GAN Research with Dual Generators

This project presents a novel approach to Generative Adversarial Networks (GANs) by employing two generators in competition with each other and against a common discriminator.

Overview

Traditional GANs use a single generator and discriminator to learn and generate realistic images. This project explores the use of two generators that both compete against a common discriminator. The aim is to study the performance and characteristics of dual generators in GAN training.

Features

• Dual Generator Architecture.
• Use of gradient penalty for improved training stability.
• Dynamic directory creation for model checkpoints and generated samples.
• Configurable parameters through a JSON file.
• GPU support detection.

Getting Started

Prerequisites

• Python 3.x
• PyTorch
• torchvision
• tqdm

Installation

1. Clone the repository.
2. Install the required packages.

*See my other projects for more details about setup and configuration

Usage

1. Update the src/settings/settings.py with the correct paths.

2. Configure the training parameters in src/json/params.json.

3. Execute the training:

python run.py

Structure

• run.py: Entry point for training.
• src/app/training.py: Contains training-related functions.
• src/utils/utils.py: Utility functions.
• src/json/params.json: Training parameters in JSON format.
• src/settings/settings.py: Path settings.

License

This project is licensed under the MIT License.