This repository holds the code for our paper [PAPER-NAME] by Andrew Wang, Gopal Nookula, Wyatt Mayor and Ryan Smith. The defenses are based off the adversarial attacks against road sign recognition models in [THIS PAPER] by Zhong et. al (2022). Our defenses achieve 78% robustness with a simple edge profile channel retrained on the paper's original CNN. Please see our paper [HERE] for more details.

⚠️ Note that for AndrewNet, requirements.txt was generated assuming an Anaconda installation. Please read the instructions inside requirements.txt for usage instructions.

Note that a number of files are stored with git-lfs, so you should first install git lfs and then enter git lfs install into the root directory of this repository.

Setup instructions can be found in ./AndrewNet/instructions.md.

The

Please see ./AndrewNet/instructions.md for attributions to others' work.