Face-Off is a privacy-preserving framework that introduces strategic perturbations to images of the user’s face to prevent it from being correctly recognized. By leveraging adversarial examples generated on faces in a black-box setting, we find that our perturbations transfer to proprietary face recognition APIs such as AWS Rekognition, Azure Face, and Face++.
Using python version 3.5.2
git clone https://github.com/wi-pi/face-off.git
python -m pip install venv
python3.5 -m venv /path/to/new/virtual/environment
source /path/to/new/virtual/environment/bin/activate
pip install -e .
pip install -r requirements.txt
https://drive.google.com/drive/folders/1qE21bgqeCjtqyUrCqehRd8MKt4zGn-i4?usp=sharing
Download each of the weights to the weights folder. These are used for generating adversarial examples and evaluating transferability offline.
Results reported in the paper were obtained using a server with 40 CPU cores, 2 Nvidia TITAN Xp's, and 1 Quadro P6000, 125 GB Memory, Ubuntu version 16.04 LTS, CUDA 10.0, NVIDIA Driver 410.104.
Disclaimer: The code has not yet been tested on a variety of platforms.
See https://stackoverflow.com/questions/50622525/which-tensorflow-and-cuda-version-combinations-are-compatible for CUDA - TensorFlow compatibility.
Note that any API evaluation requires accounts, keys, and an AWS S3 bucket. Below are some links to resources helpful for setting up keys. Follow the step-by-step instructions found in the below links.
AWS S3: https://docs.aws.amazon.com/AmazonS3/latest/dev/access-points.html
AWS S3 - Public Read Access: https://aws.amazon.com/premiumsupport/knowledge-center/read-access-objects-s3-bucket/
Follow the instructions in these 3 links to obtain your public and private keys. Add them to the FaceAPI/credentials.py file.
AWS Rekognition: https://docs.aws.amazon.com/rekognition/latest/dg/getting-started.html
Azure Face: https://azure.microsoft.com/en-us/services/cognitive-services/face/#get-started
Face++: https://console.faceplusplus.com/documents/7079083
./scripts/create_directories.sh
Creates the necessary subdirectories. Code creates subdirectories in data/new_adv_imgs.
./scripts/attack.sh
Generates perturbations on a single image-class pair. Code generates perturbations on faces, and outputs results in data/new_adv_imgs. Successful generation will be printed through each iteration.
./scripts/mask_my_face.sh
Generates perturbations on the set of faces in data/test_imgs/myface/. Code generates perturbations on faces, and outputs results in data/new_adv_imgs. Successful generation will be printed through each iteration.
NOTE: If you want to use hinge loss, you must align (detect, crop, and resize) a bucket of your own faces to sizes 160x160 or 96x96. You can use MTCNN to do so. We will integrate support for this shortly.
./scripts/api_eval.sh
Feeds generated perturbations into the APIs and stores results in data/new_api_results. Success scores will be printed through each iteration.
./scripts/analyze_api_results.sh
Reads and interprets the API scores.
https://arxiv.org/abs/2003.08861
@misc{chandrasekaran2020faceoff,
title={Face-Off: Adversarial Face Obfuscation},
author={Varun Chandrasekaran and Chuhan Gao and Brian Tang and Kassem Fawaz and Somesh Jha and Suman Banerjee},
year={2020},
eprint={2003.08861},
archivePrefix={arXiv},
primaryClass={cs.CR}
}
Questions? Contact [email protected] or [email protected] with subject: Face-Off