Simple GAN for MNIST Dataset

This project implements a simple Generative Adversarial Network (GAN) to generate MNIST-like images using PyTorch. It includes:

• A Discriminator class to distinguish between real and fake images.
• A Generator class to generate new images from random noise.
• A main.py script to train the GAN and log results to TensorBoard.

Code Source: This implementation is inspired by this YouTube tutorial by the channel.

Features

• Utilizes PyTorch for model implementation and training.
• Trains on the MNIST dataset with normalized input data.
• Logs loss values and visualizes generated images in TensorBoard.
• Supports GPU training for faster performance.

Project Structure

.
├── src/
│   ├── models/
│   │   ├── discriminator.py  # Defines the Discriminator class
│   │   ├── generator.py      # Defines the Generator class
├── main.py                   # Main script for training the GAN
├── dataset/                  # Directory where MNIST data is downloaded
├── runs/                     # TensorBoard logs for real and fake images
├── requirements.txt          # Python package dependencies
└── README.md                 # Project documentation

Requirements

All dependencies are listed in the requirements.txt file. Install them using the following command:

pip install -r requirements.txt

Installation

1. Clone the repository:

   git clone https://github.com/your-username/simpleGAN.git
   cd simpleGAN

2. Create a virtual environment and activate it:

   python -m venv env
   source env/bin/activate  # For Linux/Mac
   env\Scripts\activate   # For Windows

3. Install the required Python packages:

   pip install -r requirements.txt

Training the GAN

1. Run the main.py script to start training:

   python main.py

2. View training progress and generated images using TensorBoard:

   tensorboard --logdir=runs

   Open the URL provided by TensorBoard to visualize logs and images.

Outputs

• Loss Values: Discriminator and generator loss values printed in the console during training.
• Generated Images: Fake images generated at each epoch, logged in TensorBoard.

Customization

• Modify hyperparameters such as learning rate, batch size, or the number of epochs in main.py.
• Adjust the architecture of the Discriminator or Generator in their respective files under src/models/.

Acknowledgments

This project is inspired by this YouTube tutorial.

License

This project is licensed under the MIT License. See the LICENSE.md file for details.

Badges

• Python Version: 3.8+
• PyTorch Version: 1.12+
• TensorBoard: Enabled