• Code for paper ``Towards Robust Person Re-identification by Defending Against Universal Attackers" Paper.

• python 3.7.9
• CUDA==11.2
• Market1501,DukeMTMC-reID,MSMT-17,PersonX456,UnrealPerson,RandPerson
• torch==1.3.1
• torchvision==0.2.1

• Market1501,DukeMTMC-reID,MSMT-17,PersonX456 are the same as MetaAttack described
• Download UnrealPerson from link
  • zip unreal_vX.Y and put them to ./data/unrealperson/raw
  • final structure as follows:

.
+-- data 
|   +-- unrealperson
|       +-- images
|       +-- meta.json
|       +-- splits.json
|       +-- raw
|           +-- unreal_vX.Y
|               +-- images

• Download RandPerson(all images) from link
  • only download randperson/images/subet and zip it to ./data/randperson/raw/
  • final structure as follows:

.
+-- data 
|   +-- unrealperson
|       +-- images
|       +-- meta.json
|       +-- splits.json
|       +-- raw
|           +-- randperson_subset
|               +-- randperson_subset

• Download attacked re-ID models from BaiduYun (Password:7q0o)
• Put models under ./pretrained_models

• See runAttackMar.sh for more information

• Preparing perturbation models to ./attackModel, you can pre-download our attacker from BaiduYun (Password:d9bj)
• Preparing corresponding pre-trained model from BaiduYun (Password:7q0o)
• See runDefenseMar.sh for more information

• Using 'resMeta' to create model, then load defense models
• You can download our defense models from BaiduYun (Password:gbot)

Our code is based on MetaAttack, if you use our code, please also cite their paper.

@inproceedings{yang2021learning,
  title={Learning to Attack Real-World Models for Person Re-identification via Virtual-Guided Meta-Learning},
  author={Yang, Fengxiang and Zhong, Zhun and Liu, Hong and Wang, Zheng and Luo, Zhiming and Li, Shaozi and Sebe, Nicu and Satoh, Shin’ichi},
  booktitle={AAAI},
  volume={35},
  number={4},
  pages={3128--3135},
  year={2021}
}

If you find this repo useful for your research, please consider citing the paper

@article{yang2022towards,
  title={Towards Robust Person Re-Identification by Defending Against Universal Attackers},
  author={Yang, Fengxiang and Weng, Juanjuan and Zhong, Zhun and Liu, Hong and Wang, Zheng and Luo, Zhiming and Cao, Donglin and Li, Shaozi and Satoh, Shin'ichi and Sebe, Nicu},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2022},
  publisher={IEEE}
}