Cycle-Dehaze: Enhanced CycleGAN for Single Image Dehazing

This reposotory is our project for NTIRE 2018 Challenge on Image Dehazing.

Our paper published in CVPR 2018 Workshop (3rd NTIRE). Please cite our paper, if it is helpful for your research.

@inproceedings{engin2018cycle,
  title={Cycle-Dehaze: Enhanced CycleGAN for Single Image Dehazing},
  author={Engin, Deniz and Gen{\c{c}}, An{\i}l and Ekenel, Haz{\i}m Kemal},
  booktitle={The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  year={2018}
}

Model Architecture

Prerequisites

• TensorFlow 1.4.1 or later
• Python 3
• MATLAB

Our code is tested under Ubuntu 16.04 environment with Titan X GPUs.

Demo

• Test the model for Track 1: Indoor

 sh demo.sh data/indoor results/indoor models/Hazy2GT_indoor.pb

• Test the model for Track 2: Outdoor

sh demo.sh data/outdoor results/outdoor models/Hazy2GT_outdoor.pb

• You can use this model for your own images.

sh demo.sh input_folder output_folder model_name

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

The code is based on CycleGAN-TensorFlow implementation.