Skip to content

Latest commit

 

History

History
24 lines (17 loc) · 1.28 KB

README.md

File metadata and controls

24 lines (17 loc) · 1.28 KB

Model-Inversion-Attack

This a TensorFlow Implementation of the Model Inversion Attack introduced with Model Inversion Attacks that Exploit Confidence Information and Basic Countermeasures (Fredrikson Et al.)

The gradient step and the final output of the attack loop is pre-processed with ZCA whitening and Global Contrast Normalization with Pylearn2, this helps to preserve the facial features present in the input dataset.

Because of tricky dependencies and the need to build pylearn2 yourself, the notebook has been converted to a docker-container. The project can be build like this:

docker build -t mia . -f Dockerfile

and run as a tensorflow/jupyter-notebook:

docker run -it -p 8888:8888 mia

Directions to Use

  1. Download the AT&T Face Dataset from here
  2. Extract the dataset and replace the path variable in the 3rd cell of the inversion notebook.

The dataset is no longer available on the aforementioned website and has been added to this repository. The dataset was preserved and downloaded from Mohamed Elsayed@github.

Thanks to AT&T Laboratories Cambridge for providing this dataset.