Articles producció científica> Enginyeria Informàtica i Matemàtiques

Monocular depth map estimation based on a multi-scale deep architecture and curvilinear saliency feature boosting

  • Identification data

    Identifier: imarina:9280528
    Authors:
    Abdulwahab, SaddamRashwan, Hatem AGarcia, Miguel AngelMasoumian, ArminPuig, Domenec
    Abstract:
    Estimating depth from a monocular camera is a must for many applications, including scene understanding and reconstruction, robot vision, and self-driving cars. However, generating depth maps from single RGB images is still a challenge as object shapes are to be inferred from intensity images strongly affected by viewpoint changes, texture content and light conditions. Therefore, most current solutions produce blurry approximations of low-resolution depth maps. We propose a novel depth map estimation technique based on an autoencoder network. This network is endowed with a multi-scale architecture and a multi-level depth estimator that preserve high-level information extracted from coarse feature maps as well as detailed local information present in fine feature maps. Curvilinear saliency, which is related to curvature estimation, is exploited as a loss function to boost the depth accuracy at object boundaries and raise the performance of the estimated high-resolution depth maps. We evaluate our model on the public NYU Depth v2 and Make3D datasets. The proposed model yields superior performance on both datasets compared to the state-of-the-art, achieving an accuracy of 86% and showing exceptional performance at the preservation of object boundaries and small 3D structures. The code of the proposed model is publicly available at https://github.com/SaddamAbdulrhman/MDACSFB.
  • Others:

    Author, as appears in the article.: Abdulwahab, Saddam; Rashwan, Hatem A; Garcia, Miguel Angel; Masoumian, Armin; Puig, Domenec
    Department: Enginyeria Informàtica i Matemàtiques
    URV's Author/s: Abdellatif Fatahallah Ibrahim Mahmoud, Hatem / Abdulwahab, Saddam Abdulrhman Hamed / GARCIA GARCIA, MIGUEL ANGEL / Masoumian, Armin / Puig Valls, Domènec Savi
    Keywords: Multi-scale networks Monocular depth map estimation Deep autoencoders Curvilinear saliency multi-scale networks deep autoencoders curvilinear saliency
    Abstract: Estimating depth from a monocular camera is a must for many applications, including scene understanding and reconstruction, robot vision, and self-driving cars. However, generating depth maps from single RGB images is still a challenge as object shapes are to be inferred from intensity images strongly affected by viewpoint changes, texture content and light conditions. Therefore, most current solutions produce blurry approximations of low-resolution depth maps. We propose a novel depth map estimation technique based on an autoencoder network. This network is endowed with a multi-scale architecture and a multi-level depth estimator that preserve high-level information extracted from coarse feature maps as well as detailed local information present in fine feature maps. Curvilinear saliency, which is related to curvature estimation, is exploited as a loss function to boost the depth accuracy at object boundaries and raise the performance of the estimated high-resolution depth maps. We evaluate our model on the public NYU Depth v2 and Make3D datasets. The proposed model yields superior performance on both datasets compared to the state-of-the-art, achieving an accuracy of 86% and showing exceptional performance at the preservation of object boundaries and small 3D structures. The code of the proposed model is publicly available at https://github.com/SaddamAbdulrhman/MDACSFB.
    Thematic Areas: Zootecnia / recursos pesqueiros Software Matemática / probabilidade e estatística Interdisciplinar Engenharias iv Engenharias iii Engenharias i Computer science, artificial intelligence Ciências biológicas ii Ciências biológicas i Ciências ambientais Ciências agrárias i Ciência da computação Biotecnología Artificial intelligence Administração pública e de empresas, ciências contábeis e turismo
    licence for use: https://creativecommons.org/licenses/by/3.0/es/
    Author's mail: saddam.abdulwahab@urv.cat miguelangel.garciag@urv.cat hatem.abdellatif@urv.cat armin.masoumian@estudiants.urv.cat armin.masoumian@estudiants.urv.cat saddam.abdulwahab@urv.cat domenec.puig@urv.cat
    Author identifier: 0000-0001-9972-2182 0000-0001-5421-1637 0000-0002-0562-4205
    Record's date: 2024-09-21
    Papper version: info:eu-repo/semantics/acceptedVersion
    Link to the original source: https://link.springer.com/article/10.1007/s00521-022-07663-x
    Licence document URL: https://repositori.urv.cat/ca/proteccio-de-dades/
    Papper original source: Neural Computing & Applications. 34 (19): 16423-16440
    APA: Abdulwahab, Saddam; Rashwan, Hatem A; Garcia, Miguel Angel; Masoumian, Armin; Puig, Domenec (2022). Monocular depth map estimation based on a multi-scale deep architecture and curvilinear saliency feature boosting. Neural Computing & Applications, 34(19), 16423-16440. DOI: 10.1007/s00521-022-07663-x
    Article's DOI: 10.1007/s00521-022-07663-x
    Entity: Universitat Rovira i Virgili
    Journal publication year: 2022
    Publication Type: Journal Publications
  • Keywords:

    Artificial Intelligence,Computer Science, Artificial Intelligence,Software
    Multi-scale networks
    Monocular depth map estimation
    Deep autoencoders
    Curvilinear saliency
    multi-scale networks
    deep autoencoders
    curvilinear saliency
    Zootecnia / recursos pesqueiros
    Software
    Matemática / probabilidade e estatística
    Interdisciplinar
    Engenharias iv
    Engenharias iii
    Engenharias i
    Computer science, artificial intelligence
    Ciências biológicas ii
    Ciências biológicas i
    Ciências ambientais
    Ciências agrárias i
    Ciência da computação
    Biotecnología
    Artificial intelligence
    Administração pública e de empresas, ciências contábeis e turismo
  • Documents:

  • Cerca a google

    Search to google scholar