Image-to-Image Translation for Domain Adaptation
JoliGAN provides models and machine learning tools for unpaired image to image translation and domain adaptation.
- Multiple models based on adversarial generation: CycleGAN, CyCADA, CUT and more
- Many generator architectures such as styleGAN2 decoder / mobile resnet, attention resnet,...
- Semantic consistency allowing to perform domain adaptation e.g. from synthetic to real-like data
Use cases
- Image to image translation while preserving semantic
- Image to image translation to cope with scarce data
Models
| Name | Paper |
|---|---|
| CycleGAN | https://arxiv.org/abs/1703.10593 |
| CyCADA | https://arxiv.org/abs/1711.03213 |
| CUT | https://arxiv.org/abs/2007.15651 |
| RecycleGAN | https://arxiv.org/abs/1808.05174 |
| StyleGAN2 | https://arxiv.org/abs/1912.04958 |
Generator architectures
| Architecture | Number of parameters |
|---|---|
| Resnet 9 blocks | 11.378M |
| Mobile resnet 9 blocks | 1.987M |
| Resnet attn | 11.823M |
| Mobile resnet attn | 2.432M |
| Segformer b0 | 4.158M |
| Segformer attn b0 | 4.60M |
| Segformer attn b1 | 14.724M |
| Segformer attn b5 | 83.016M |
Quick Start
Prerequisites
- Linux
- Python 3
- CPU or NVIDIA GPU + CUDA CuDNN
Installation
Clone this repo:
git clone --recursive https://github.com/jolibrain/joliGAN.git
cd joliGANInstall PyTorch and other dependencies (torchvision, visdom and dominate, FID).
For pip users, please type the command pip install -r requirements.txt.
JoliGAN training
With a dataset located in directory dataroot:
- Train with cut :
You can modify the hyperparameters in ./scripts/train_cut.sh and then use the following line command.
bash ./scripts/train_cut.sh dataroot
- Train with cut and labels :
You can modify the hyperparameters in ./scripts/train_cut_semantic.sh and then use the following line command.
bash ./scripts/train_cut_semantic.sh dataroot
- Train with cut and mask labels :
You can modify the hyperparameters in ./scripts/train_cut_semantic_mask.sh and then use the following line command.
bash ./scripts/train_cut_semantic_mask.sh dataroot
- Train with cycleGAN :
You can modify the hyperparameters in ./scripts/train_cyclegan.sh and then use the following line command.
bash ./scripts/train_cyclegan.sh dataroot
- Train with cycleGAN and labels :
You can modify the hyperparameters in ./scripts/train_cyclegan_semantic.sh and then use the following line command.
bash ./scripts/train_cyclegan_semantic.sh dataroot
- Train with cycleGAN and mask labels :
You can modify the hyperparameters in ./scripts/train_cyclegan_semantic_mask.sh and then use the following line command.
bash ./scripts/train_cyclegan_semantic_mask.sh dataroot
JoliGAN server
Ensure everything is installed
python3 -m pip install fastapi uvicornThen run server:
server/run.sh --host localhost --port 8000Tests
To launch tests before new commits:
bash scripts/run_tests.sh /path/to/dir
Docker build
To build a docker for joliGAN server:
docker build -t jolibrain/joligan_build -f docker/Dockerfile.build .
docker build -t jolibrain/joligan_server -f docker/Dockerfile.server .
To run the joliGAN docker:
nvidia-docker run jolibrain/myjoligan
Code format
If you want to contribute please use black code format. Install:
pip install black
Usage :
black .
If you want to format the code automatically before every commit :
pip install pre-commit
pre-commit install
Authors
JoliGAN is created and maintained by Jolibrain.
Our code is inspired by pytorch-CycleGAN-and-pix2pix, CUT and AttentionGAN among others.