TaintRadar

Official Implementation of Detecting Localized Adversarial Examples: A Generic Approach using Critical Region Analysis published at IEEE International Conference on Computer Communications (INFOCOM) 2021.

Installation

Please download the code:

To use our code, first download the repository:

git clone https://github.com/FengtingLI/TaintRadar.git

To install the dependencies:

conda create -n taint_radar -y python=3.7 tensorflow-gpu=1.13 keras=2.3.1
conda activate taint_radar
pip install -r requirements.txt

Running

We provided an example of our method. The resources (model and images) can be found here (google drive). Download the .zip files and extract them directly under the main folder, like this:

--> TaintRadar
    --> models
        --> vgg16.h5
    --> images
        --> *.png   # The attacked images
        --> *_origin.png # The corresponding original images

Then, run the following code:

python run.py

Citation

If you find this code useful, please consider citing the following paper:

@inproceedings{li2021detecting,
  title     = {Detecting Localized Adversarial Examples: A Generic Approach using Critical Region Analysis},
  author    = {Fengting Li, Xuankai Liu, Xiaoli Zhang, Qi Li, Kun Sun, Kang Li},
  booktitle = {{IEEE} Conference on Computer Communications, {INFOCOM}},
  year      = {2021}
}