Overlap Training to Mitigate Inconsistencies Caused by Image Tiling in CNNs

DC FieldValueLanguage
dc.contributor.authorAn, Yu-
dc.contributor.authorYe, Qing-
dc.contributor.authorGuo, Jiulin-
dc.contributor.authorDong, Ruihai-
dc.date.accessioned2021-11-03T09:07:46Z-
dc.date.available2021-11-03T09:07:46Z-
dc.date.copyright2020 Springeren_US
dc.date.issued2020-12-08-
dc.identifier.isbn978-3-030-63798-9-
dc.identifier.urihttp://hdl.handle.net/10197/12583-
dc.descriptionThe 40th SGAI International Conference on Artificial Intelligence (AI-2020), Cambridge, United Kingdom (held online due to Coronavirus outbreak), 15–17 December 2020en_US
dc.description.abstractThis paper focuses on the problem of inconsistent predictions of modern convolutional neural networks (CNN) at patch (i.e. sub-image) boundaries. Limited by the graphics processing unit (GPU) resources, image tiling and stitching countermeasure have been applied for most megapixel images, that is, cutting images into overlapping tiles as CNN input, and then stitching CNN outputs together. However, we found that stitched (i.e. recovered) predictions have discontinuous grid-like noise. We propose a simple yet efficient overlap training framework to mitigate the inconsistent prediction at patch boundaries without changing the model architecture while improving the stability, robustness of the model. We have applied our solution to various CNNs (such as U-Net, DeepLab, RCF) and tested them on two real-world datasets. Extensive experiments suggest that the new framework is sufficient in reducing inconsistency and outperform these countermeasures. The source code and coloured figures are made publicly available online at: https://github.com/anyuzoey/Overlap-Training.git.en_US
dc.description.sponsorshipScience Foundation Irelanden_US
dc.language.isoenen_US
dc.publisherSpringeren_US
dc.relation.ispartofBramer, M., Ellis, R. (eds.). Artificial Intelligence XXXVII: 40th SGAI International Conference on Artificial Intelligence, AI 2020, Cambridge, UK, December 15–17, 2020, Proceedings. SGAI 2020en_US
dc.relation.ispartofseriesLecture Notes in Computer Scienceen_US
dc.relation.ispartofseries12498en_US
dc.relation.ispartofseriesLecture Notes in Artificial Intelligenceen_US
dc.relation.ispartofseries12498en_US
dc.rightsThe final publication is available at www.springerlink.com.en_US
dc.subjectConvolutional neural networksen_US
dc.subjectComputer visionen_US
dc.subjectImage segmentationen_US
dc.subjectFault recognitionen_US
dc.titleOverlap Training to Mitigate Inconsistencies Caused by Image Tiling in CNNsen_US
dc.typeConference Publicationen_US
dc.internal.webversionshttp://bcs-sgai.org/ai2020/-
dc.statusPeer revieweden_US
dc.identifier.doi10.1007/978-3-030-63799-6_3-
dc.neeo.contributorAn|Yu|aut|-
dc.neeo.contributorYe|Qing|aut|-
dc.neeo.contributorGuo|Jiulin|aut|-
dc.neeo.contributorDong|Ruihai|aut|-
dc.date.embargo2021-12-08en_US
dc.description.othersponsorshipInsight Research Centreen_US
dc.description.admin2021-02-24 JG: broken PDF replaceden_US
dc.date.updated2020-09-21T14:03:37Z-
dc.identifier.grantidSFI/12/RC/2289 P2-
dc.rights.licensehttps://creativecommons.org/licenses/by-nc-nd/3.0/ie/en_US
item.fulltextWith Fulltext-
item.grantfulltextopen-
Appears in Collections:Insight Research Collection
Files in This Item:
 File SizeFormat
DownloadOverlap Training to Mitigate Inconsistencies Caused by Image Tiling in CNNs_.pdf1.08 MBAdobe PDF
Show simple item record

Page view(s)

89
Last Week
4
Last month
20
checked on Jan 20, 2022

Download(s)

11
checked on Jan 20, 2022

Google ScholarTM

Check

Altmetric


If you are a publisher or author and have copyright concerns for any item, please email research.repository@ucd.ie and the item will be withdrawn immediately. The author or person responsible for depositing the article will be contacted within one business day.