Moire-Backdoor-Attack (MBA)

Official Pytorch implementation for our ACM MM 2023 paper: Moiré Backdoor Attack (MBA): A Novel Trigger for Pedestrian Detectors in the Physical World

Samples

Preparation

• Python
• Pytorch
• YOLOv5
• MMDetection

Usage

Please download the COCO2017 dataset in this link and the OCHuman in this link.

• Step 0:
  Creating COCOPerson and OCHuman datasets in YOLO format, complete with mask labels, based on the aforementioned two datasets.

• Step 1:
  Generating poisoned data samples.

  python moire2img.py --source imageDir

• Step 2:
  Generating poisoned dataset and put in the dataset folder.

  python move2dataset.py

• Step 3:
  Train on the poisoned dataset using the corresponding scripts of the two libraries YOLOv5 and MMDetection.

Citation

If you find the papers are useful for your research, please cite our paper as follows:

@inproceedings{wei2023moire,
  title={Moir{\'e} Backdoor Attack (MBA): A Novel Trigger for Pedestrian Detectors in the Physical World},
  author={Wei, Hui and Yu, Hanxun and Zhang, Kewei and Wang, Zhixiang and Zhu, Jianke and Wang, Zheng},
  booktitle={Proceedings of the 31st ACM International Conference on Multimedia},
  pages={8828--8838},
  year={2023}
}

Acknowledgements

We would like to acknowledge the YOLOv5 open-source library (https://github.com/ultralytics/yolov5) and MMDetection open-source library (https://github.com/open-mmlab/mmdetection).