FilterFool

This is the official repository of Semantically Adversarial Learnable Filters.

Example of results

Original Image	Adversarial Image	Original Image	Adversarial Image	Original Image	Adversarial Image
macaw	Irish setter	crane	mower	Irish terrier	orang

Setup

1. Create conda virtual-environment

    module load python3/anaconda
    conda create --name FilterFool python=3.6.8
   

2. Activate conda environment

    conda activate FilterFool
   

3. Extract the tar file

   tar -zxvf https://github.com/AliShahin/FilterFool.git
   

4. Go to the working directory

   cd FilterFool_code
   

5. Install requirements (please make sure your GPU is enabled)

    pip install -r requirements.txt
   

Generate adversarial images

1. In the script.sh set the desired filter among "Nonlinear_detail", "Gamma" or "Log"

2. Generate the FilterFool adversarial image

   bash script.sh
   

3. The FilterFool adversarial image and log file are stored in the Results_{filter} (within the root directory) with the same name as their corresponding original images

Authors

• Ali Shahin Shamsabadi
• Changjae Oh
• Andrea Cavallaro

References

If you use our code, please cite the following paper:

  @article{shamsabadi2021filterfool,
    title = {Semantically Adversarial Learnable Filters},
    author = {Shamsabadi, Ali Shahin and Oh, Changjae and Cavallaro, Andrea},
    journal={arXiv preprint arXiv:2008.06069},
    year = {2021}
  }

License

The content of this project itself is licensed under the Creative Commons Non-Commercial (CC BY-NC).