search for




 

A study on the performance of generative adversarial networks
Journal of the Korean Data & Information Science Society 2018;29:1155-67
Published online September 30, 2018
© 2018 Korean Data and Information Science Society.

Yeongjae Lee1 · Kyungha Seok2

12Department of Statistics, Inje University
Correspondence to: Professor, Institute of Statistical Information, Department of Statistics, Inje University, Gyungnam 50834, Korea. E-mail: statskh@inje.ac.kr
This work was supported by the 2017 Inje University research grant.
Received July 5, 2018; Revised August 10, 2018; Accepted August 20, 2018.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
Generative Adversarial Networks (GAN) is one of the most popular models in generative deep learning models. Many derivatives have been published and researches have been conducted in various fields. In this study, we review the derivatives of GAN and compare them. We determine the proper dimension of the latent space and compare the metrics Fréchet Inception distance (FID) and Inception score (IS) which are used for evaluating generated data. The experiments show that GAN-NS and LSGAN works well and FID is superior to IS. And the 10 dimensional latent spaces yield good results, which is not much different from the result of typical 100 dimensions.
Keywords : Deep learning, Fréchet Inception distance, generative adversarial networks, Inception score, latent space.