Cargando…

Saliency guided data augmentation strategy for maximally utilizing an object’s visual information

Among the various types of data augmentation strategies, the mixup-based approach has been particularly studied. However, in existing mixup-based approaches, object loss and label mismatching can occur if random patches are utilized when constructing augmented images, and additionally, patches that...

Descripción completa

Detalles Bibliográficos
Autores principales: An, Junhyeok, Jang, Soojin, Kwon, Junehyoung, Jin, Kyohoon, Kim, YoungBin
Formato: Online Artículo Texto
Lenguaje:English
Publicado: Public Library of Science 2022
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9560504/
https://www.ncbi.nlm.nih.gov/pubmed/36227912
http://dx.doi.org/10.1371/journal.pone.0274767
_version_ 1784807762194071552
author An, Junhyeok
Jang, Soojin
Kwon, Junehyoung
Jin, Kyohoon
Kim, YoungBin
author_facet An, Junhyeok
Jang, Soojin
Kwon, Junehyoung
Jin, Kyohoon
Kim, YoungBin
author_sort An, Junhyeok
collection PubMed
description Among the various types of data augmentation strategies, the mixup-based approach has been particularly studied. However, in existing mixup-based approaches, object loss and label mismatching can occur if random patches are utilized when constructing augmented images, and additionally, patches that do not contain objects might be included, which degrades performance. In this paper, we propose a novel augmentation method that mixes patches in a non-overlapping manner after they are extracted from the salient regions in an image. The suggested method can make effective use of object characteristics, because the constructed image consists only of visually important regions and is robust to noise. Since the patches do not occlude each other, the semantically meaningful information in the salient regions can be fully utilized. Additionally, our method is more robust to adversarial attack than the conventional augmentation method. In the experimental results, when Wide ResNet was trained on the public datasets, CIFAR-10, CIFAR-100 and STL-10, the top-1 accuracy was 97.26%, 83.99% and 82.40% respectively, which surpasses other augmentation methods.
format Online
Article
Text
id pubmed-9560504
institution National Center for Biotechnology Information
language English
publishDate 2022
publisher Public Library of Science
record_format MEDLINE/PubMed
spelling pubmed-95605042022-10-14 Saliency guided data augmentation strategy for maximally utilizing an object’s visual information An, Junhyeok Jang, Soojin Kwon, Junehyoung Jin, Kyohoon Kim, YoungBin PLoS One Research Article Among the various types of data augmentation strategies, the mixup-based approach has been particularly studied. However, in existing mixup-based approaches, object loss and label mismatching can occur if random patches are utilized when constructing augmented images, and additionally, patches that do not contain objects might be included, which degrades performance. In this paper, we propose a novel augmentation method that mixes patches in a non-overlapping manner after they are extracted from the salient regions in an image. The suggested method can make effective use of object characteristics, because the constructed image consists only of visually important regions and is robust to noise. Since the patches do not occlude each other, the semantically meaningful information in the salient regions can be fully utilized. Additionally, our method is more robust to adversarial attack than the conventional augmentation method. In the experimental results, when Wide ResNet was trained on the public datasets, CIFAR-10, CIFAR-100 and STL-10, the top-1 accuracy was 97.26%, 83.99% and 82.40% respectively, which surpasses other augmentation methods. Public Library of Science 2022-10-13 /pmc/articles/PMC9560504/ /pubmed/36227912 http://dx.doi.org/10.1371/journal.pone.0274767 Text en © 2022 An et al https://creativecommons.org/licenses/by/4.0/This is an open access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/) , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
spellingShingle Research Article
An, Junhyeok
Jang, Soojin
Kwon, Junehyoung
Jin, Kyohoon
Kim, YoungBin
Saliency guided data augmentation strategy for maximally utilizing an object’s visual information
title Saliency guided data augmentation strategy for maximally utilizing an object’s visual information
title_full Saliency guided data augmentation strategy for maximally utilizing an object’s visual information
title_fullStr Saliency guided data augmentation strategy for maximally utilizing an object’s visual information
title_full_unstemmed Saliency guided data augmentation strategy for maximally utilizing an object’s visual information
title_short Saliency guided data augmentation strategy for maximally utilizing an object’s visual information
title_sort saliency guided data augmentation strategy for maximally utilizing an object’s visual information
topic Research Article
url https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9560504/
https://www.ncbi.nlm.nih.gov/pubmed/36227912
http://dx.doi.org/10.1371/journal.pone.0274767
work_keys_str_mv AT anjunhyeok saliencyguideddataaugmentationstrategyformaximallyutilizinganobjectsvisualinformation
AT jangsoojin saliencyguideddataaugmentationstrategyformaximallyutilizinganobjectsvisualinformation
AT kwonjunehyoung saliencyguideddataaugmentationstrategyformaximallyutilizinganobjectsvisualinformation
AT jinkyohoon saliencyguideddataaugmentationstrategyformaximallyutilizinganobjectsvisualinformation
AT kimyoungbin saliencyguideddataaugmentationstrategyformaximallyutilizinganobjectsvisualinformation