GAN, SSGAN, WGAN, and VAE are neural networks for content generation. GAN generates realistic images, SSGAN improves quality, WGAN ensures stability, and VAE compresses data to learn features. Applications include image generation, quality enhancement, and fraud detection.

Using NN (Neural Networks) to tackle continuous optimal transport problem is a promising approach especially for unpaired style-transfer problem. This method learns a one-to-one mapping (OT map) between the source and target data distributions but uses adversarial training (similar to GANs), which is not very stable.

This repository contains PyTorch implementations of modern generative models, with extensive experiments on cat image generation.