Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation

Accurate training of deep neural networks for semantic segmentation requires a large number of pixel-level annotations of real images, which are expensive to generate or not even available. In this context, Unsupervised Domain Adaptation (UDA) can transfer knowledge from unlimited synthetic annotati...

Descripción completa

Detalles Bibliográficos
Autores: Alcover Couso, Roberto, San Miguel Avedillo, Juan Carlos, Escudero Viñolo, Marcos, Carballeira López, Pablo
Tipo de recurso: artículo
Fecha de publicación:2024
País:España
Institución:Universidad Autónoma de Madrid
Repositorio:Biblos-e Archivo. Repositorio Institucional de la UAM
Idioma:inglés
OAI Identifier:oai:repositorio.uam.es:10486/712091
Acceso en línea:http://hdl.handle.net/10486/712091
https://dx.doi.org/10.1007/s00371-024-03373-8
Access Level:acceso abierto
Palabra clave:Semantic Segmentation
Unsupervised Domain Adaptation
Curriculum Learning
Synthetic Data
Informática
id ES_620f7991fb3dee410ac0ab0c21de08f5
oai_identifier_str oai:repositorio.uam.es:10486/712091
network_acronym_str ES
network_name_str España
repository_id_str
spelling Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentationAlcover Couso, RobertoSan Miguel Avedillo, Juan CarlosEscudero Viñolo, MarcosCarballeira López, PabloSemantic SegmentationUnsupervised Domain AdaptationCurriculum LearningSynthetic DataInformáticaAccurate training of deep neural networks for semantic segmentation requires a large number of pixel-level annotations of real images, which are expensive to generate or not even available. In this context, Unsupervised Domain Adaptation (UDA) can transfer knowledge from unlimited synthetic annotations to unlabeled real images of a given domain. UDA methods are composed of an initial training stage with labeled synthetic data followed by a second stage for feature alignment between labeled synthetic and unlabeled real data. In this paper, we propose a novel approach for UDA focusing the initial training stage, which leads to increased performance after adaptation. We introduce a curriculum strategy where each semantic class is learned progressively. Thereby, better features are obtained for the second stage. This curriculum is based on: (1) a classscoring function to determine the difficulty of each semantic class, (2) a strategy for incremental learning based on scoring and pacing functions that limits the required training time unlike standard curriculum-based training and (3) a training loss to operate at class level. We extensively evaluate our approach as the first stage of several state-of-the-art UDA methods for semantic segmentation. Our results demonstrate significant performance enhancements across all methods: improvements of up to 10% for entropy-based techniques and 8% for adversarial methods. These findings underscore the dependency of UDA on the accuracy of the initial training. The implementation is available at https://github.com/vpulab/PCCLThis work has been partially supported by the Spanish Government through its TED2021-131643A-I00 (HVD) and the PID2021-125051OB-I00 (SEGA-CV) projectsSpringerDepartamento de Tecnología Electrónica y de las ComunicacionesEscuela Politécnica Superior20242024-04-15research articlehttp://purl.org/coar/resource_type/c_2df8fbb1VoRhttp://purl.org/coar/version/c_970fb48d4fbd8a85info:eu-repo/semantics/articleapplication/pdfhttp://hdl.handle.net/10486/712091https://dx.doi.org/10.1007/s00371-024-03373-8reponame:Biblos-e Archivo. Repositorio Institucional de la UAMinstname:Universidad Autónoma de MadridInglésengopen accesshttp://purl.org/coar/access_right/c_abf2info:eu-repo/semantics/openAccessoai:repositorio.uam.es:10486/7120912026-06-23T12:46:27Z
dc.title.none.fl_str_mv Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
title Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
spellingShingle Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
Alcover Couso, Roberto
Semantic Segmentation
Unsupervised Domain Adaptation
Curriculum Learning
Synthetic Data
Informática
title_short Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
title_full Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
title_fullStr Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
title_full_unstemmed Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
title_sort Per-class curriculum for Unsupervised Domain Adaptation in semantic segmentation
dc.creator.none.fl_str_mv Alcover Couso, Roberto
San Miguel Avedillo, Juan Carlos
Escudero Viñolo, Marcos
Carballeira López, Pablo
author Alcover Couso, Roberto
author_facet Alcover Couso, Roberto
San Miguel Avedillo, Juan Carlos
Escudero Viñolo, Marcos
Carballeira López, Pablo
author_role author
author2 San Miguel Avedillo, Juan Carlos
Escudero Viñolo, Marcos
Carballeira López, Pablo
author2_role author
author
author
dc.contributor.none.fl_str_mv Departamento de Tecnología Electrónica y de las Comunicaciones
Escuela Politécnica Superior
dc.subject.none.fl_str_mv Semantic Segmentation
Unsupervised Domain Adaptation
Curriculum Learning
Synthetic Data
Informática
topic Semantic Segmentation
Unsupervised Domain Adaptation
Curriculum Learning
Synthetic Data
Informática
description Accurate training of deep neural networks for semantic segmentation requires a large number of pixel-level annotations of real images, which are expensive to generate or not even available. In this context, Unsupervised Domain Adaptation (UDA) can transfer knowledge from unlimited synthetic annotations to unlabeled real images of a given domain. UDA methods are composed of an initial training stage with labeled synthetic data followed by a second stage for feature alignment between labeled synthetic and unlabeled real data. In this paper, we propose a novel approach for UDA focusing the initial training stage, which leads to increased performance after adaptation. We introduce a curriculum strategy where each semantic class is learned progressively. Thereby, better features are obtained for the second stage. This curriculum is based on: (1) a classscoring function to determine the difficulty of each semantic class, (2) a strategy for incremental learning based on scoring and pacing functions that limits the required training time unlike standard curriculum-based training and (3) a training loss to operate at class level. We extensively evaluate our approach as the first stage of several state-of-the-art UDA methods for semantic segmentation. Our results demonstrate significant performance enhancements across all methods: improvements of up to 10% for entropy-based techniques and 8% for adversarial methods. These findings underscore the dependency of UDA on the accuracy of the initial training. The implementation is available at https://github.com/vpulab/PCCL
publishDate 2024
dc.date.none.fl_str_mv 2024
2024-04-15
dc.type.none.fl_str_mv research article
http://purl.org/coar/resource_type/c_2df8fbb1
VoR
http://purl.org/coar/version/c_970fb48d4fbd8a85
dc.type.openaire.fl_str_mv info:eu-repo/semantics/article
format article
dc.identifier.none.fl_str_mv http://hdl.handle.net/10486/712091
https://dx.doi.org/10.1007/s00371-024-03373-8
url http://hdl.handle.net/10486/712091
https://dx.doi.org/10.1007/s00371-024-03373-8
dc.language.none.fl_str_mv Inglés
eng
language_invalid_str_mv Inglés
language eng
dc.rights.none.fl_str_mv open access
http://purl.org/coar/access_right/c_abf2
dc.rights.openaire.fl_str_mv info:eu-repo/semantics/openAccess
rights_invalid_str_mv open access
http://purl.org/coar/access_right/c_abf2
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.publisher.none.fl_str_mv Springer
publisher.none.fl_str_mv Springer
dc.source.none.fl_str_mv reponame:Biblos-e Archivo. Repositorio Institucional de la UAM
instname:Universidad Autónoma de Madrid
instname_str Universidad Autónoma de Madrid
reponame_str Biblos-e Archivo. Repositorio Institucional de la UAM
collection Biblos-e Archivo. Repositorio Institucional de la UAM
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869409467382104064
score 15,198674

Similares en LA Referencia