An examination of synthetic images produced with DCGAN according to the size of data and epoch

Abstract

In recent years, the popular network of adversarial networks has increased in studies for computer vision. The lack of data used in the studies and the lack of good training for the resulting model draw attention to techniques such as data enhancement and synthetic data generation. In this article, synthetic data was produced using Generative Adversarial Networks (GANs). The data in the dataset used consists of 10000 faces from the CelebA dataset available online. The impact of the increase in the number of data on fake images created by DCGAN, one of the GANs, is the main topic of the article. In the study, the data is divided into two parts. In the first study, fake data were generated from 5000 data, and in the next study, fake data images were forged using all of the data meaning 10000 data. The result was found that the number of data and the increase in epoch were accurately proportional to the success of the fraudulent images created.