Configurable GAN Trainer

A Python application for training Generative Adversarial Networks (GANs) with configurable architectures and training parameters, featuring a Tkinter-based graphical user interface.

This project allows users to:

• Define Generator and Discriminator architectures via a JSON configuration file.
• Set training parameters (learning rates, loss functions, epochs, batch sizes) through the UI.
• Load and preview network architectures.
• Start, pause, resume, and switch active training between the Generator and Discriminator.
• Monitor training progress with a simple log.
• Visualize training losses.
• Generate sample images from the current Generator.

Project Structure

GAN_Project_Root/
├── .git/
├── .venv/
├── GAN/
│ ├── **init**.py
│ ├── main.py
│ ├── controller.py
│ ├── model_builder.py
│ ├── train_manager.py
│ ├── data_loader.py
│ └── config.json
├── data/
│ └── MNIST/
│ └── raw/
│ └── ...
├── .gitignore
├── pyproject.toml
├── requirements.txt
├── README.md
├── LICENSE
└── uv.lock

Prerequisites

• Python 3.8+
• uv (or pip and venv) for environment and package management

Setup and Installation

1. Clone the repository (if applicable):

   # git clone <repository-url>
   # cd <repository-name>

2. Create and activate a virtual environment: Using uv:

   uv venv .venv
   source .venv/bin/activate

   Or using standard venv:

   python3 -m venv .venv
   source .venv/bin/activate

3. Install dependencies: Using uv:

   uv pip install -r requirements.txt

   Or using pip:

   pip install -r requirements.txt

   If you also want to install the project itself in editable mode (useful for development):

   uv pip install -e . # or pip install -e .

Usage

1. Ensure you have a config.json file (an example is provided in GAN/config.json) defining the GAN architecture and default training settings.

2. Run the application:

   python -m GAN.main

   This will launch the Tkinter GUI.

3. Using the GUI:

   • Network Preview Tab: Load your config.json to see a textual representation of the Generator and Discriminator architectures.
   • Training Parameters Tab:
     • Adjust learning rates, loss functions, epochs, and batch sizes for both Generator and Discriminator.
     • Select the data folder (defaults to ./data for MNIST).
     • Choose which network (Generator or Discriminator) should be active initially.
     • Use the control buttons (Start, Pause, Resume, Switch, Stop) to manage the training process.
     • Training logs will appear in the text area.
   • Summary & Generation Tab:
     • View a summary of the loaded configuration.
     • Plot the training losses (Generator vs. Discriminator).
     • Generate a sample image using the current state of the Generator.

Configuration (config.json)

The config.json file defines:

• generator: Input dimension, layer configurations (type, units/filters, activation, etc.), output size.
• discriminator: Input shape, layer configurations, output size.
• training: Default training parameters for generator and discriminator (loss function, learning rate, epochs, batch size), data folder path, and initial active network.

See GAN/config.json for a detailed example.

Contributing

Contributions are welcome! Please feel free to submit a pull request or open an issue.

License

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