MartinHahner / FoggySynscapes

Semantic Understanding of Foggy Scenes with Purely Synthetic Data

Forked from github.com/sakaridis/fog_simulation-SFSU_synthetic.
Adapted by Martin Hahner at the Computer Vision Lab of ETH Zurich.

Introduction

This is the source code we used in our ITSC 2019 paper titled
Semantic Understanding of Foggy Scenes with Purely Synthetic Data,
to create Foggy Synscapes, a foggy version of the original Synscapes dataset.

The pipeline computes transmittance maps,
which can be combined with the original clear-weather image
to obtain foggy versions of the clear-weather image.

Pipeline Overview

Getting Started

This demo runs our fog simulation pipeline on an example clear-weather image from Synscapes and writes the results to ./output/demo/.

Requirements

• MATLAB
  The code has been developed in MATLAB R2018b and tested in MATLAB 2019a, too.

How to run the demo

1. Clone this repository

   git clone https://github.com/MartinHahner88/FoggySynscapes.git
   

2. Go to MATLAB's Command Window and assign FOG_SIMULATION_ROOT to the path you have cloned this repository into

   FOG_SIMULATION_ROOT = '/path/to/FoggySynscapes';
   

3. Change MATLAB's current folder

   cd(fullfile(FOG_SIMULATION_ROOT, 'source'));
   

4. Run Demo.m

   Demo;
   

Generating Foggy Synscapes all by yourself

Due to the license agreement of the Synscapes dataset,
we unfortunatelly cannot directly distribute Foggy Synscapes.

But if you follow the instructions below, you can reproduce Foggy Synscapes all by yourself. *

* given you have access to the original Synscapes dataset

Requirements

• Synscapes
  You need the original dataset as a starting point.

• Python
  The code has been developed and tested using Python 3.7.4.

• MATLAB
  The code has been developed in MATLAB R2018b and tested in MATLAB 2019a, too.

How to reproduce Foggy Synscapes:

1. Request the original Synscapes dataset by sending a kind email to the authors.

2. Then you need to convert the .exr depth files provided by the Synscapes dataset to .mat depth files that are compatible to our fog simulation pipeline. You can do this for example by using the provided exr_to_mat.py python script.

3. Finally, you can modify FoggySynscapes.m according to your needs and generate your own version of
   Foggy Synscapes.

Note: In our paper we used beta values of [0.005, 0.01, 0.02, 0.03, 0.06].

Citations

If you use our work, please cite our publication and the Synscapes White Paper:

@inproceedings{FoggySynscapes,
  author    = {Hahner, Martin and Dai, Dengxin and Sakaridis, Christos and Zaech, Jan-Nico and Van Gool, Luc},
  title     = {Semantic Understanding of Foggy Scenes with Purely Synthetic Data},
  booktitle = {IEEE International Conference on Intelligent Transportation Systems (ITSC)}, 
  year      = {2019},
}

@article{Synscapes,
    author    = {Magnus Wrenninge and Jonas Unger},
    title     = {Synscapes: A Photorealistic Synthetic Dataset for Street Scene Parsing},
    url       = {http://arxiv.org/abs/1810.08705},
    year      = {2018},
}

License

Our fog simulation pipeline is made available for non-commercial use only.
For details, please refer to the license agreement.

Contact

Martin Hahner
martin.hahner[at]vision.ee.ethz.ch
https://www.trace.ethz.ch/foggy_synscapes