Cross-Modality Perturbation Synergy Attack for Person Re-identification

Contribution

Finds transferable perturbations that break person re-ID across RGB and infrared.

Abstract

We study cross-modality person re-identification attack and formulate universal perturbation crafting as a synergy-driven optimization problem. By combining gradients from RGB and infrared modalities with a grayscale bridge, the method learns a transferable perturbation that degrades re-ID performance across modalities while remaining visually subtle.

BibTeX

@inproceedings{gong2024cmps,
  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},
  booktitle={Advances in Neural Information Processing Systems 37},
  year={2024},
  doi={10.52202/079017-0735},
}