Open AccessResearch on the Application of Double Conditional General Advantage Nets (DCGAN) to generate Simple Background Image with Specified Hue
Author(s) -
Suohe Yang,
Haicheng Bai
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1486/2/022044
Subject(s) - simple (philosophy) , hue , image (mathematics) , computer science , stability (learning theory) , graphics , face (sociological concept) , artificial intelligence , algorithm , machine learning , computer graphics (images) , social science , philosophy , epistemology , sociology
Since the General Advanced Nets (GAN) proposed by Ian Goodfellow in 2014, the research on GAN has become a hot spot of deep learning. People can use GAN to generate images, but there are many problems in GAN, such as the inability to generate images of specified categories, poor model stability, and high training difficulty. Wasserstein GAN improves the disadvantages of the original GAN. Meanwhile, Conditional General Advanced Nets proposes to add condition vectors on the basis of the original GAN, so that the GAN can generate specified pictures, making the original GAN more and more easy to use. Most of the research work of GAN focuses on the real face and real scene generation, but the generation of the commonly used background image in the design is less. In this paper, we use two conditional GAN (DCGAN) and WGAN model framework to generate a simple background image of specified hue and random graphics, which can provide designers with fast and available background images in some simple projects.