DeepJDOT

Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation

Bharath Bhushan Damodaran*, Benjamin Kellenberger, Rémi Flamary, Devis Tuia, Nicolas Courty

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference paper

3 Citations (Scopus)

Abstract

In computer vision, one is often confronted with problems of domain shifts, which occur when one applies a classifier trained on a source dataset to target data sharing similar characteristics (e.g. same classes), but also different latent data structures (e.g. different acquisition conditions). In such a situation, the model will perform poorly on the new data, since the classifier is specialized to recognize visual cues specific to the source domain. In this work we explore a solution, named DeepJDOT, to tackle this problem: through a measure of discrepancy on joint deep representations/labels based on optimal transport, we not only learn new data representations aligned between the source and target domain, but also simultaneously preserve the discriminative information used by the classifier. We applied DeepJDOT to a series of visual recognition tasks, where it compares favorably against state-of-the-art deep domain adaptation methods.

Original languageEnglish
Title of host publicationComputer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
PublisherSpringer Verlag
Pages467-483
Number of pages17
ISBN (Electronic)9783030012250
ISBN (Print)9783030012243
DOIs
Publication statusPublished - 6 Oct 2018
Event15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany
Duration: 8 Sep 201814 Sep 2018

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11208 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th European Conference on Computer Vision, ECCV 2018
CountryGermany
CityMunich
Period8/09/1814/09/18

Fingerprint

Optimal Transport
Joint Distribution
Classifiers
Classifier
Computer vision
Target
Data Sharing
Data structures
Labels
Computer Vision
Discrepancy
Data Structures
Series
Vision

Keywords

  • Deep domain adaptation
  • Optimal transport

Cite this

Damodaran, B. B., Kellenberger, B., Flamary, R., Tuia, D., & Courty, N. (2018). DeepJDOT: Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation. In Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings (pp. 467-483). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11208 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-01225-0_28
Damodaran, Bharath Bhushan ; Kellenberger, Benjamin ; Flamary, Rémi ; Tuia, Devis ; Courty, Nicolas. / DeepJDOT : Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation. Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Springer Verlag, 2018. pp. 467-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{08795adee86d4cc999b57dbba049f018,
title = "DeepJDOT: Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation",
abstract = "In computer vision, one is often confronted with problems of domain shifts, which occur when one applies a classifier trained on a source dataset to target data sharing similar characteristics (e.g. same classes), but also different latent data structures (e.g. different acquisition conditions). In such a situation, the model will perform poorly on the new data, since the classifier is specialized to recognize visual cues specific to the source domain. In this work we explore a solution, named DeepJDOT, to tackle this problem: through a measure of discrepancy on joint deep representations/labels based on optimal transport, we not only learn new data representations aligned between the source and target domain, but also simultaneously preserve the discriminative information used by the classifier. We applied DeepJDOT to a series of visual recognition tasks, where it compares favorably against state-of-the-art deep domain adaptation methods.",
keywords = "Deep domain adaptation, Optimal transport",
author = "Damodaran, {Bharath Bhushan} and Benjamin Kellenberger and R{\'e}mi Flamary and Devis Tuia and Nicolas Courty",
year = "2018",
month = "10",
day = "6",
doi = "10.1007/978-3-030-01225-0_28",
language = "English",
isbn = "9783030012243",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "467--483",
booktitle = "Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings",

}

Damodaran, BB, Kellenberger, B, Flamary, R, Tuia, D & Courty, N 2018, DeepJDOT: Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation. in Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11208 LNCS, Springer Verlag, pp. 467-483, 15th European Conference on Computer Vision, ECCV 2018, Munich, Germany, 8/09/18. https://doi.org/10.1007/978-3-030-01225-0_28

DeepJDOT : Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation. / Damodaran, Bharath Bhushan; Kellenberger, Benjamin; Flamary, Rémi; Tuia, Devis; Courty, Nicolas.

Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Springer Verlag, 2018. p. 467-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11208 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference paper

TY - GEN

T1 - DeepJDOT

T2 - Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation

AU - Damodaran, Bharath Bhushan

AU - Kellenberger, Benjamin

AU - Flamary, Rémi

AU - Tuia, Devis

AU - Courty, Nicolas

PY - 2018/10/6

Y1 - 2018/10/6

N2 - In computer vision, one is often confronted with problems of domain shifts, which occur when one applies a classifier trained on a source dataset to target data sharing similar characteristics (e.g. same classes), but also different latent data structures (e.g. different acquisition conditions). In such a situation, the model will perform poorly on the new data, since the classifier is specialized to recognize visual cues specific to the source domain. In this work we explore a solution, named DeepJDOT, to tackle this problem: through a measure of discrepancy on joint deep representations/labels based on optimal transport, we not only learn new data representations aligned between the source and target domain, but also simultaneously preserve the discriminative information used by the classifier. We applied DeepJDOT to a series of visual recognition tasks, where it compares favorably against state-of-the-art deep domain adaptation methods.

AB - In computer vision, one is often confronted with problems of domain shifts, which occur when one applies a classifier trained on a source dataset to target data sharing similar characteristics (e.g. same classes), but also different latent data structures (e.g. different acquisition conditions). In such a situation, the model will perform poorly on the new data, since the classifier is specialized to recognize visual cues specific to the source domain. In this work we explore a solution, named DeepJDOT, to tackle this problem: through a measure of discrepancy on joint deep representations/labels based on optimal transport, we not only learn new data representations aligned between the source and target domain, but also simultaneously preserve the discriminative information used by the classifier. We applied DeepJDOT to a series of visual recognition tasks, where it compares favorably against state-of-the-art deep domain adaptation methods.

KW - Deep domain adaptation

KW - Optimal transport

U2 - 10.1007/978-3-030-01225-0_28

DO - 10.1007/978-3-030-01225-0_28

M3 - Conference paper

SN - 9783030012243

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 467

EP - 483

BT - Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings

PB - Springer Verlag

ER -

Damodaran BB, Kellenberger B, Flamary R, Tuia D, Courty N. DeepJDOT: Deep Joint Distribution Optimal Transport for Unsupervised Domain Adaptation. In Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Springer Verlag. 2018. p. 467-483. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-01225-0_28