KIT | KIT-Bibliothek | Impressum | Datenschutz

Improving generative adversarial networks for patch-based unpaired image-to-image translation

Böhland, Moritz 1; Bruch, Roman ORCID iD icon 1; Bäuerle, Simon 1; Rettenberger, Luca ORCID iD icon 1; Reischl, Markus ORCID iD icon 1
1 Institut für Automation und angewandte Informatik (IAI), Karlsruher Institut für Technologie (KIT)

Abstract:

Deep learning models for image segmentation achieve high-quality results, but need large amounts of training data. Training data is primarily annotated manually, which is time-consuming and often not feasible for large-scale 2D and 3D images. Manual annotation can be reduced using synthetic training data generated by generative adversarial networks that perform unpaired image-to-image translation. As of now, large images need to be processed patch-wise during inference, resulting in local artifacts in border regions after merging the individual patches. To reduce these artifacts, we propose a new method that integrates overlapping patches into the training process. We incorporated our method into CycleGAN and tested it on our new 2D tiling strategy benchmark dataset. The results show that the artifacts are reduced by 85% compared to state-of-the-art weighted tiling. Additionally, we demonstrate transferability to real-world 3D biological image data, receiving a high-quality synthetic dataset. Increasing the quality of synthetic training datasets can reduce manual annotation, increase the quality of model output, and can help develop and evaluate deep learning models


Verlagsausgabe §
DOI: 10.5445/IR/1000164195
Veröffentlicht am 13.11.2023
Cover der Publikation
Zugehörige Institution(en) am KIT Institut für Automation und angewandte Informatik (IAI)
Publikationstyp Zeitschriftenaufsatz
Publikationsjahr 2023
Sprache Englisch
Identifikator ISSN: 2169-3536
KITopen-ID: 1000164195
HGF-Programm 43.31.02 (POF IV, LK 01) Devices and Applications
Weitere HGF-Programme 47.14.02 (POF IV, LK 01) Information Storage and Processing in the Cell Nucleus
Erschienen in IEEE Access
Verlag Institute of Electrical and Electronics Engineers (IEEE)
Band 11
Seiten 127895-127906
Bemerkung zur Veröffentlichung Gefördert durch den KIT-Publikationsfonds
Nachgewiesen in Web of Science
Scopus
Dimensions
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page