Balanced Discriminative Transfer Feature Learning for Visual Domain Adaptation

doi:10.12142/ZTECOM.202004011

ZTE Communications ›› 2020, Vol. 18 ›› Issue (4): 78-83.DOI: 10.12142/ZTECOM.202004011

• Research Paper • Previous Articles Next Articles

Balanced Discriminative Transfer Feature Learning for Visual Domain Adaptation

SU Limin¹, ZHANG Qiang², LI Shuang¹(), LIU Chi Harold¹

^1.Beijing Institute of Technology, Beijing 100081, China
^2.ZTE Corporation, Shenzhen 518057, China

Received:2019-12-28 Online:2020-12-25 Published:2021-01-13
About author:SU Limin is pursuing the M.Sc. degree in computer science at Beijing Institute of Technology, China. Her research interests include machine learning and transfer learning.|ZHANG Qiang received his B.S. degree in computer science from Hefei University of Technology, China in 1995. Working with ZTE Corporation, he was the director of the research team of TMN Standards from 2003 to 2006 and has been an NMS/OSS solution manager since 2007.|LI Shuang (shuangli@bit.edu.cn) received his Ph.D. degree in Department of Automation, Tsinghua University, China in 2018. He is an assistant professor in the School of Computer Science and Technology, Beijing Institute of Technology, China. He was a visiting research scholar at the Department of Computer Science, Cornell University, USA from November 2015 to June 2016. His main research interests include machine learning and deep learning, especially in transfer learning and domain adaptation.|Chi Harold LIU receives the Ph.D. degree from Imperial College, UK in 2010, and the B.Eng. degree from Tsinghua University, China in 2006. He is currently a full professor and the vice dean at the School of Computer Science and Technology, Beijing Institute of Technology, China. He is also the director of IBM Mainframe Excellence Center (Beijing), IBM Big Data Technology Center, and National Laboratory of Data Intelligence for China Light Industry. Before moving to academia, he joined IBM Research, China as a staff researcher and project manager, after working as a postdoctoral researcher at Deutsche Telekom Laboratories, Germany and a visiting scholar at IBM T. J. Watson Research Center, USA. His current research interests include the Internet of Things (IoT), big data analytics, mobile computing and deep learning. He received the Distinguished Young Scholar Award in 2013, IBM First Plateau Invention Achievement Award in 2012, and IBM First Patent Application Award in 2011, and was interviewed by EEWeb.com as the Featured Engineer in 2011. He has published more than 80 prestigious conference and journal papers and owned more than 14 EU/USA/UK/China patents. He serves as the area editor for KSII Transactions on Internet and Information Systems and the book editor for six books published by Taylor & Francis Group, USA and China Machinery Press. He also served as the general chair of IEEE SECON’13 workshop on IoT Networking and Control, IEEE WCNC’12 workshop on IoT Enabling Technologies, and ACM UbiComp’11 Workshop on Networking and Object Memories for IoT. Moreover, he served as the consultant to Asian Development Bank, Bain & Company, and KPMG, USA, and the peer reviewer for Qatar National Research Foundation and National Science Foundation, China. He is a senior member of IEEE.

Abstract

Abstract:

Transfer learning aims to transfer source models to a target domain. Leveraging the feature matching can alleviate the domain shift effectively, but this process ignores the relationship of the marginal distribution matching and the conditional distribution matching. Simultaneously, the discriminative information of both domains is also neglected, which is important for improving the performance on the target domain. In this paper, we propose a novel method called Balanced Discriminative Transfer Feature Learning for Visual Domain Adaptation (BDTFL). The proposed method can adaptively balance the relationship of both distribution matchings and capture the category discriminative information of both domains. Therefore, balanced feature matching can achieve more accurate feature matching and adaptively adjust itself to different scenes. At the same time, discriminative information is exploited to alleviate category confusion during feature matching. And with assistance of the category discriminative information captured from both domains, the source classifier can be transferred to the target domain more accurately and boost the performance of target classification. Extensive experiments show the superiority of BDTFL on popular visual cross-domain benchmarks.

Key words: transfer learning, domain adaptation, distribution adaptation, discriminative information

SU Limin, ZHANG Qiang, LI Shuang, LIU Chi Harold. Balanced Discriminative Transfer Feature Learning for Visual Domain Adaptation[J]. ZTE Communications, 2020, 18(4): 78-83.

Figures/Tables 3

References 21

1	PAN S J, YANG Q. A survey on transfer learning [J]. IEEE transactions on knowledge and data engineering, 2010, 22(10): 1345–1359. DOI: 10.1109/TKDE.2009.191 DOI
2	GONG B Q, GRAUMAN K, SHA F. Connecting the dots with landmarks: discriminatively learning domain⁃invariant features for unsupervised domain adaptation [C]//30th International Conference on International Conference on Machine Learning. Atlanta, USA: Curran Associates, 2013: 222–230
3	DUAN L X, TSANG I W, XU D. Domain transfer multiple kernel learning [J]. IEEE transactions on pattern analysis and machine intelligence, 2012, 34(3): 465–479. DOI: 10.1109/TPAMI.2011.114 DOI
4	TAN B, ZHANG Y, PAN S J, et al. Distant domain transfer learning [C]//31st AAAI Conference on Artificial Intelligence. San Francisco, USA: AAAI Press, 2017: 2604–2610
5	CHANG W⁃C, WU Y X, LIU H X, et al. Cross⁃domain kernel induction for transfer learning [C]//31st AAAI Conference on Artificial Intelligence. San Francisco, USA: AAAI Press, 2017: 1763–1769
6	PAN S J, TSANG I W, KWOK J T, et al. Domain adaptation via transfer component analysis [J]. IEEE transactions on neural networks, 2011, 22(2): 199–210. DOI: 10.1109/TNN.2010.2091281 DOI
7	GRETTON A, BORGWARDT K M, RASCH M, et al. A kernel method for the two⁃sample⁃problem [C]//19th International Conference on Neural Information Processing Systems. Kitakyushu, Japan: Springer, 2007: 513–520
8	LONG M S, WANG J M, DING G G, et al. Transfer feature learning with joint distribution adaptation [C]//IEEE International Conference on Computer Vision. Sydney, Australia: IEEE, 2013: 2200–2207. DOI: 10.1109/ICCV.2013.274 DOI
9	LONG M S, WANG J M, DING G G, et al. Transfer joint matching for unsupervised domain adaptation [C]//IEEE Conference on Computer Vision and Pattern Recognition. Columbus, USA: IEEE, 2014: 1410–1417. DOI: 10.1109/CVPR.2014.183 DOI
10	BAKTASHMOTLAGH M, HARANDI M T, LOVELL B C, et al. Unsupervised domain adaptation by domain invariant projection [C]//IEEE International Conference on Computer Vision. Sydney, Australia: IEEE, 2013: 769–776. DOI: 10.1109/ICCV.2013.100 DOI
11	LI S, SONG S J, HUANG G, et al. Domain invariant and class‑discriminative feature learning for visual domain adaptation [J]. IEEE transactions on image processing, 2018, 27(9): 4260–4273. DOI: 10.1109/TIP.2018.2839528 DOI
12	SCHROFF F, KALENICHENKO D, PHILBIN J. FaceNet: A unified embedding for face recognition and clustering [EB/OL]. (2015⁃03⁃12)[2018⁃01⁃01].
13	XIAO Q Q, LUO H, ZHANG C. Margin sample mining loss: a deep learning based method for person Re⁃identification [EB/OL]. (2017⁃10⁃02)[2018⁃01⁃01].
14	LI X Y, HERRANZ L, JIANG S Q. Multifaceted analysis of fine⁃tuning in a deep model for visual recognition [J]. IMS transactions on data science, 2020, 1(1): 1–22. DOI: 10.1145/3319500 DOI
15	SAENKO K, KULIS B, FRITZ M, et al. Adapting visual category models to new domains [C]//11th European Conference on Computer Vision. Heraklion, Greece: IEEE, 2010: 213–226
16	SIM T, BAKER S, BSAT M, pose The CMU, illumination, and expression (PIE) database [C]//Fifth IEEE International Conference on Automatic Face and Gesture Recognition. Washington, USA: IEEE, 2002. DOI: 10.1109/AFGR.2002.1004130 DOI
17	FUKUNAGA K, NARENDRA P M. A branch and bound algorithm for computing k⁃nearest neighbors [J]. IEEE transactions on computers, 1975, C⁃24(7): 750–753. DOI: 10.1109/T-C.1975.224297 DOI
18	JOLLIFFE I T. Principal component analysis [M]. New York, USA: Springer⁃Verlag, 2011: 1094–1096. DOI: 10.1007/b98835 DOI
19	GONG B, SHI Y, SHA F, et al. Geodesic flow kernel for unsupervised domain adaptation [C]//IEEE Conference on Computer Vision and Pattern Recognition. Providence, USA: IEEE, 2012: 2066–2073
20	FERNANDO B, HABRARD A, SEBBAN M, et al. Unsupervised visual domain adaptation using subspace alignment [C]//IEEE International Conference on Computer Vision. Sydney, Australia: IEEE, 2013: 2960–2967. DOI: 10.1109/ICCV.2013.368 DOI
21	XU Y, FANG X Z, WU J, et al. Discriminative transfer subspace learning via low⁃rank and sparse representation [J]. IEEE transactions on image processing, 2016, 25(2): 850–863. DOI: 10.1109/TIP.2015.2510498 DOI

Task/Method	1NN	PCA	GFK	TCA	JDA	SA	DTSL	BDTFL
A→D	59.6	60.6	52.0	56.2	56.8	61.0	60.0	62.65
A→W	54.0	55.6	48.2	54.8	58.1	59.5	54.5	58.62
D→A	42.4	44.6	41.8	44.3	44.8	46.9	46.6	54.56
D→W	90.9	91.8	86.5	92.1	95.7	95.1	94.3	96.60
W→A	40.8	42.0	38.6	42.1	46.2	46.6	45.6	51.79
W→D	97.8	98.6	87.5	95.6	97.6	98.2	94.0	98.59
Average	64.3	65.5	59.1	64.2	66.5	67.9	65.8	70.47

Task/Method	1NN	PCA	GFK	TCA	JDA	SA	DTSL	BDTFL
A→D	59.6	60.6	52.0	56.2	56.8	61.0	60.0	62.65
A→W	54.0	55.6	48.2	54.8	58.1	59.5	54.5	58.62
D→A	42.4	44.6	41.8	44.3	44.8	46.9	46.6	54.56
D→W	90.9	91.8	86.5	92.1	95.7	95.1	94.3	96.60
W→A	40.8	42.0	38.6	42.1	46.2	46.6	45.6	51.79
W→D	97.8	98.6	87.5	95.6	97.6	98.2	94.0	98.59
Average	64.3	65.5	59.1	64.2	66.5	67.9	65.8	70.47

Task/Method	1NN	PCA	GFK	TCA	JDA	SA	DTSL	BDTFL
C05 →C07	26.1	24.8	26.2	40.8	58.6	26.8	65.9	67.3
C05 →C09	26.6	25.2	27.3	41.8	52.0	28.2	64.1	68.1
C05→C27	30.7	29.3	31.2	59.6	83.7	30.9	82.0	89.9
C05 →C29	16.7	16.3	17.6	29.4	47.7	19.6	54.9	58.6
C07→C05	24.5	24.2	25.2	41.8	60.6	26.4	45.0	59.8
C07→C09	46.6	45.5	47.4	51.5	60.2	48.0	53.5	62.7
C07→C27	54.1	53.4	54.3	64.7	75.4	54.3	71.4	80.4
C07 →C29	26.5	25.4	27.1	33.7	40.9	28.2	48.0	49.6
C09 →C05	21.4	21.0	21.8	34.7	50.9	23.2	52.5	57.7
C09→C07	41.0	40.5	43.2	47.7	56.1	44.3	55.6	61.1
C09 →C27	46.5	46.1	46.4	56.2	68.0	46.2	77.5	76.1
C09 →C29	26.2	25.3	26.7	33.2	40.3	28.9	54.1	53.2
C27→C05	33.0	32.0	34.2	55.6	81.0	36.3	81.5	86.8
C27 →C07	62.7	61.0	62.9	67.8	82.8	63.8	85.4	86.6
C27→C09	73.2	72.2	73.4	75.9	87.2	73.2	82.2	86.5
C27 →C29	37.2	35.1	37.4	40.3	49.9	38.1	72.6	72.8
C29→C05	18.5	18.9	20.4	27.0	47.5	23.4	52.2	59.1
C29 →C07	24.2	23.4	24.6	30.0	44.8	25.5	49.4	53.6
C29→C09	28.3	27.2	28.5	30.0	48.1	28.6	58.5	57.6
C29 →C27	31.2	30.3	31.3	33.6	56.5	31.2	64.3	66.0
Average	34.8	33.9	35.4	44.8	59.6	36.3	63.5	67.7

Task/Method	1NN	PCA	GFK	TCA	JDA	SA	DTSL	BDTFL
C05 →C07	26.1	24.8	26.2	40.8	58.6	26.8	65.9	67.3
C05 →C09	26.6	25.2	27.3	41.8	52.0	28.2	64.1	68.1
C05→C27	30.7	29.3	31.2	59.6	83.7	30.9	82.0	89.9
C05 →C29	16.7	16.3	17.6	29.4	47.7	19.6	54.9	58.6
C07→C05	24.5	24.2	25.2	41.8	60.6	26.4	45.0	59.8
C07→C09	46.6	45.5	47.4	51.5	60.2	48.0	53.5	62.7
C07→C27	54.1	53.4	54.3	64.7	75.4	54.3	71.4	80.4
C07 →C29	26.5	25.4	27.1	33.7	40.9	28.2	48.0	49.6
C09 →C05	21.4	21.0	21.8	34.7	50.9	23.2	52.5	57.7
C09→C07	41.0	40.5	43.2	47.7	56.1	44.3	55.6	61.1
C09 →C27	46.5	46.1	46.4	56.2	68.0	46.2	77.5	76.1
C09 →C29	26.2	25.3	26.7	33.2	40.3	28.9	54.1	53.2
C27→C05	33.0	32.0	34.2	55.6	81.0	36.3	81.5	86.8
C27 →C07	62.7	61.0	62.9	67.8	82.8	63.8	85.4	86.6
C27→C09	73.2	72.2	73.4	75.9	87.2	73.2	82.2	86.5
C27 →C29	37.2	35.1	37.4	40.3	49.9	38.1	72.6	72.8
C29→C05	18.5	18.9	20.4	27.0	47.5	23.4	52.2	59.1
C29 →C07	24.2	23.4	24.6	30.0	44.8	25.5	49.4	53.6
C29→C09	28.3	27.2	28.5	30.0	48.1	28.6	58.5	57.6
C29 →C27	31.2	30.3	31.3	33.6	56.5	31.2	64.3	66.0
Average	34.8	33.9	35.4	44.8	59.6	36.3	63.5	67.7

Balanced Discriminative Transfer Feature Learning for Visual Domain Adaptation

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 3

References 21

Related Articles 1

Recommended Articles 0

Metrics