Welcome to the Generative Adversarial Networks (GAN) repository! This repository features a variety of GAN architectures and experiments, showcasing their potential in tasks like image generation, domain translation, and high-resolution image synthesis. Explore implementations of DCGAN, CycleGAN, Pix2Pix, Progressive Growing GAN, and more.
Generative Adversarial Networks (GANs) are a groundbreaking class of machine learning models. A GAN consists of two networks:
- Generator: Creates synthetic data that mimics real data.
- Discriminator: Differentiates between real and generated data.
These two networks are trained in a zero-sum game, pushing the Generator to produce increasingly realistic outputs. GANs have applications in:
- Image Synthesis: Generating high-quality, photorealistic images.
- Image-to-Image Translation: Transforming images from one domain to another.
- Unpaired Domain Adaptation: Learning mappings between datasets without direct pairings.
- Progressive Learning: Generating high-resolution images via stepwise training.
Generative-Adversarial-Network/
├── Basic_GAN.ipynb # Basic GAN implementation
├── Conditional_GAN.ipynb # Conditional GAN implementation
├── CycleGAN for Unpaired Image Translation.ipynb # CycleGAN implementation
├── Deep_Convolutional_GAN.ipynb # DCGAN implementation
├── Pix2Pix for Paired Image Translation.ipynb # Pix2Pix implementation
├── Progressive Growing GAN.ipynb # Progressive GAN for high-res image data
├── README.md # Repository documentation
🔧 Features
Basic GAN: Demonstrates the fundamental GAN concept with a simple generator and discriminator.
Deep Convolutional GAN (DCGAN): Uses convolutional layers for stable, high-quality image generation.
Conditional GAN (cGAN): Generates data conditioned on additional information like class labels.
CycleGAN: Performs unpaired image-to-image translation tasks, e.g., converting horses to zebras.
Pix2Pix: Executes paired image-to-image translation, e.g., transforming sketches to photos.
Progressive Growing GAN: Produces high-resolution images by progressively increasing model complexity.
Visual Demonstrations
🌟 Highlights
GAN Architectures:
Basic GAN: Introduces the core GAN framework.
DCGAN: Employs deep convolutional networks for generating photorealistic images.
Conditional GAN: Adds control over the generation process through input conditions.
CycleGAN: Enables translation between two unpaired datasets.
Pix2Pix: Translates between two paired datasets with pixel-level mappings.
Progressive GAN: Scales GANs to synthesize high-resolution images.
🛠️ Technologies Used
Python 3.x
TensorFlow / PyTorch
NumPy
Matplotlib
OpenCV
PIL (Python Imaging Library)
📊 Getting Started
Clone the repository:
git clone https://github.com/Someshdiwan/Generative-Adversarial-Network-
cd Generative-Adversarial-Network
Install the required Python libraries: pip install
Explore and execute the Jupyter notebooks to experience the power of GANs.
📚 Resources
Tutorials and Research Papers:
Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks: https://junyanz.github.io/CycleGAN/
Style based GAN: https://www.naukri.com/code360/library/style-based-gan
Original GAN Paper/Generative Adversarial Nets (Goodfellow et al., 2014): https://arxiv.org/abs/1406.2661
Progressive Growing of GANs: https://arxiv.org/abs/1710.10196
🤝 Contributions
We welcome your contributions to enhance this repository!
You can: Report issues or suggest features.
Fork the repository, make improvements, and submit a pull request.
🌟 Support
If you find this repository helpful, please consider giving it a ⭐ on GitHub to show your support!