Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection (NeurIPS 2024)

Project | Paper

This repository enables the generation of adversarial patches that remain effective under multiple camera devices.

Hui Wei¹, Zhixiang Wang², Kewei Zhang¹, Jiaqi Hou¹, Yuanwei Liu¹, Hao Tang³, Zheng Wang¹

¹Wuhan University,  ²University of Tokyo,  ³Peking University.

Update

• 2024.10.10: Repo is released.
• 2024.09.26: Paper is accepted to NeurIPS 2024.

🧰 Setup

1. Clone this repo:

git clone https://github.com/weihui1308/CAP.git
cd CAP

2. Install dependencies:

pip install -r requirements.txt

3. Download the weight files:

• Finetuned YOLOv5 model: finetune_yolov5s_onINRIA.pt
• Camera ISP proxy network model: checkpoint_ISPNet.pth

🌈 Training and Evaluation

1. Optimize the adversarial patch:

python train.py --data config/data_config.yaml --weights checkpoints/finetune_yolov5s_onINRIA.pt --batch_size 32 --epochs 1000

2. Evaluate a given adversarial patch:

python val.py --patch_name patch/onePatch.png

Citation

If you find our work useful, please kindly cite as:

@inproceedings{wei2024cap,
      title={Revisiting Adversarial Patches for Designing Camera-Agnostic Attacks against Person Detection},
      author={Wei, Hui and Wang, Zhixiang and Zhang, Kewei and Hou, Jiaqi and Liu, Yuanwei and Tang, Hao and Wang, Zheng},
      booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
      year={2024}
    }