GAN_Intro

This repository provides an introduction to Generative Adversarial Networks (GANs) using PyTorch. It contains two Google Colab notebooks that demonstrate the implementation of:

1. Simple GAN on the MNIST dataset.
2. Wasserstein GAN (WGAN) on the CelebA dataset using Wasserstein Loss.

Contents

• GAN_MNIST.ipynb: A basic implementation of a GAN on the MNIST dataset. This notebook walks through the process of creating a simple GAN architecture, training it, and generating new images of handwritten digits.

• WGAN_CelebA.ipynb: An implementation of a Wasserstein GAN (WGAN) on the CelebA dataset. This notebook demonstrates the usage of the Wasserstein Loss function to improve the stability of GAN training and generate realistic images of faces.

Requirements

To run these notebooks, you will need the following dependencies:

• Python 3.x
• PyTorch
• torchvision
• numpy
• matplotlib
• Google Colab (recommended for ease of use)

Getting Started

1. Clone the repository:

   git clone https://github.com/Daimon5/GAN_Intro.git

2. Open the desired notebook in Google Colab by uploading it or using the GitHub link directly.

How to Use

• Simple GAN:

  • Follow the notebook to understand how a basic GAN works.
  • Train the GAN on the MNIST dataset.
  • Generate new images of handwritten digits.

• Wasserstein GAN (WGAN):

  • Explore the advantages of using Wasserstein Loss in GANs.
  • Train the WGAN on the CelebA dataset.
  • Generate realistic images of human faces.

References

• Original GAN Paper
• Wasserstein GAN Paper
• PyTorch Documentation

License

This project is licensed under the MIT License - see the LICENSE file for details.