ScAR: Scaling-Adversarial-Robustness-for-LiDAR-Object-Detection

Official implementation of the IROS2023 paper "ScAR: Scaling Adversarial Robustness for LiDAR Object Detection". https://arxiv.org/abs/2312.03085.

Prerequisites:

1. PCDet https://github.com/open-mmlab/OpenPCDet

Usage:

1. Follow the instruction of PCDet to install the pre-requirements.
2. Choose a 3D detector, and train it on a dataset.
3. Use the attack.py file to generate three types of adversarial attacks: model-aware attack, distribution-aware attack, and blind attack.
4. Use the scar.py file to generate adversarial robust training samples.
5. Test the baseline and the scar-trained baseline on three types of attacked dataset.

Performance:

Notice: attack.py and scar.py can be applied to any baselines, you can simply modified the code to adapt to your method.

Citation:

Please cite the following paper if this you feel this code helpful.

@inproceedings{lu2023scar,
  title={ScAR: Scaling Adversarial Robustness for LiDAR Object Detection},
  author={Lu, Xiaohu and Radha, Hayder},
  booktitle={2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={5758--5764},
  year={2023},
  organization={IEEE}
}