Unsupervised Visual Domain Adaptation Using Subspace Alignment

Basura Fernando, Amaury Habrard, Marc Sebban and Tinne Tuytelaars

*KU Leuven ESAT-PSI iMinds Belgium

**UMR CNRS 5516 Laboratoire Hubert Curien University of Saint-Etienne France

Abstract

In this paper, we introduce a new domain adaptation (DA) algorithm where the source and target domains are represented by subspaces described by eigenvectors. In this context, our method seeks a domain adaptation solution by learning a mapping function which aligns the source subspace with the target one. We show that the solution of the corresponding optimization problem can be obtained in a simple closed form, leading to an extremely fast algorithm. We use a theoretical result to tune the unique hyperparameter corresponding to the size of the subspaces. We run our method on various datasets and show that, despite its intrinsic simplicity, it outperforms state of the art DA methods.

PDFs
PDF
Supplementary material

PDFs

PDF

Supplementary material

Source Code	size
DA_SA.tar Subspace Alignment Code with a demo. (Last updated 2014 September 23)	11.3KB
getGFKDim.m Subspace disagreement measure code	2.9KB
MLE_IntrinsicDim.m SA-MLE subspace estimation code. (Last updated 2014 September 23)	2.9KB
JCSL.m Joint cross-domain classification code.	3.6KB

Bibtex
@inproceedings{fernando2013SA, title={Unsupervised Visual Domain Adaptation Using Subspace Alignment}, author={Fernando, Basura and Habrard, Amaury and Sebban, Marc and Tuytelaars, Tinne}, booktitle={ICCV}, year={2013} }

Bibtex

@inproceedings{fernando2013SA, title={Unsupervised Visual Domain Adaptation Using Subspace Alignment}, author={Fernando, Basura and Habrard, Amaury and Sebban, Marc and Tuytelaars, Tinne}, booktitle={ICCV}, year={2013} }

Acknowledgements
We would like to thank Dr. Karl Markus Moll for his valuable discussions and feedback during this work.

Acknowledgements

We would like to thank Dr. Karl Markus Moll for his valuable discussions and feedback during this work.

Related Publications
	Subspace Alignment For Domain Adaptation Basura Fernando, Amaury Habrard, Marc Sebban and Tinne Tuytelaars arXiv.org PDF
	Unsupervised Visual Domain Adaptation Using Subspace Alignment Basura Fernando, Amaury Habrard, Marc Sebban and Tinne Tuytelaars ICCV 2013 PDF Suppl. Material
	Location Recognition Over Large Time Lags Basura Fernando, Tatiana Tommasi and Tinne Tuytelaars Computer Vision and Image Understanding arxiv PDF Project Bibtex @article{Fernando2015CVIU, title = "Location recognition over large time lags ", journal = "Computer Vision and Image Understanding ", volume = "139", number = "", pages = "21-28", year = "2015", note = "", issn = "1077-3142", doi = "http://dx.doi.org/10.1016/j.cviu.2015.05.016", url = "http://www.sciencedirect.com/science/article/pii/S107731421500137X", author = "Basura Fernando and Tatiana Tommasi and Tinne Tuytelaars", }
	Joint cross-domain classification and subspace learning for unsupervised adaptation Basura Fernando, Tatiana Tommasi and Tinne Tuytelaars Pattern Recognition Letters PDF Code Bibtex @article{Fernando2015PRL, title = "Joint cross-domain classification and subspace learning for unsupervised adaptation", journal = "Pattern Recognition Letters", volume = "", number = "", pages = " - ", year = "2015", note = "", issn = "", doi = " 10.1016/j.patrec.2015.07.009", url = "", author = "Basura Fernando and Tatiana Tommasi and Tinne Tuytelaars", }
	Does Evolution cause a Domain Shift? Konstantinos Rematas, Basura Fernando, Tatiana Tommasi and Tinne Tuytelaars International Workshop on Visual Domain Adaptation and Dataset Bias - ICCV 2013 PDF Project