Centre de documentation
scientifique

Domain adaptation for urban scene segmentation / Antoine Saporta (2022)  

Public Titre : Domain adaptation for urban scene segmentation Type de document : Thèse/HDR Auteurs : Antoine Saporta, Auteur ; Matthieu Cord, Directeur de thèse Editeur : Paris : Sorbonne Université Année de publication : 2022 Importance : 147 p. Format : 21 x 30 cm Note générale : bibliographie Thèse de Doctorat de Sorbonne Université, spécialité Informatique Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Traitement d'image optique [Termes IGN] apprentissage profond [Termes IGN] classification non dirigée [Termes IGN] entropie [Termes IGN] Mapillary [Termes IGN] navigation autonome [Termes IGN] réseau antagoniste génératif [Termes IGN] scène urbaine [Termes IGN] segmentation sémantique [Termes IGN] vision par ordinateur Index. décimale : THESE Thèses et HDR Résumé : (auteur) This thesis tackles some of the scientific locks of perception systems based on neural networks for autonomous vehicles. This dissertation discusses domain adaptation, a class of tools aiming at minimizing the need for labeled data. Domain adaptation allows generalization to so-called target data that share structures with the labeled so-called source data allowing supervision but nevertheless following a different statistical distribution. First, we study the introduction of privileged information in the source data, for instance, depth labels. The proposed strategy, BerMuDA, bases its domain adaptation on a multimodal representation obtained by bilinear fusion, modeling complex interactions between segmentation and depth. Next, we examine self-supervised learning strategies in domain adaptation, relying on selecting predictions on the unlabeled target data, serving as pseudo-labels. We propose two new selection criteria: first, an entropic criterion with ESL; then, with ConDA, using an estimate of the true class probability. Finally, the extension of adaptation scenarios to several target domains as well as in a continual learning framework is proposed. Two approaches are presented to extend traditional adversarial methods to multi-target domain adaptation: Multi-Dis. and MTKT. In a continual learning setting for which the target domains are discovered sequentially and without rehearsal, the proposed CTKT approach adapts MTKT to this new problem to tackle catastrophic forgetting. Note de contenu : 1- Introduction 2- Unsupervised domain adaptation 3- Leveraging priviledge information for unsupervised domain adaptation 4- Estimating and exploiting confident pseudo-labels for self-training 5- Adaptation to multiple domains 6- Conclusion Numéro de notice : 24079 Affiliation des auteurs : non IGN Thématique : IMAGERIE/INFORMATIQUE Nature : Thèse française Note de thèse : Thèse de Doctorat : Informatique : Sorbonne Université : 2022 Organisme de stage : Institut des Systèmes Intelligents et de Robotique DOI : sans En ligne : https://theses.hal.science/tel-03886201 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=102213