Photo-To-Cartoon Translation with Generative Adversarial Network
Date
2022
Student ID
Research Supervisor
Editor
Journal Title
Volume
Issue
Journal Title
Journal ISSN
Volume Title
Abstract
Cartoons are a popular art form in our daily lives, and the ability to automatically create cartoon graphics from photos is highly desired. Cartoon images have a more vibrant and lively appearance than traditional naive pictures. This study aims to explain the process of translating real-world photos into cartoon-like images. While converting pictures to cartoons, there were a few difficulties, including fine hair edges, mismatched colors, and texture concerns. Photos were converted to cartoon-style images using generative adversarial networks (GAN). Various neural network-based GAN networks, DCGAN, CycleGAN, and AnimeGAN, have been applied in this work for cartoon conversion. Among them, CycleGAN performs better in transforming actual photographs into colorful, eye-catching cartoons. This project's approach is based on learning-based methodologies, which have lately gained popularity for stylizing images in artistic forms like painting. The results may be used to convert real-world photographs to high-quality cartoon graphics quickly. This project provides a web API that contains training weights derived from the models outlined below. Based on that API, we created a web app that converts real-world images into high-quality cartoon graphics for various cartoon styles. In these experiments, it outperforms state-of-the-art approaches to producing high-quality cartoon graphics from real-world photos. Numerical results show that the CycleGAN approach has the lowest training time per epoch and requires the minimum number of trainable parameters.
Description
Keywords
Citation
Department Name
Electrical and Computer Engineering
Publisher
North South University