TY - JOUR
T1 - Improving the Generalization Ability of Deep Neural Networks for Cross-Domain Visual Recognition
AU - Zheng, Jianwei
AU - Lu, Chao
AU - Hao, Cong
AU - Chen, Deming
AU - Guo, Donghui
N1 - Funding Information:
Manuscript received October 8, 2019; revised December 9, 2019; accepted January 5, 2020. Date of publication January 9, 2020; date of current version September 9, 2021. This work was supported in part by the National Natural Science Foundation of China under Grant 61274133 and Grant 61836010. (Corresponding author: Donghui Guo.) Jianwei Zheng was with the Department of Electrical and Computer Engineering, University of Illinois at Urbana–Champaign, Urbana, IL, USA. He is now with the Department of Electronic Engineering, Xiamen University, Xiamen 360005, China (e-mail: jianweizheng2013@gmail.com).
Funding Information:
ACKNOWLEDGMENT The work was completed during the time when the Jianwei Zheng studied as a visiting Ph.D. student in the Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, which was supported by the China Scholarship Council.
Publisher Copyright:
© 2016 IEEE.
PY - 2021/9
Y1 - 2021/9
N2 - Feature learning with deep neural networks (DNNs) has made remarkable progress in recent years. However, its data-driven nature makes the collection of labeled training data expensive or impossible when the testing domain changes. Here, we propose a method of transferable feature learning and instance-level adaptation to improve the generalization ability of DNNs so as to mitigate the domain shift challenge for cross-domain visual recognition. When less labeled information is available, our proposed method shows attractive results in the new target domain and outperforms the typical fine-tuning method. Two DNNs are chosen as the representatives working with our proposed method, to do a comprehensive study about the generalization ability on the tasks of image-to-image transfer, image-to-video transfer, multidomain image classification, and weakly supervised detection. The experimental results show that our proposed method is superior to other existing works in the literature. In addition, a large scale of cross-domain database is merged from three different domains, providing a quantitative platform to evaluate different approaches in the field of cross-domain object detection.
AB - Feature learning with deep neural networks (DNNs) has made remarkable progress in recent years. However, its data-driven nature makes the collection of labeled training data expensive or impossible when the testing domain changes. Here, we propose a method of transferable feature learning and instance-level adaptation to improve the generalization ability of DNNs so as to mitigate the domain shift challenge for cross-domain visual recognition. When less labeled information is available, our proposed method shows attractive results in the new target domain and outperforms the typical fine-tuning method. Two DNNs are chosen as the representatives working with our proposed method, to do a comprehensive study about the generalization ability on the tasks of image-to-image transfer, image-to-video transfer, multidomain image classification, and weakly supervised detection. The experimental results show that our proposed method is superior to other existing works in the literature. In addition, a large scale of cross-domain database is merged from three different domains, providing a quantitative platform to evaluate different approaches in the field of cross-domain object detection.
KW - Cross-domain
KW - deep neural networks (DNNs)
KW - domain adaptation
KW - generalization ability
KW - visual recognition
UR - http://www.scopus.com/inward/record.url?scp=85078022024&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85078022024&partnerID=8YFLogxK
U2 - 10.1109/TCDS.2020.2965166
DO - 10.1109/TCDS.2020.2965166
M3 - Article
AN - SCOPUS:85078022024
SN - 2379-8920
VL - 13
SP - 607
EP - 620
JO - IEEE Transactions on Cognitive and Developmental Systems
JF - IEEE Transactions on Cognitive and Developmental Systems
IS - 3
M1 - 8954791
ER -