UI2IT

Reimplement Unsupervised Image-to-Image Translation Networks (PyTorch)

Introduction

• Replaced Visdom with TensorBoard.
• Restructured in a clear way.
• Less GPU usage than the original code.

.
├── config              # Model config
├── datasets            # Data location
├── experiments         # Log, save model...
├── lib
│   ├── datasets        # Dataset classes
│   ├── layers          # Layer classes
│   ├── models          # Model classes
│   ├── config.py       # Class to load config
│   ├── experiment.py   # Experiment class, define callback here
│   └── runner.py       # Train/val loop
├── tensorboard
├── utils
├── main.py
└── README.md

Usage

python main.py train --exp_name <experiment_name> --cfg <config_path>

Roadmap

Method	Venue	Status
CycleGAN	ICCV 2017	Done
Unsupervised Image-to-Image Translation Networks	NIPS 2017	In progress
Contrastive Learning for Unpaired Image-to-Image Translation	ECCV 2020	To-do
Dual Contrastive Learning for Unsupervised Image-to-Image Translation	CVPRW 2021	To-do