DSpace Repository

Distribution-preserving data augmentation

Show simple item record

dc.contributor.author Saran, Nurdan Ayşe
dc.contributor.author Saran, Murat
dc.contributor.author Nar, Fatih
dc.date.accessioned 2022-04-06T11:21:25Z
dc.date.available 2022-04-06T11:21:25Z
dc.date.issued 2021-05-27
dc.identifier.citation Saran, Nurdan Ayşe; Saran, Murat; Nar, Fatih (2021). "Distribution-preserving data augmentation", Peerj Computer Science. tr_TR
dc.identifier.issn 2376-5992
dc.identifier.uri http://hdl.handle.net/20.500.12416/5286
dc.description.abstract In the last decade, deep learning has been applied in a wide range of problems with tremendous success. This success mainly comes from large data availability, increased computational power, and theoretical improvements in the training phase. As the dataset grows, the real world is better represented, making it possible to develop a model that can generalize. However, creating a labeled dataset is expensive, time-consuming, and sometimes not likely in some domains if not challenging. Therefore, researchers proposed data augmentation methods to increase dataset size and variety by creating variations of the existing data. For image data, variations can be obtained by applying color or spatial transformations, only one or a combination. Such color transformations perform some linear or nonlinear operations in the entire image or in the patches to create variations of the original image. The current color-based augmentation methods are usually based on image processing methods that apply color transformations such as equalizing, solarizing, and posterizing. Nevertheless, these color-based data augmentation methods do not guarantee to create plausible variations of the image. This paper proposes a novel distribution-preserving data augmentation method that creates plausible image variations by shifting pixel colors to another point in the image color distribution. We achieved this by defining a regularized density decreasing direction to create paths from the original pixels' color to the distribution tails. The proposed method provides superior performance compared to existing data augmentation methods which is shown using a transfer learning scenario on the UC Merced Land-use, Intel Image Classification, and Oxford-IIIT Pet datasets for classification and segmentation tasks. tr_TR
dc.language.iso eng tr_TR
dc.relation.isversionof 10.7717/peerj-cs.571 tr_TR
dc.rights info:eu-repo/semantics/openAccess tr_TR
dc.subject Machine Learning tr_TR
dc.subject Deep Learning tr_TR
dc.subject Data Augmentation tr_TR
dc.subject Color-Based Augmentation tr_TR
dc.subject Transfer Learning tr_TR
dc.title Distribution-preserving data augmentation tr_TR
dc.type article tr_TR
dc.relation.journal Peerj Computer Science tr_TR
dc.contributor.authorID 20868 tr_TR
dc.contributor.authorID 17753 tr_TR
dc.contributor.department Çankaya Üniversitesi, Mühendislik Fakültesi, Bilgisayar Mühendisliği Bölümü tr_TR


Files in this item

This item appears in the following Collection(s)

Show simple item record