maxingan2412/pass-reid

[ECCV2022] PASS: Part-Aware Self-Supervised Pre-Training for Person Re-Identification

PASS: Part-Aware Self-Supervised Pre-Training for Person Re-Identification [pdf]

The official repository for PASS: Part-Aware Self-Supervised Pre-Training for Person Re-Identification ECCV2022.

Requirements

Installation

pip install -r requirements.txt

We recommend to use /torch=1.8.0 /torchvision=0.9.0 /timm=0.3.4 /cuda>11.1 /faiss-gpu=1.7.2/ A100 for training and evaluation. If you find some packages are missing, please install them manually. You can refer to DINO, TransReID and cluster-contrast-reid to install the environment of pre-training, supervised ReID and unsupervised ReID, respectively. You can also refer to TransReID-SSL to install the whole environments.

Prepare Datasets

mkdir data

Download the datasets:

• Market-1501
• MSMT17
• LUPerson. We don't have the copyright of the LUPerson dataset. Please contact authors of LUPerson to get this dataset.

Then unzip them and rename them under the directory like

data
├── market1501
│   └── bounding_box_train
│   └── bounding_box_test
│   └── ..
├── MSMT17
│   └── train
│   └── test
│   └── ..
└── LUP
    └── images 

Pre-trained Models

Model	Download
ViT-S/16	link
ViT-B/16	link

Please download pre-trained models and put them into your custom file path.

ReID performance

We have reproduced the performance to verify the reproducibility. The reproduced results may have a gap of about 0.1~0.2% with the numbers in the paper.

Supervised ReID

Market-1501

Model	Image Size	mAP	Rank-1	Download
ViT-S/16	256*128	92.2	96.3	model / log
ViT-S/16	384*128	92.6	96.8	model / log
ViT-B/16	256*128	93.0	96.8	model / log
ViT-B/16	384*128	93.3	96.9	model / log

MSMT17

Model	Image Size	mAP	Rank-1	Download
ViT-S/16	256*128	69.1	86.5	model / log
ViT-S/16	384*128	71.7	87.9	model / log
ViT-B/16	256*128	71.8	88.2	model / log
ViT-B/16	384*128	74.3	89.7	model / log

UDA ReID

MSMT2Market

Model	Image Size	mAP	Rank-1	Download
ViT-S/16	256*128	90.2	95.8	model / log

Market2MSMT

Model	Image Size	mAP	Rank-1	Download
ViT-S/16	256*128	49.1	72.7	model / log

USL ReID

Market-1501

Model	Image Size	mAP	Rank-1	Download
ViT-S/16	256*128	88.7	95.0	model / log

MSMT17

Model	Image Size	mAP	Rank-1	Download
ViT-S/16	256*128	41.0	67.0	model / log

Acknowledgment

Our implementation is mainly based on the following codebases. We gratefully thank the authors for their wonderful works.

TransReID-SSL, LUPerson, DINO, TransReID, cluster-contrast-reid.

Citation

If you find this code useful for your research, please cite our paper

@article{zhu2022part,
  title={PASS: Part-Aware Self-Supervised Pre-Training for Person Re-Identification},
  author={Zhu, Kuan and Guo, Haiyun and Yan, Tianyi and Zhu, Yousong and Wang, Jinqiao and Tang, Ming},
  journal={arXiv preprint arXiv:2203.03931},
  year={2022}
}

Contact

If you have any question, please feel free to contact us. E-mail: [email protected]. 环境配置 如果发生 No module named 'torch._six' 进行如下替换

#from torch._six import container_abcs TORCH_MAJOR = int(torch.version.split('.')[0]) TORCH_MINOR = int(torch.version.split('.')[1]) if TORCH_MAJOR == 1 and TORCH_MINOR < 8: from torch._six import container_abcs else: import collections.abc as container_abcs

环境 1.conda建立一个 3.10的python虚拟环境 2.pip3 install torch torchvision torchaudio 3.pip install -r requ.txt