

	A Comparative Study of CNN- and Transformer-Based Visual Style Transfer
	Wei, Hua-Peng 1; Deng, Ying-Ying2 ; Tang, Fan1 ; Pan, Xing-Jia3 ; Dong, Wei-Ming2
发表期刊	JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY
ISSN	1000-9000
	2022-06-01
卷号	37 期号:3 页码:601-614
通讯作者	Tang, Fan(tangfan@jlu.edu.cn)
摘要	Vision Transformer has shown impressive performance on the image classification tasks. Observing that most existing visual style transfer (VST) algorithms are based on the texture-biased convolution neural network (CNN), here raises the question of whether the shape-biased Vision Transformer can perform style transfer as CNN. In this work, we focus on comparing and analyzing the shape bias between CNN- and transformer-based models from the view of VST tasks. For comprehensive comparisons, we propose three kinds of transformer-based visual style transfer (Tr-VST) methods (Tr-NST for optimization-based VST, Tr-WCT for reconstruction-based VST and Tr-AdaIN for perceptual-based VST). By engaging three mainstream VST methods in the transformer pipeline, we show that transformer-based models pre-trained on ImageNet are not proper for style transfer methods. Due to the strong shape bias of the transformer-based models, these Tr-VST methods cannot render style patterns. We further analyze the shape bias by considering the inuence of the learned parameters and the structure design. Results prove that with proper style supervision, the transformer can learn similar texture-biased features as CNN does. With the reduced shape bias in the transformer encoder, Tr-VST methods can generate higher-quality results compared with state-of-the-art VST methods.
关键词	transformer convolution neural network visual style transfer comparative study
DOI	10.1007/s11390-022-2140-7
收录类别	SCI
语种	英语
资助项目	National Key Research and Development Program of China[2020AAA0106200] ; National Natural Science Foundation of China[62102162] ; National Natural Science Foundation of China[61832016] ; National Natural Science Foundation of China[U20B2070] ; National Natural Science Foundation of China[6210070958] ; CASIA-Tencent Youtu Joint Research Project ; Open Projects Program of the National Laboratory of Pattern Recognition
项目资助者	National Key Research and Development Program of China ; National Natural Science Foundation of China ; CASIA-Tencent Youtu Joint Research Project ; Open Projects Program of the National Laboratory of Pattern Recognition
WOS研究方向	Computer Science
WOS类目	Computer Science, Hardware & Architecture ; Computer Science, Software Engineering
WOS记录号	WOS:000812520400008
出版者	SCIENCE PRESS
引用统计	被引频次：11[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/173211/49223
专题	多模态人工智能系统全国重点实验室_多媒体计算
通讯作者	Tang, Fan
作者单位	1.Jilin Univ, Sch Artificial Intelligence, Changchun 130012, Peoples R China 2.Chinese Acad Sci, Inst Automat, Natl Lab Pattern Recognit, Beijing 100190, Peoples R China 3.Tencent Inc, Youtu Lab, Shanghai 200233, Peoples R China
推荐引用方式 GB/T 7714	Wei, Hua-Peng,Deng, Ying-Ying,Tang, Fan,et al. A Comparative Study of CNN- and Transformer-Based Visual Style Transfer[J]. JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY,2022,37(3):601-614.
APA	Wei, Hua-Peng,Deng, Ying-Ying,Tang, Fan,Pan, Xing-Jia,&Dong, Wei-Ming.(2022).A Comparative Study of CNN- and Transformer-Based Visual Style Transfer.JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY,37(3),601-614.
MLA	Wei, Hua-Peng,et al."A Comparative Study of CNN- and Transformer-Based Visual Style Transfer".JOURNAL OF COMPUTER SCIENCE AND TECHNOLOGY 37.3(2022):601-614.