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 language | English |
|---|---|
| Title of host publication | Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings |
| Publisher | Springer Verlag |
| Pages | 467-483 |
| Number of pages | 17 |
| ISBN (Electronic) | 9783030012250 |
| ISBN (Print) | 9783030012243 |
| DOIs | |
| Publication status | Published - 6 Oct 2018 |
| Event | 15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany Duration: 8 Sep 2018 → 14 Sep 2018 |
Publication series
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Volume | 11208 LNCS |
| ISSN (Print) | 0302-9743 |
| ISSN (Electronic) | 1611-3349 |
Conference
| Conference | 15th European Conference on Computer Vision, ECCV 2018 |
|---|---|
| Country | Germany |
| City | Munich |
| Period | 8/09/18 → 14/09/18 |
Fingerprint
Keywords
- Deep domain adaptation
- Optimal transport
Cite this
}
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 proceeding › Conference 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 -