Investigating the effect of loss functions on single-image GAN performance
Loss functions are crucial in training generative adversarial networks (GANs) and shaping the resulting outputs. These functions, specifically designed for GANs, optimize generator and discriminator networks together but in opposite directions. GAN models, which typically handle large datasets, have...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Bursa Technical University
2024-12-01
|
| Series: | Journal of Innovative Science and Engineering |
| Subjects: | |
| Online Access: | http://jise.btu.edu.tr/en/download/article-file/3991473 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832586971534327808 |
|---|---|
| author | Eyyup YİLDİZ Mehmet Erkan YUKSEL Selcuk SEVGEN |
| author_facet | Eyyup YİLDİZ Mehmet Erkan YUKSEL Selcuk SEVGEN |
| author_sort | Eyyup YİLDİZ |
| collection | DOAJ |
| description | Loss functions are crucial in training generative adversarial networks (GANs) and shaping the resulting outputs. These functions, specifically designed for GANs, optimize generator and discriminator networks together but in opposite directions. GAN models, which typically handle large datasets, have been successful in the field of deep learning. However, exploring the factors that influence the success of GAN models developed for limited data problems is an important area of research. In this study, we conducted a comprehensive investigation into the loss functions commonly used in GAN literature, such as binary cross entropy (BCE), Wasserstein generative adversarial network (WGAN), least squares generative adversarial network (LSGAN), and hinge loss. Our research focused on examining the impact of these loss functions on improving output quality and ensuring training convergence in single-image GANs. Specifically, we evaluated the performance of a single-image GAN model, SinGAN, using these loss functions in terms of image quality and diversity. Our experimental results demonstrated that loss functions successfully produce high-quality, diverse images from a single training image. Additionally, we found that the WGAN-GP and LSGAN-GP loss functions are more effective for single-image GAN models. |
| format | Article |
| id | doaj-art-f79733079baa4f6ebf9813c54ad307d6 |
| institution | Kabale University |
| issn | 2602-4217 |
| language | English |
| publishDate | 2024-12-01 |
| publisher | Bursa Technical University |
| record_format | Article |
| series | Journal of Innovative Science and Engineering |
| spelling | doaj-art-f79733079baa4f6ebf9813c54ad307d62025-01-24T19:11:04ZengBursa Technical UniversityJournal of Innovative Science and Engineering2602-42172024-12-018221322510.38088/jise.1497968Investigating the effect of loss functions on single-image GAN performanceEyyup YİLDİZ0https://orcid.org/0000-0002-7051-3368Mehmet Erkan YUKSEL1https://orcid.org/0000-0001-8976-9964Selcuk SEVGEN2https://orcid.org/0000-0003-1443-1779ERZINCAN BINALI YILDIRIM UNIVERSITYBURDUR MEHMET AKİF ERSOY UNIVERSITYİSTANBUL UNIVERSITY-CERRAHPASALoss functions are crucial in training generative adversarial networks (GANs) and shaping the resulting outputs. These functions, specifically designed for GANs, optimize generator and discriminator networks together but in opposite directions. GAN models, which typically handle large datasets, have been successful in the field of deep learning. However, exploring the factors that influence the success of GAN models developed for limited data problems is an important area of research. In this study, we conducted a comprehensive investigation into the loss functions commonly used in GAN literature, such as binary cross entropy (BCE), Wasserstein generative adversarial network (WGAN), least squares generative adversarial network (LSGAN), and hinge loss. Our research focused on examining the impact of these loss functions on improving output quality and ensuring training convergence in single-image GANs. Specifically, we evaluated the performance of a single-image GAN model, SinGAN, using these loss functions in terms of image quality and diversity. Our experimental results demonstrated that loss functions successfully produce high-quality, diverse images from a single training image. Additionally, we found that the WGAN-GP and LSGAN-GP loss functions are more effective for single-image GAN models.http://jise.btu.edu.tr/en/download/article-file/3991473generative adversarial networkslow data regimesingle-image ganloss functionsimage diversity |
| spellingShingle | Eyyup YİLDİZ Mehmet Erkan YUKSEL Selcuk SEVGEN Investigating the effect of loss functions on single-image GAN performance Journal of Innovative Science and Engineering generative adversarial networks low data regime single-image gan loss functions image diversity |
| title | Investigating the effect of loss functions on single-image GAN performance |
| title_full | Investigating the effect of loss functions on single-image GAN performance |
| title_fullStr | Investigating the effect of loss functions on single-image GAN performance |
| title_full_unstemmed | Investigating the effect of loss functions on single-image GAN performance |
| title_short | Investigating the effect of loss functions on single-image GAN performance |
| title_sort | investigating the effect of loss functions on single image gan performance |
| topic | generative adversarial networks low data regime single-image gan loss functions image diversity |
| url | http://jise.btu.edu.tr/en/download/article-file/3991473 |
| work_keys_str_mv | AT eyyupyildiz investigatingtheeffectoflossfunctionsonsingleimageganperformance AT mehmeterkanyuksel investigatingtheeffectoflossfunctionsonsingleimageganperformance AT selcuksevgen investigatingtheeffectoflossfunctionsonsingleimageganperformance |