porky12 / cyclegan_pytorch

pytroch implementation of CycleGan used for rain removal

CycleGAN PyTorch

What is this repository for?

Implementation of CycleGan model in PyTorch (original implementation link). The implementation is used to remove rain from rainy images.

How do I get set up ?

Step by step:

Install:

• PyTorch and dependencies
• Torch vision
• visdom and dominate

git clone https://github.com/pytorch/vision
cd vision
python setup.py install

pip install visdom
pip install dominate

How do I train CycleGAN with new images ?

you may have information on how to run train.py by:

python train.py --help

you can train your own model by running (N.B.: example):

python train.py --dataroot ./data --name cyclegan_custom --model cycle_gan --no_dropout

How do I test CycleGAN after training?

you can test the model on a given collection, in order to transform A to B or B to A (Possible only after training).

python test.py --dataroot ./data --name cyclegan_custom --model cycle_gan --no_dropout --phase test --results_dir ./result_folder

Contents

└── cyclegan
    ├── data                          # data folder contaning both A and B images
         ├── testA                    # test images belonging to class A
         ├── testB                    # test images belonging to class B
         ├── trainA                   # train images belonging to class A
         └── trainB                   # train images belonging to class B
    ├── images                        # images ... 
    ├── models
        └── ...                       # cycle gan model implementation .py
    ├── options                      
        └── ...                       # options : base, train, test .py
    ├── util    
        └── ...                       # utils .py               
    ├── test.py                       # to test
    ├── train.py                      # to train
    ├── README.md                     # Readme

Demonstration: De-raining images

The example below present 12 rainy images where cycleGAN has been used to de-rain.

Acknowledgement

Based on two implementations:

• https://github.com/junyanz/CycleGAN by Jun-Yan Zhu and Taesung Park
• https://github.com/EliasVansteenkiste/CycleGANwithPerceptionLoss by Elias Vansteenkiste