Single-image-dehazing-using-improved-cycleGAN

This is the code for the journal paper Single image dehazing using improved cycleGAN published in JVCI.

Acknowledgements

The base code of cycleGAN's implementation in tensorflow has been taken from this Repo by xhujoy. 😊

Dataset

• Download and place the NYU or reside-β dataset's into dataset directory.
• Split the hazy and clear images into train, test sets.
• Rename the directory name containing the training images of hazy, clear images as trainA, trainB respectively.
• Do the same for the corresponding directories containing the test set images.

The final file structure should be as follows

project
│   README.md
│   get_ssim.py
│   main.py
│   model.py
│   module.py
│   ops.py
│   requirements.txt
│   utils.py
│
└───dataset
│   └───<dataset_name>
│       └───trainA
│       |   | img1.jpg
│       |   | img2.jpg
│       |   | ...
│       |
│       └───trainB
│       |   | img1.jpg
│       |   | img2.jpg
│       |   | ...
│       |
│       └───testA
│       |   | img1.jpg
│       |   | img2.jpg
│       |   | ...
│       |
│       └───testB
│           | img1.jpg
│           | img2.jpg
│           | ...
|
└───checkpoint
    └───<dataset_name>_<img size>

The rest of the directories required to store the saved model, sample results and log files will be created automatically when main.py is executed.