• español
  • English
  • français
  • Deutsch
  • português (Brasil)
  • italiano
    • español
    • English
    • français
    • Deutsch
    • português (Brasil)
    • italiano
    • español
    • English
    • français
    • Deutsch
    • português (Brasil)
    • italiano
    JavaScript is disabled for your browser. Some features of this site may not work without it.

    Parcourir

    Tout UVaDOCCommunautésPar date de publicationAuteursSujetsTitres

    Mon compte

    Ouvrir une session

    Statistiques

    Statistiques d'usage de visualisation

    Compartir

    Voir le document 
    •   Accueil de UVaDOC
    • PUBLICATIONS SCIENTIFIQUES
    • Institutos de Investigación
    • Instituto de las Tecnologías Avanzadas en la Producción (ITAP)
    • ITAP - Artículos de revista
    • Voir le document
    •   Accueil de UVaDOC
    • PUBLICATIONS SCIENTIFIQUES
    • Institutos de Investigación
    • Instituto de las Tecnologías Avanzadas en la Producción (ITAP)
    • ITAP - Artículos de revista
    • Voir le document
    • español
    • English
    • français
    • Deutsch
    • português (Brasil)
    • italiano

    Exportar

    RISMendeleyRefworksZotero
    • edm
    • marc
    • xoai
    • qdc
    • ore
    • ese
    • dim
    • uketd_dc
    • oai_dc
    • etdms
    • rdf
    • mods
    • mets
    • didl
    • premis

    Citas

    Por favor, use este identificador para citar o enlazar este ítem:https://uvadoc.uva.es/handle/10324/73711

    Título
    Training Fully Convolutional Neural Networks for Lightweight, Non-Critical Instance Segmentation Applications
    Autor
    Veganzones, Miguel
    Cisnal De La Rica, AnaAutoridad UVA Orcid
    Fuente López, Eusebio de laAutoridad UVA Orcid
    Fraile Marinero, Juan CarlosAutoridad UVA Orcid
    Año del Documento
    2024-12-05
    Editorial
    MDPI
    Documento Fuente
    Applied Science, December, vol. 14, n. 23, p. 11357
    Résumé
    Augmented reality applications involving human interaction with virtual objects often rely on segmentation-based hand detection techniques. Semantic segmentation can then be enhanced with instance-specific information to model complex interactions between objects, but extracting such information typically increases the computational load significantly. This study proposes a training strategy that enables conventional semantic segmentation networks to preserve some instance information during inference. This is accomplished by introducing pixel weight maps into the loss calculation, increasing the importance of boundary pixels between instances. We compare two common fully convolutional network (FCN) architectures, U-Net and ResNet, and fine-tune the fittest to improve segmentation results. Although the resulting model does not reach state-of-the-art segmentation performance on the EgoHands dataset, it preserves some instance information with no computational overhead. As expected, degraded segmentations are a necessary trade-off to preserve boundaries when instances are close together. This strategy allows approximating instance segmentation in real-time using non-specialized hardware, obtaining a unique blob for an instance with an intersection over union greater than 50% in 79% of the instances in our test set. A simple FCN, typically used for semantic segmentation, has shown promising instance segmentation results by introducing per-pixel weight maps during training for light-weight applications.
    Palabras Clave
    computer vision
    convolutional neural networks
    deep learning
    hand segmentation
    semantic segmentation
    ISSN
    2076-3417
    Revisión por pares
    SI
    DOI
    10.3390/app142311357
    Version del Editor
    https://www.mdpi.com/2076-3417/14/23/11357
    Idioma
    spa
    URI
    https://uvadoc.uva.es/handle/10324/73711
    Tipo de versión
    info:eu-repo/semantics/publishedVersion
    Derechos
    openAccess
    Aparece en las colecciones
    • ITAP - Artículos de revista [53]
    Afficher la notice complète
    Fichier(s) constituant ce document
    Nombre:
    applsci-14-11357.pdf
    Tamaño:
    6.652Mo
    Formato:
    Adobe PDF
    Descripción:
    articulo
    Thumbnail
    Voir/Ouvrir
    Attribution-NonCommercial-NoDerivatives 4.0 InternacionalExcepté là où spécifié autrement, la license de ce document est décrite en tant que Attribution-NonCommercial-NoDerivatives 4.0 Internacional

    Universidad de Valladolid

    Powered by MIT's. DSpace software, Version 5.10