CLIP-Driven Fine-grained Text-Image Person Re-identification

Official PyTorch implementation of the paper "CLIP-Driven Fine-grained Text-Image Person Re-identification".

Usage

Requirements

we use single RTX3090 24G GPU for training and evaluation.

Python 3.6.9
pytorch 1.7.0
torchvision 0.8.1
scipy 1.2.1

Dataset Preparation

Download the CUHK-PEDES dataset from here, ICFG-PEDES dataset from here and RSTPReid dataset form here

Organize them in your dataset root dir folder as follows:

|-- your dataset root dir/
|   |-- <CUHK-PEDES>/
|       |-- imgs
|            |-- cam_a
|            |-- cam_b
|            |-- ...
|       |-- reid_raw.json
|
|   |-- <ICFG-PEDES>/
|       |-- imgs
|            |-- test
|            |-- train 
|       |-- ICFG_PEDES.json
|
|   |-- <RSTPReid>/
|       |-- imgs
|       |-- data_captions.json

Data Preparation

1. Run data.sh (or Download from here)
2. Copy files test_reid.json, train_reid.json and val_reid.json to project_directory/cuhkpedes/processed_data/

Training

python train.py 

Testing

python test.py

Acknowledgments

Our code is extended from the following repositories. We sincerely appreciate for their contributions.

• CLIP
• SAF

Citation

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

@article{CFine,
   title={CLIP-Driven Fine-grained Text-Image Person Re-identification}, 
   author={Shuanglin Yan and Neng Dong and Liyan Zhang and Jinhui Tang},
   journal={IEEE Transactions on Image Processing}, 
   year={2023},
   volume={},
   number={},
   pages={1-14},
   doi={10.1109/TIP.2023.3327924}
}

Contact

If you have any question, please feel free to contact us. E-mail: shuanglinyan@njust.edu.cn.