Generic Model-Agnostic Convolutional Neural Network(GMAN) is a convolutional neural network proposed for haze removal and clear image restoration. It is an end-to-end deep learning system that employs the encoder-decoder network for denoising image. I've used Kaggle notebook for the purpose of implementation and training. Dataset used for training and validation is SOTS outdoor available here.

Detailed explanation and documentation here. Modified model along with web app and deployment code will be there soon.

Note: Incase notebook is not loading on GitHub, you can check notebook with validation output upto 10 epochs here.

• Python(3.6+)
• Tensorflow(2+)
• GPU: Nvidia Tesla P100(provided by Kaggle)

1. Download the saved model.
2. pip install -r requirements.txt
3. Give model path and image path to test.py and run.
   (Note: Saved model folder, test.py and images should be in same folder.)

I've used naturally hazed images downloaded randomly from google and some images are from dataset. You can see the dehazed test images against hazy images here, some of them are below. Dehazed test images with good resolution are available here.

@article{liu2019single,
  title={Single Image Dehazing with a Generic Model-Agnostic Convolutional Neural Network},
  author={Liu, Zheng and Xiao, Botao and Alrabeiah, Muhammad and Wang, Keyan and Chen, Jun},
  journal={IEEE Signal Processing Letters},
  volume={26},
  number={6},
  pages={833--837},
  year={2019},
  publisher={IEEE}
}

https://github.com/Seanforfun/GMAN_Net_Haze_Removal