Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++

Semantic segmentation of airborne LiDAR point clouds enables a broad range of urban and environmental applications. However, domain shifts between training and operational data, as well asthefrequentemergenceofnewsemanticclasses,posesignificantchallengesfordeploying deep learning models effectively....

Descripción completa

Detalles Bibliográficos
Autores: Carós, Mariona, Just, Ariadna, Seguí Mesquida, Santi, Vitrià i Marca, Jordi
Tipo de recurso: artículo
Estado:Versión publicada
Fecha de publicación:2026
País:España
Institución:Universidad de Barcelona
Repositorio:Dipòsit Digital de la UB
OAI Identifier:oai:diposit.ub.edu:2445/227057
Acceso en línea:https://hdl.handle.net/2445/227057
Access Level:acceso abierto
Palabra clave:Visualització tridimensional
Visió per ordinador
Teledetecció
Three-dimensional display systems
Computer vision
Remote sensing
id ES_bb7e05973de9e5f5b71a29657dc1c43c
oai_identifier_str oai:diposit.ub.edu:2445/227057
network_acronym_str ES
network_name_str España
repository_id_str
spelling Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++Carós, MarionaJust, AriadnaSeguí Mesquida, SantiVitrià i Marca, JordiVisualització tridimensionalVisió per ordinadorTeledeteccióThree-dimensional display systemsComputer visionRemote sensingSemantic segmentation of airborne LiDAR point clouds enables a broad range of urban and environmental applications. However, domain shifts between training and operational data, as well asthefrequentemergenceofnewsemanticclasses,posesignificantchallengesfordeploying deep learning models effectively. In this work, we explore the integration of Low-Rank Adaptation (LoRA), a parameter-efficient fine-tuning technique, into the PointNet++ architecture to address these challenges. We evaluate LoRA in two realistic scenarios: domain adaptation and incremental learning with novel classes, using subsets of large-scale LiDAR datasets under constrained labeled data settings. Our experiments show that LoRA outperforms traditional full f ine-tuning, achieving notable gains (+3.1 IoU for specific classes and +0.3 mIoUonTerLiDAR, +2.7 mIoU on DALES), while exhibiting greater resistance to catastrophic forgetting and improved generalization, particularly for underrepresented classes. Furthermore, LoRA exceeds baseline accuracy with substantially fewer trainable parameters (73.4% reduction), highlighting its suitability for resource-constrained deployment scenarios. We also present TerLiDAR, a publicly available annotated airborne LiDAR dataset covering 51.4 km2 along the Ter River in Catalonia, Spain. It contributes to increasing the diversity of semantic segmentation benchmarks and advancing 3D scene understanding in remote sensing.Elsevier2026info:eu-repo/semantics/articleinfo:eu-repo/semantics/publishedVersionapplication/pdfhttps://hdl.handle.net/2445/227057Articles publicats en revistes (Matemàtiques i Informàtica)reponame:Dipòsit Digital de la UBinstname:Universidad de BarcelonaReproducció del document publicat a: https://doi.org/10.1016/j.ophoto.2026.100119ISPRS Journal of Photogrammetry and Remote Sensing, 2026, vol. 19https://doi.org/10.1016/j.ophoto.2026.100119cc-by-nc-nd (c) Mariona Carós, et al., 2026http://creativecommons.org/licenses/by-nc-nd/4.0/info:eu-repo/semantics/openAccessoai:diposit.ub.edu:2445/2270572026-05-27T06:46:51Z
dc.title.none.fl_str_mv Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
title Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
spellingShingle Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
Carós, Mariona
Visualització tridimensional
Visió per ordinador
Teledetecció
Three-dimensional display systems
Computer vision
Remote sensing
title_short Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
title_full Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
title_fullStr Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
title_full_unstemmed Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
title_sort Enhancing point cloud semantic segmentation via scalable domain adaptation with LoRA-enabled PointNet++
dc.creator.none.fl_str_mv Carós, Mariona
Just, Ariadna
Seguí Mesquida, Santi
Vitrià i Marca, Jordi
author Carós, Mariona
author_facet Carós, Mariona
Just, Ariadna
Seguí Mesquida, Santi
Vitrià i Marca, Jordi
author_role author
author2 Just, Ariadna
Seguí Mesquida, Santi
Vitrià i Marca, Jordi
author2_role author
author
author
dc.subject.none.fl_str_mv Visualització tridimensional
Visió per ordinador
Teledetecció
Three-dimensional display systems
Computer vision
Remote sensing
topic Visualització tridimensional
Visió per ordinador
Teledetecció
Three-dimensional display systems
Computer vision
Remote sensing
description Semantic segmentation of airborne LiDAR point clouds enables a broad range of urban and environmental applications. However, domain shifts between training and operational data, as well asthefrequentemergenceofnewsemanticclasses,posesignificantchallengesfordeploying deep learning models effectively. In this work, we explore the integration of Low-Rank Adaptation (LoRA), a parameter-efficient fine-tuning technique, into the PointNet++ architecture to address these challenges. We evaluate LoRA in two realistic scenarios: domain adaptation and incremental learning with novel classes, using subsets of large-scale LiDAR datasets under constrained labeled data settings. Our experiments show that LoRA outperforms traditional full f ine-tuning, achieving notable gains (+3.1 IoU for specific classes and +0.3 mIoUonTerLiDAR, +2.7 mIoU on DALES), while exhibiting greater resistance to catastrophic forgetting and improved generalization, particularly for underrepresented classes. Furthermore, LoRA exceeds baseline accuracy with substantially fewer trainable parameters (73.4% reduction), highlighting its suitability for resource-constrained deployment scenarios. We also present TerLiDAR, a publicly available annotated airborne LiDAR dataset covering 51.4 km2 along the Ter River in Catalonia, Spain. It contributes to increasing the diversity of semantic segmentation benchmarks and advancing 3D scene understanding in remote sensing.
publishDate 2026
dc.date.none.fl_str_mv 2026
dc.type.none.fl_str_mv info:eu-repo/semantics/article
info:eu-repo/semantics/publishedVersion
format article
status_str publishedVersion
dc.identifier.none.fl_str_mv https://hdl.handle.net/2445/227057
url https://hdl.handle.net/2445/227057
dc.language.none.fl_str_mv
language_invalid_str_mv
dc.relation.none.fl_str_mv Reproducció del document publicat a: https://doi.org/10.1016/j.ophoto.2026.100119
ISPRS Journal of Photogrammetry and Remote Sensing, 2026, vol. 19
https://doi.org/10.1016/j.ophoto.2026.100119
dc.rights.none.fl_str_mv cc-by-nc-nd (c) Mariona Carós, et al., 2026
http://creativecommons.org/licenses/by-nc-nd/4.0/
info:eu-repo/semantics/openAccess
rights_invalid_str_mv cc-by-nc-nd (c) Mariona Carós, et al., 2026
http://creativecommons.org/licenses/by-nc-nd/4.0/
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.publisher.none.fl_str_mv Elsevier
publisher.none.fl_str_mv Elsevier
dc.source.none.fl_str_mv Articles publicats en revistes (Matemàtiques i Informàtica)
reponame:Dipòsit Digital de la UB
instname:Universidad de Barcelona
instname_str Universidad de Barcelona
reponame_str Dipòsit Digital de la UB
collection Dipòsit Digital de la UB
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869418030708031488
score 15,812429

Similares en LA Referencia