Exploring DeshuffleGANs in Self-Supervised Generative Adversarial Networks

Gulcin Baykal*, Furkan Ozcelik, Gozde Unal

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

11 Citations (Scopus)

Abstract

Generative Adversarial Networks (GANs) have become the most used networks towards solving the problem of image generation. Self-supervised GANs are later proposed to avoid the catastrophic forgetting of the discriminator and to improve the image generation quality without needing the class labels. However, the generalizability of the self-supervision tasks on different GAN architectures is not studied before. To that end, we extensively analyze the contribution of a previously proposed self-supervision task, deshuffling of the DeshuffleGANs in the generalizability context. We assign the deshuffling task to two different GAN discriminators and study the effects of the task on both architectures. We extend the evaluations compared to the previously proposed DeshuffleGANs on various datasets. We show that the DeshuffleGAN obtains the best FID results for several datasets compared to the other self-supervised GANs. Furthermore, we compare the deshuffling with the rotation prediction that is firstly deployed to the GAN training and demonstrate that its contribution exceeds the rotation prediction. We design the conditional DeshuffleGAN called cDeshuffleGAN to evaluate the quality of the learnt representations. Lastly, we show the contribution of the self-supervision tasks to the GAN training on the loss landscape and present that the effects of these tasks may not be cooperative to the adversarial training in some settings. Our code can be found at https://github.com/gulcinbaykal/DeshuffleGAN.

Original languageEnglish
Article number108244
JournalPattern Recognition
Volume122
DOIs
Publication statusPublished - Feb 2021

Bibliographical note

Publisher Copyright:
© 2021 Elsevier Ltd

Funding

In this work, Gulcin Baykal and Furkan Ozcelik were supported by the Turkcell-ITU Researcher Funding Program. Gulcin Baykal was further supported by DeepMind Scholarship Program at ITU. This work was also supported by the Scientific Research Project Unit of Istanbul Technical University [project number MOA-2019-42321]. Gozde Unal is a professor of Artificial Intelligence and Data Engineering / Computer Engineering at Istanbul Technical University (ITU). She is the founder of ITU-AI Center. She is the recipient of the Marie Curie Alumni Association (MCAA) Career Award 2016 of European Commission. Her main research interests are in learning models of AI, particularly representation learning, and uncertainty in deep learning.

FundersFunder number
Marie Curie Alumni Association
International Technological University
European Commission
Istanbul Teknik ÜniversitesiMOA-2019-42321

    Keywords

    • DeshuffleGANs
    • Deshuffling
    • Generative adversarial networks
    • Self-Supervised generative adversarial networks
    • Self-supervised learning

    Fingerprint

    Dive into the research topics of 'Exploring DeshuffleGANs in Self-Supervised Generative Adversarial Networks'. Together they form a unique fingerprint.

    Cite this