Tan, Jia Ler (2024) Performance comparison between generative adversarial networks (GAN) variants in generating comic character images. Final Year Project, UTAR.
Abstract
Generative Adversarial Networks (GANs) have emerged as a powerful framework for generating realistic and diverse data, including images. This project aims to provide a comprehensive understanding of GANs and their applications in anime face generation. Through theoretical investigation, practical implementation, and empirical analysis, the project explores the working principles of GANs, including their architecture, training dynamics, and variants. The focus is on prominent GAN architectures such as Deep Convolutional GANs (DCGAN), CycleGAN, and Spectral Normalization GAN (SNGAN). The project conducts a thorough performance analysis of these GAN architectures in anime face generation tasks. This involves collecting and preprocessing anime face datasets, training GAN models, and evaluating their performance using quantitative metrics. The quality and diversity of generated anime face images are analyzed using FID and IS score. Furthermore, a comparative analysis of DCGAN, CycleGAN, and SNGAN is conducted to identify their strengths and weaknesses. This comparative study provides insights into the suitability of different GAN architectures for anime face generation applications. The project aims to contribute to the advancement of knowledge in the field of GANs.
Actions (login required)
![[img]](http://eprints.utar.edu.my/style/images/fileicons/application_pdf.png)
