International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 07 | July 2024
www.irjet.net
p-ISSN: 2395-0072
Comparative Analysis of CycleGAN and StyleGAN in Unpaired Image-toImage Translation and High-Quality Image Synthesis Aswathy Ashokan1 1Assistant Professor, Dept. of CSE College of Engineering Munnar, Kerala, India
---------------------------------------------------------------------***--------------------------------------------------------------------Abstract - This paper provides a comparative analysis of two advanced Generative Adversarial Network (GAN) architectures, CycleGAN and StyleGAN, focusing on their applications in unpaired image-to-image translation and high-quality image synthesis. By examining their underlying architectures, training methodologies, and practical applications, aim to elucidate the strengths and limitations of each model. Experimental results on various datasets will be presented to highlight the performance differences, providing insights into their suitability for specific tasks in computer vision.
Key Words: Generative Adversarial Network (GAN), CycleGAN, StyleGAN
1.INTRODUCTION
Fig -1: Basic GAN architecture
Generative Adversarial Networks (GANs) have revolutionized the field of computer vision, enabling the generation of highly realistic images. Among the myriad of GAN variants, CycleGAN and StyleGAN have emerged as two prominent models, each excelling in different applications. CycleGAN is renowned for its ability to perform unpaired image-to-image translation, making it suitable for tasks where paired training data is unavailable. Conversely, StyleGAN is celebrated for its ability to generate high-quality images with fine-grained control over style and attributes, making it ideal for tasks requiring high-fidelity image synthesis.
1.2 CycleGAN CycleGAN is designed for unpaired image-to-image translation. It uses a dual-generator and dual-discriminator architecture to transform images from one domain to another and back again, ensuring consistency through a cycle-consistency loss. This loss penalizes discrepancies between the original images and those reconstructed after a cycle of translations. Additionally, an identity loss is employed to preserve key characteristics of the input images during translation [2][3]. Applications of CycleGAN include style transfer and object transfiguration, where direct paired data is not available [2][3].
1.1 Generative Adversarial Networks (GANs) Generative Adversarial Networks (GANs) consist of two neural networks, a generator and a discriminator, which are trained simultaneously through adversarial training. The generator aims to create realistic images from random noise, while the discriminator attempts to distinguish between real and generated images [1]. The adversarial training process leads to the generator producing increasingly realistic images as it tries to fool the discriminator [1].
Fig 2- CycleGAN architecture
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 274