Single Image Dehazing Using Deep Learning

This repository contains the deep learning models used in the DeepDive: An End-to-End Dehazing Method Using Deep Learning paper.

The datasets used to train this models are:

• NYU Depth V2
• Reside

Citation

If you use this model in your research, please cite:

@inproceedings{goncalves2017deepdive,
    title={DeepDive: An End-to-End Dehazing Method Using Deep Learning},
    author={Goncalves, Lucas Teixeira and Gaya, Joel De Oliveira and Drews, Paulo and Botelho, Silvia Silva Da Costa},
    booktitle={Graphics, Patterns and Images (SIBGRAPI), 2017 30th SIBGRAPI Conference on},
    pages={436--441},
    year={2017},
    organization={IEEE}
}

Prerequisites

To run this model, you will need:

• Tensorflow
• Python Imaging Library (PIL)
• Numpy

Running the model

To run this model, simply run the main.py python code.

Arguments:

• -h, --help: View the help message and exit
• -m, --mode MODE: Specify one of the possible modes:
  • train
  • evaluate
  • restore
  • dataset_manage
• -a, --architecture ARCHITECTURE: Specify the architecture used in the model
• -d, --dataset DATASET: Specify the dataset implementation used to train the model
• -l, --loss LOSS: Specify the loss implementation used to train the model
• -o, --optimizer OPTIMIZER: Specify the optimizer implementation used to train the model
• -g, --dataset_manager DATASET_MANAGER
• -e, --evaluate EVALUATE
• --evaluate_path EVALUATE_PATH
• -p, --execution_path EXECUTION_PATH

The items highlighted in bold are obligatory for all modes except dataset_manage.