GANs implementation

• GANs implementation
  • Overview
  • Stack
  • Testing
  • License
  • Credits

Overview

In this repository you can find implementation of some GAN written from scratch on pytorch. For now, there you can find such architectures, as:

• Deep Convolutional (DC) GAN;
• Conditional Variation of DC GAN that works with finite number of classes;
• Wasserstein GAN;
• Wasserstein GAN with gradient penalty;
• Pix2Pix model;

Later I'll also add:

• CycleGAN;
• StyleGAN;
• ProGAN;
• SuperResolution (SR) GAN;
• ESRGAN;

And a banch of metrics, such as:

• Frechet inception distance;
• Leave-one-out 1NN score;

Stack

My stack here is:

• Models from scratch via PyTorch;
• Mypy as a type checker;
• Testing with pytest;
• Ruff as a linter;

Testing

All the code was tested on Ubuntu 20.04 and Windownd 11.

License

Here I use MIT License so feel free to use this code for any your purpose.

Credits

Many thanks to Aladdin Persson for his awesome GAN playlist that I find very helpful.