Dual Discriminator Generative Adversarial Nets

Nguyen, Tu Dinh; Le, Trung; Vu, Hung V.; Phung, Dinh

doi:10.48550/arxiv.1709.03831

Cited by 7 publications

(8 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that the relationship among generators (discriminators) can be explored through weight sharing as explained in Section 5. 3.…”

Section: Methodsmentioning

confidence: 99%

“…As a supervised learning method, in its implementation, the label of each image needs to be available, which means that it has stronger or stricter requirements than ours. GAN with Dual Discriminators (D2GAN*) This model has the identical structure with the D2GAN model proposed by Nguyen et al [3]. Two discriminators in the original D2GAN learn from the same datasets.…”

Section: Baselinesmentioning

confidence: 99%

See 1 more Smart Citation

IntersectGAN

Yao¹,

Zhang²,

Wang³

et al. 2019

Proceedings of the 27th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Generative adversarial networks (GANs) have demonstrated great success in generating various visual content. However, images generated by existing GANs are often of attributes (e.g., smiling expression) learned from one image domain. As a result, generating images of multiple attributes requires many real samples possessing multiple attributes which are very resource expensive to be collected. In this paper, we propose a novel GAN, namely Intersect-GAN, to learn multiple attributes from different image domains through an intersecting architecture. For example, given two image domains X 1 and X 2 with certain attributes, the intersection X 1 ∩ X 2 denotes a new domain where images possess the attributes from both X 1 and X 2 domains. The proposed IntersectGAN consists of two discriminators D 1 and D 2 to distinguish between generated and real samples of different domains, and three generators where the intersection generator is trained against both discriminators. And an overall adversarial loss function is defined over three generators. As a result, our proposed IntersectGAN can be trained on multiple domains of which each presents one specific attribute, and eventually eliminates the need of real sample images simultaneously possessing multiple attributes. By using the CelebFaces Attributes dataset, our proposed IntersectGAN is able to produce high quality face images possessing multiple attributes (e.g., a face with black hair and a smiling expression). Both qualitative and quantitative evaluations are conducted to compare our proposed IntersectGAN with other baseline methods. Besides, several different applications of IntersectGAN have been explored with promising results. CCS CONCEPTS• Computing methodologies → Computer vision tasks; Learning paradigms.

show abstract

“…Note that the relationship among generators (discriminators) can be explored through weight sharing as explained in Section 5. 3.…”

Section: Methodsmentioning

confidence: 99%

Section: Baselinesmentioning

confidence: 99%

IntersectGAN

Yao¹,

Zhang²,

Wang³

et al. 2019

Proceedings of the 27th ACM International Conference on Multimedia

View full text Add to dashboard Cite

show abstract

“…Recent attempts have been made to solve the mode collapsing problem by improving the training [19]. The idea of applying a dual discriminator GAN (D2GAN) as a method to tackle the mode collapse has been discussed in [20]. We propose here a new approach motivated by [20] and based on the Wasserstein loss function.…”

Section: Model Architecturementioning

confidence: 99%

“…The idea of applying a dual discriminator GAN (D2GAN) as a method to tackle the mode collapse has been discussed in [20]. We propose here a new approach motivated by [20] and based on the Wasserstein loss function. We term our proposed model as Dual Discriminator Wasserstein GAN (D2WGAN).…”

Section: Model Architecturementioning

confidence: 99%

Audio inpainting with generative adversarial network

Ebner

Eltelt

2020

Preprint

View full text Add to dashboard Cite

We study the ability of Wasserstein Generative Adversarial Network (WGAN) to generate missing audio content which is, in context, (statistically similar) to the sound and the neighboring borders. We deal with the challenge of audio inpainting long range gaps (500 ms) using WGAN models. We improved the quality of the inpainting part using a new proposed WGAN architecture that uses a short-range and a long-range neighboring borders compared to the classical WGAN model. The performance was compared with two different audio instruments (piano and guitar) and on virtuoso pianists together with a string orchestra. The objective difference grading (ODG) was used to evaluate the performance of both architectures. The proposed model outperforms the classical WGAN model and improves the reconstruction of high-frequency content. Further, we got better results for instruments where the frequency spectrum is mainly in the lower range where small noises are less annoying for human ear and the inpainting part is more perceptible. Finally, we could show that better test results for audio dataset were reached where a particular instrument is accompanist by other instruments if we train the network only on this particular instrument neglecting the other instruments.

show abstract

“…Improve the structure of GAN model, such as training the interaction between multiple discriminator D and multiple generator G, such as DualGAN[16] model consisting of two D and two G.…”

mentioning

confidence: 99%

Study of Restrained Network Structures for Wasserstein Generative Adversarial Networks (WGANs) on Numeric Data Augmentation

Wang

Cui

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Some recent studies have suggested using Generative Adversarial Network (GAN) for numeric data over-sampling, which is to generate data for completing the imbalanced numeric data. Compared with the conventional over-sampling methods, taken SMOTE as an example, the recently-proposed GAN schemes fail to generate distinguishable augmentation results for classifiers. In this paper, we discuss the reason for such failures, based on which we further study the restrained conditions between G and D theoretically, and propose a quantitative indicator of the restrained structure, called Similarity of the Restrained Condition (SRC) to measure the restrained conditions. Practically, we propose several candidate solutions, which are isomorphic (IWGAN) mirror (MWGAN) and self-symmetric WGAN (SWGAN) for restrained conditions. Besides, the restrained WGANs enhance the classification performance in AUC on five classifiers compared with the original data as the baseline, conventional SMOTE, and other GANs add up to 20 groups of experiments in four datasets. The restrained WGANs outperform all others in 17/20 groups, among which IWGAN accounted for 15/17 groups and the SRC is an effective measure in evaluating the restraints so that further GAN structures with G-D restrains could be designed on SRC. Multidimensional scaling (MDS) is introduced to eliminate the impact of datasets and evaluation of the AUC in a composite index and IWGAN decreases the MDS distance by 20% to 40%. Moreover, the convergence speed of IWGAN is increased, and the initial error of loss function is reduced. INDEX TERMS Restrained network structures, generative adversarial network, numeric data augmentation.

show abstract

Dual Discriminator Generative Adversarial Nets

Cited by 7 publications

References 14 publications

IntersectGAN

IntersectGAN

Audio inpainting with generative adversarial network

Study of Restrained Network Structures for Wasserstein Generative Adversarial Networks (WGANs) on Numeric Data Augmentation

Contact Info

Product

Resources

About