PatchCURE: Improving Certifiable Robustness, Model Utility, and Computation Efficiency of Adversarial Patch Defenses

Code for "PatchCURE: Improving Certifiable Robustness, Model Utility, and Computation Efficiency of Adversarial Patch Defenses" in USENIX Security Symposium 2024.

Update 04/2024: Released code and model checkpoints.

Overview

Main results

Check out our paper list for adversarial patch research and leaderboard for certifiable robust image classification for fun!

Dependency

Tested with torch==1.13.1 and timm==0.9.16. This repository should be compatible with newer version of packages. requirements.txt lists other required packages (with version numbers commented out).

Files

├── README.md                        # this file 
├── requirements.txt                 # required packages
├── example_cmds.sh                  # command to reproduce PatchCURE results reported in the paper
├── reproducibility.md               # detailed guide for experiments. used for artifact evaluation at USENIX Security 
├── get_imagenet_val.sh              # script for downloading ImageNet val dataset.  
| 
├── main.py                          # PatchCURE entry point.  
| 
├── utils
|   ├── builder.py                   # utils for building models and getting data loaders
|   ├── pcure.py                     # utils for PatchCURE inference algorithms 
|   ├── split.py                     # utils for splitting models into sub-models for PatchCURE construction
|   ├── bagnet.py                    # BagNet model; adapted from https://github.com/wielandbrendel/bag-of-local-features-models/blob/master/bagnets/pytorchnet.py
|   └── vit_srf.py                   # ViT-SRF model; based on timm/models/vision_transformer.py 
|
| 
├── data   
|   └── imagenet                     # data directory for imagenet # use torchvision.datasets.ImageFolder
|
└── checkpoints                      # directory for checkpoints
    ├── README.md                    # details of checkpoints
    └── ...                          # model checkpoints

Dataset

• ImageNet (ILSVRC2012).

Getting Started

1. See Files for details of each file.
2. Download data in Datasets to data/.
3. Read checkpoints/README.md and download checkpoints from Google Drive link and move them to checkpoints.
4. See example_cmd.sh for example commands for running the code and reproducing PatchCURE results reported in the paper.
5. See reproducibility.md for a more detailed guide for running experiments. It was used for artifact evaluation in USENIX Security.

If anything is unclear, please open an issue or contact Chong Xiang ([email protected]).

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{xiang2024patchcure,
  title={PatchCURE: Improving Certifiable Robustness, Model Utility, and Computation Efficiency of Adversarial Patch Defenses},
  author={Xiang, Chong and Wu, Tong and Dai, Sihui and Petit, Jonathan and Jana, Suman and Mittal, Prateek},
  booktitle = {33rd {USENIX} Security Symposium ({USENIX} Security)},
  year={2024}
}