TY - GEN
T1 - Domain adaptive fusion for adaptive image classification
AU - Dudley, Andrew
AU - Nagabandi, Bhadrinath
AU - Venkateswara, Hemanth
AU - Panchanathan, Sethuraman
N1 - Funding Information:
Foundation for their funding support. This material is partially based upon work supported by Adidas and by the National Science Foundation under Grant No. 1828010.
Publisher Copyright:
© Springer Nature Switzerland AG 2020.
PY - 2020
Y1 - 2020
N2 - Recent works in the development of deep adaptation networks have yielded progressive improvement on unsupervised domain adaptive classification tasks by reducing the distribution discrepancy between source and target domains. In parallel, the unification of dominant semi-supervised learning techniques has illustrated the unprecedented potential for utilizing unlabeled data to train a classification model in defiance of a discouragingly meager labeled dataset. In this paper, we propose Domain Adaptive Fusion (DAF), a novel domain adaptation algorithm that encourages a domain-invariant linear relationship between the pixel-space of different domains and the prediction-space while being trained under a domain adversarial signal. The thoughtful combination of key components in unsupervised domain adaptation and semi-supervised learning enable DAF to effectively bridge the gap between source and target domains. Experiments performed on computer vision benchmark datasets for domain adaptation endorse the efficacy of our hybrid approach, outperforming all of the baseline architectures on most of the transfer tasks.
AB - Recent works in the development of deep adaptation networks have yielded progressive improvement on unsupervised domain adaptive classification tasks by reducing the distribution discrepancy between source and target domains. In parallel, the unification of dominant semi-supervised learning techniques has illustrated the unprecedented potential for utilizing unlabeled data to train a classification model in defiance of a discouragingly meager labeled dataset. In this paper, we propose Domain Adaptive Fusion (DAF), a novel domain adaptation algorithm that encourages a domain-invariant linear relationship between the pixel-space of different domains and the prediction-space while being trained under a domain adversarial signal. The thoughtful combination of key components in unsupervised domain adaptation and semi-supervised learning enable DAF to effectively bridge the gap between source and target domains. Experiments performed on computer vision benchmark datasets for domain adaptation endorse the efficacy of our hybrid approach, outperforming all of the baseline architectures on most of the transfer tasks.
KW - Domain adaptation
KW - Domain-shift
KW - Entropy regularization
KW - Semi supervised learning
UR - http://www.scopus.com/inward/record.url?scp=85089615199&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85089615199&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-54407-2_30
DO - 10.1007/978-3-030-54407-2_30
M3 - Conference contribution
AN - SCOPUS:85089615199
SN - 9783030544065
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 357
EP - 371
BT - Smart Multimedia - 2nd International Conference, ICSM 2019, Revised Selected Papers
A2 - McDaniel, Troy
A2 - Berretti, Stefano
A2 - Curcio, Igor D.D.
A2 - Basu, Anup
PB - Springer
T2 - 2nd International Conference on Smart Multimedia, ICSM 2019
Y2 - 16 December 2019 through 18 December 2019
ER -