Overlap Training to Mitigate Inconsistencies Caused by Image Tiling in CNNs

Files in This Item:
Access to this item has been restricted by the copyright holder until:2021-12-08
 File SizeFormat
    Request a copyOverlap Training to Mitigate Inconsistencies Caused by Image Tiling in CNNs_.pdf1.08 MBAdobe PDF
Title: Overlap Training to Mitigate Inconsistencies Caused by Image Tiling in CNNs
Authors: An, YuYe, QingGuo, JiulinDong, Ruihai
Permanent link: http://hdl.handle.net/10197/12583
Date: 8-Dec-2020
Online since: 2021-11-03T09:07:46Z
Abstract: This 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.
Funding Details: Science Foundation Ireland
Funding Details: Insight Research Centre
Type of material: Conference Publication
Publisher: Springer
Series/Report no.: Lecture Notes in Computer Science; 12498; Lecture Notes in Artificial Intelligence; 12498
Copyright (published version): 2020 Springer
Keywords: Convolutional neural networksComputer visionImage segmentationFault recognition
DOI: 10.1007/978-3-030-63799-6_3
Other versions: http://bcs-sgai.org/ai2020/
Language: en
Status of Item: Peer reviewed
Is part of: Bramer, M., Ellis, R. (eds.). Artificial Intelligence XXXVII: 40th SGAI International Conference on Artificial Intelligence, AI 2020, Cambridge, UK, December 15–17, 2020, Proceedings. SGAI 2020
Conference Details: The 40th SGAI International Conference on Artificial Intelligence (AI-2020), Cambridge, United Kingdom (held online due to Coronavirus outbreak), 15–17 December 2020
ISBN: 978-3-030-63798-9
This item is made available under a Creative Commons License: https://creativecommons.org/licenses/by-nc-nd/3.0/ie/
Appears in Collections:Insight Research Collection

Show full item record

Page view(s)

60
Last Week
2
Last month
checked on Nov 30, 2021

Download(s)

7
checked on Nov 30, 2021

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.