Adversarial YOLO

This repository was originally based on the marvis YOLOv2 inplementation: https://github.com/marvis/pytorch-yolo2

This work corresponds to the following paper: https://arxiv.org/abs/1904.08653:

@inproceedings{thysvanranst2019,
    title={Fooling automated surveillance cameras: adversarial patches to attack person detection},
    author={Thys, Simen and Van Ranst, Wiebe and Goedem\'e, Toon},
    booktitle={CVPRW: Workshop on The Bright and Dark Sides of Computer Vision: Challenges and Opportunities for Privacy and Security},
    year={2019}
}

If you use this work, please cite this paper.

We at KSU have now adapted this repository to use YOLOv8 rather than YOLOv2 for the inference implementation

What you need

We use Python 3.10. Make sure that you have a working implementation of PyTorch installed, to do this see: https://pytorch.org/

This project has prerequisite packages that must be installed using pip before use:

pip install -r requirements.txt

No other installation for this project is necessary, you can simply run the python code straight from this directory.

(no longer used) original YOLOv2 model weights used on INRIA dataset

mkdir weights; curl https://pjreddie.com/media/files/yolov2.weights -o weights/yolo.weights

(no longer used) Get the INRIA dataset:

curl ftp://ftp.inrialpes.fr/pub/lear/douze/data/INRIAPerson.tar -o inria.tar
tar xf inria.tar
mv INRIAPerson inria
cp -r yolo-labels inria/Train/pos/

Generating a patch

patch_config.py contains configuration of different experiments. You can design your own experiment by inheriting from the base BaseConfig class or an existing experiment. ReproducePaperObj reproduces the patch that minimizes object score from the paper (With a lower batch size to fit on a desktop GPU).

Here is a config for patch training on a small subset of satellite imagery of aircraft dataset:

python train_patch.py airbus-subset-8

Here is another config for patch training (not working currently):

python train_patch.py aircraft