Code for the NeurIPS 2024 paper ``Cross-Modality Perturbation Synergy Attack for Person Re-identification".

Paper

Requirements:

• python 3.7
• CUDA==11.2
• faiss-gpu==1.6.0

Preparing Data

• There is a processed tar file in BaiduYun (Password: 9pmk) with all needed files.

Run our code

See run.sh for more information.

If you find this code useful in your research, please cite:

@article{gong2024cross,
	title={Cross-modality perturbation synergy attack for person re-identification},
	author={Gong, Yunpeng and Zhong, Zhun and Qu, Yansong and Luo, Zhiming and Ji, Rongrong and Jiang, Min},
	journal={Advances in Neural Information Processing Systems},
	volume={37},
	pages={23352--23377},
	year={2024}
}

Acknowledgments

The code is based on LTA, Random Color Erasing and Mata_attack.

If you use the code, please cite the paper.

@inproceedings{colorAttack2022,
  title={Person re-identification method based on color attack and joint defence},
  author={Gong, Yunpeng and Huang, Liqing and Chen, Lifei},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4313--4322},
  year={2022}
}

@inproceedings{gong2026theory,
  title={A theory-inspired framework for few-shot cross-modal sketch person re-identification},
  author={Gong, Yunpeng and Hou, Yongjie and Shi, Jiangming and Diep, Kim Long and Jiang, Min},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={40},
  number={6},
  pages={4284--4292},
  year={2026}

Question

1.Recently, some users have reported that running the same code now produces results that differ significantly from those obtained a few months ago. We suspect this is most likely caused by updates to packages or the runtime environment. It is a common issue that significant discrepancies in results arise due to updates in the code package and its environment. At the moment, there is no good solution, and the author does not have the time or capacity to maintain it. Please make sure that your environment and package versions are configured according to the versions used when the Mata_attack code was released.

2.In addition, questions regarding the model accuracy of AGW and DDAG are frequently asked. For such inquiries, it is recommended to consult the respective authors of the relevant papers.

Contact Me

Email: fmonkey625@gmail.com