Implementations of Image to Image Translation via Generative Models using PyTorch

0. Introduction

This repository contains implementations of fundamental Image-to-Image Translation via Generative Models, including Pix2Pix, DiscoGAN, CycleGAN, BicycleGAN, and StarGAN, Unsupervised Attention-Guided GAN, MUNIT, and U-GAT-IT.

Please note that I focused on implementation rather than deriving the best results. In other words, a set of hyper-parameters that I used may not produce the best results. For example, you can expect better CycleGAN results when increasing total epochs to 200.

Method	Modality	Papers
Supervised Learning	Uni-modal	Pix2Pix
	Multi-modal	BicycleGAN
Unsupervised Learning	Uni-modal	DiscoGAN & CycleGAN & UAG-GAN & U-GAT-IT
	Multi-modal	StarGAN & MUNIT

1. Pix2Pix : Image-to-Image Translation with Conditional Adversarial Networks

2. DiscoGAN : Learning to Discover Cross-Domain Relations with Generative Adversarial Networks

3. CycleGAN : Image-to-Image Translation with Conditional Adversarial Networks

4. BicycleGAN : Toward Multimodal Image-to-Image Translation

5. StarGAN : Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation

6. Unsupervised Attention-Guided GAN : Unsupervised Attention-guided Image to Image Translation

7. MUNIT : Multimodal Unsupervised Image-to-Image Translation

8. U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation

Development Environment

- Ubuntu 18.04 LTS
- NVIDIA GFORCE GTX 1080 ti
- CUDA 10.2
- torch 1.5.1
- torchvision 0.5.0
- etc