Anime Face Generation using Generative Adversial Networks in Deep Learning

Authors

  • Anjana M S, Dr. Dhanya N M

Keywords:

GAN, DCGAN, Style2GAN, Deep Learning, Machine Learning, Artificial Intelligence, Generation

Abstract

Anime characters are used not only in books, but also in entertainment, awareness shows, video games, etc. In the recent times, there are many systems built for anime face generations. There are also various kinds of Artificial Intelligence approaches used to solve this, but the most famous ones are Generative Adversial Networks. The main reason is that the efficiency and performance of Generative Adversial Networks are found to be increasing as days go by. This paper is written to compare the quality of images, performance and efficiency and performance of two types of Generative Adversial Networks; namely Deep Convolutional Generative Adversial Network, and Style Generative Adversial Net-work2. Fréchet Inception Distance (FID) is taken as the evaluation metric for the systems implemented. For Deep Convolutional Generative Adversial Network, an FID score of 624.04 is found; whereas for Style Generative Adversial Net-work2, an FID score of 30.6 is found. From the resultant images and the Fréchet Inception Distance score, it is evident that Style Generative Adversial Network2 is the best model for anime face generation

Downloads

Published

2022-07-25