Controllable generative adversarial network

Minhyeok Lee, Junhee Seok

Research output: Contribution to journalArticle

Abstract

Recently introduced generative adversarial networks (GANs) have been shown numerous promising results to generate realistic samples. In the last couple of years, it has been studied to control features in synthetic samples generated by the GAN. Auxiliary classifier GAN (ACGAN), a conventional method to generate conditional samples, employs a classification layer in discriminator to solve the problem. However, in this paper, we demonstrate that the auxiliary classifier can hardly provide good guidance for training of the generator, where the classifier suffers from overfitting. Since the generator learns from classification loss, such a problem has a chance to hinder the training. To overcome this limitation, here, we propose a controllable GAN (ControlGAN) structure. By separating a feature classifier from the discriminator, the classifier can be trained with data augmentation technique, which can support to make a fine classifier. Evaluated with the CIFAR-10 dataset, ControlGAN outperforms AC-WGAN-GP which is an improved version of the ACGAN, where Inception score of the ControlGAN is 8.61 ± 0.10. Furthermore, we demonstrate that the ControlGAN can generate intermediate features and opposite features for interpolated input and extrapolated input labels that are not used in the training process. It implies that the ControlGAN can significantly contribute to the variety of generated samples.

Original languageEnglish
Article number8641270
Pages (from-to)28158-28169
Number of pages12
JournalIEEE Access
Volume7
DOIs
Publication statusPublished - 2019 Jan 1

Fingerprint

Classifiers
Discriminators
Labels

Keywords

  • Artificial neural network
  • generative adversarial networks
  • generative model
  • sample generation

ASJC Scopus subject areas

  • Computer Science(all)
  • Materials Science(all)
  • Engineering(all)

Cite this

Controllable generative adversarial network. / Lee, Minhyeok; Seok, Junhee.

In: IEEE Access, Vol. 7, 8641270, 01.01.2019, p. 28158-28169.

Research output: Contribution to journalArticle

Lee, Minhyeok ; Seok, Junhee. / Controllable generative adversarial network. In: IEEE Access. 2019 ; Vol. 7. pp. 28158-28169.
@article{855abe001a424765b237c51524fb8a9b,
title = "Controllable generative adversarial network",
abstract = "Recently introduced generative adversarial networks (GANs) have been shown numerous promising results to generate realistic samples. In the last couple of years, it has been studied to control features in synthetic samples generated by the GAN. Auxiliary classifier GAN (ACGAN), a conventional method to generate conditional samples, employs a classification layer in discriminator to solve the problem. However, in this paper, we demonstrate that the auxiliary classifier can hardly provide good guidance for training of the generator, where the classifier suffers from overfitting. Since the generator learns from classification loss, such a problem has a chance to hinder the training. To overcome this limitation, here, we propose a controllable GAN (ControlGAN) structure. By separating a feature classifier from the discriminator, the classifier can be trained with data augmentation technique, which can support to make a fine classifier. Evaluated with the CIFAR-10 dataset, ControlGAN outperforms AC-WGAN-GP which is an improved version of the ACGAN, where Inception score of the ControlGAN is 8.61 ± 0.10. Furthermore, we demonstrate that the ControlGAN can generate intermediate features and opposite features for interpolated input and extrapolated input labels that are not used in the training process. It implies that the ControlGAN can significantly contribute to the variety of generated samples.",
keywords = "Artificial neural network, generative adversarial networks, generative model, sample generation",
author = "Minhyeok Lee and Junhee Seok",
year = "2019",
month = "1",
day = "1",
doi = "10.1109/ACCESS.2019.2899108",
language = "English",
volume = "7",
pages = "28158--28169",
journal = "IEEE Access",
issn = "2169-3536",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Controllable generative adversarial network

AU - Lee, Minhyeok

AU - Seok, Junhee

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Recently introduced generative adversarial networks (GANs) have been shown numerous promising results to generate realistic samples. In the last couple of years, it has been studied to control features in synthetic samples generated by the GAN. Auxiliary classifier GAN (ACGAN), a conventional method to generate conditional samples, employs a classification layer in discriminator to solve the problem. However, in this paper, we demonstrate that the auxiliary classifier can hardly provide good guidance for training of the generator, where the classifier suffers from overfitting. Since the generator learns from classification loss, such a problem has a chance to hinder the training. To overcome this limitation, here, we propose a controllable GAN (ControlGAN) structure. By separating a feature classifier from the discriminator, the classifier can be trained with data augmentation technique, which can support to make a fine classifier. Evaluated with the CIFAR-10 dataset, ControlGAN outperforms AC-WGAN-GP which is an improved version of the ACGAN, where Inception score of the ControlGAN is 8.61 ± 0.10. Furthermore, we demonstrate that the ControlGAN can generate intermediate features and opposite features for interpolated input and extrapolated input labels that are not used in the training process. It implies that the ControlGAN can significantly contribute to the variety of generated samples.

AB - Recently introduced generative adversarial networks (GANs) have been shown numerous promising results to generate realistic samples. In the last couple of years, it has been studied to control features in synthetic samples generated by the GAN. Auxiliary classifier GAN (ACGAN), a conventional method to generate conditional samples, employs a classification layer in discriminator to solve the problem. However, in this paper, we demonstrate that the auxiliary classifier can hardly provide good guidance for training of the generator, where the classifier suffers from overfitting. Since the generator learns from classification loss, such a problem has a chance to hinder the training. To overcome this limitation, here, we propose a controllable GAN (ControlGAN) structure. By separating a feature classifier from the discriminator, the classifier can be trained with data augmentation technique, which can support to make a fine classifier. Evaluated with the CIFAR-10 dataset, ControlGAN outperforms AC-WGAN-GP which is an improved version of the ACGAN, where Inception score of the ControlGAN is 8.61 ± 0.10. Furthermore, we demonstrate that the ControlGAN can generate intermediate features and opposite features for interpolated input and extrapolated input labels that are not used in the training process. It implies that the ControlGAN can significantly contribute to the variety of generated samples.

KW - Artificial neural network

KW - generative adversarial networks

KW - generative model

KW - sample generation

UR - http://www.scopus.com/inward/record.url?scp=85063259875&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85063259875&partnerID=8YFLogxK

U2 - 10.1109/ACCESS.2019.2899108

DO - 10.1109/ACCESS.2019.2899108

M3 - Article

VL - 7

SP - 28158

EP - 28169

JO - IEEE Access

JF - IEEE Access

SN - 2169-3536

M1 - 8641270

ER -