Unsupervised visual domain adaptation via discriminative dictionary evolution
- PDF / 1,873,685 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 33 Downloads / 179 Views
THEORETICAL ADVANCES
Unsupervised visual domain adaptation via discriminative dictionary evolution Songsong Wu1 · Guangwei Gao2 · Zuoyong Li3 · Fei Wu4 · Xiao‑Yuan Jing5 Received: 18 March 2019 / Accepted: 30 March 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract This work focuses on unsupervised visual domain adaptation which is still challenging in visual recognition. Most of the attention has been dedicated to seeking the domain-invariant features of cross-domain data, but they ignores the valuable discriminative information in the source domain. In this paper, we propose a Discriminative Dictionary Evolution (DDE) approach to seek discriminative features robust to domain shift. Specifically, DDE gradually adapts a discriminative dictionary learned from the source domain to the target domain through a dictionary evolving procedure, in which self-selected atoms of the dictionary are updated with 𝓁2,1-norm-based regularization. DDE produces domain-invariant representations for cross-domain visual recognition meanwhile promotes the discriminativeness of the dictionary. Empirical results on realworld data sets demonstrate the advantages of the proposed approach over existing competitive methods. Keywords Cross-domain visual classification · Domain adaptation · Discriminative dictionary evolution · Feature representation learning · Transfer learning
1 Introduction In machine learning-based visual systems, the generalization performance of learning model degrades seriously when training samples from a source domain and test samples from a target domain are mismatched in data distribution [3, 54]. The issue of domain shift is caused by various dataset bias factors, e.g., various lighting conditions and viewing angles of camera in face image acquisition. Domain * Songsong Wu [email protected] 1
Guangdong University of Petrochemical Technology, No. 139, Guandu 2 Road, Maoming 525000, China
2
Institute of Advanced Technology, Nanjing University of Posts and Telecommunications, No. 9 Wenyuan Road, Nanjing 210023, China
3
Department of Computer Science, Minjiang University, No. 200 Zhenxiyuangong Road, Shangjie District, Minhou Town, Fuzhou 350108, China
4
School of Automation, Nanjing University of Posts and Telecommunications, No. 9 Wenyuan Road, Nanjing 210023, China
5
School of Computer Science, Wuhan University, No. 299, Bayi Road, Wuchang District, Wuhan 430072, China
adaptation aims to address the domain shift issue by using both the source and target samples for knowledge transfer between domains. Domain adaptation has brought promising results in cross-domain object classification [11, 22, 39], multi-view action recognition [57], person re-identification [38],video-event detection [7], personalized handwriting recognition [56] and face recognition [18, 21]. Two representative approaches to eliminate the domain shift are feature adaptation and classifier adaptation [36]. In feature level, a domain-invariant feature space is learned so that the data distributions
Data Loading...
