Multitemporal classification without new labels: A solution with optimal transport

Abstract : —Re-using models trained on a specific image acquisition to classify landcover in another image is no easy task. Illumination effects, specific angular configurations, abrupt and simple seasonal changes make that the spectra observed, even though representing the same kind of surface, drift in a way that prevents a non-adapted model to perform well. In this paper we propose a relative normalization technique to perform domain adaptation, i.e. to make the data distribution in the images more similar before classification. We study optimal transport as a way to match the image-specific distributions and propose two regularization schemes, one supervised and one semi-supervised, to obtain more robust and semantic matchings. Code is available at Experiments a challenging triplet of WorldView2 images, comparing three neighborhoods of the city of Zurich at different time instants confirm the effectiveness of the proposed method, that can perform adaptation in these non-coregistered and very different urban case studies.
Type de document :
Communication dans un congrès
Multitemp, Jul 2015, Annecy, France. 2015, <10.1109/Multi-Temp.2015.7245773>
Liste complète des métadonnées
Contributeur : Nicolas Courty <>
Soumis le : mardi 12 janvier 2016 - 14:05:39
Dernière modification le : mercredi 2 août 2017 - 10:10:23
Document(s) archivé(s) le : jeudi 10 novembre 2016 - 23:51:06


Fichiers éditeurs autorisés sur une archive ouverte



Devis Tuia, Rémi Flamary, Alain Rakotomamonjy, Nicolas Courty. Multitemporal classification without new labels: A solution with optimal transport. Multitemp, Jul 2015, Annecy, France. 2015, <10.1109/Multi-Temp.2015.7245773>. <hal-01254329>



Consultations de
la notice


Téléchargements du document