We provide the code for reproducing result of our paper Semantically Self-Aligned Network for Text-to-Image Part-aware Person Re-identification.

1. CUHK-PEDES

   Organize them in dataset folder as follows:

   |-- dataset/
   |   |-- <CUHK-PEDES>/
   |       |-- imgs
               |-- cam_a
               |-- cam_b
               |-- ...
   |       |-- reid_raw.json
   

   Download the CUHK-PEDES dataset from here and then run the process_CUHK_data.py as follow:

   cd SSAN
   python ./dataset/process_CUHK_data.py
   

2. ICFG-PEDES

   Organize them in dataset folder as follows:

   |-- dataset/
   |   |-- <ICFG-PEDES>/
   |       |-- imgs
               |-- test
               |-- train 
   |       |-- ICFG_PEDES.json
   

   Note that our ICFG-PEDES is collect from MSMT17 and thus we keep its storage structure in order to avoid the loss of information such as camera label, shooting time, etc. Therefore, the file testand train here are not the way ICFG-PEDES is divided. The exact division of ICFG-PEDES is determined by ICFG-PDES.json. The ICFG-PDES.json is organized like the reid_raw.json in CUHK-PEDES .

   Please request the ICFG-PEDES database from 272521211@qq.com and then run the process_ICFG_data.py as follow:

   cd SSAN
   python ./dataset/process_ICFG_data.py
   

sh experiments/CUHK-PEDES/train.sh 
sh experiments/ICFG-PEDES/train.sh 

sh experiments/CUHK-PEDES/test.sh 
sh experiments/ICFG-PEDES/test.sh 

Our Results on CUHK-PEDES dataset

Our Results on ICFG-PEDES dataset

If this work is helpful for your research, please cite our work:

@article{ding2021semantically,
  title={Semantically Self-Aligned Network for Text-to-Image Part-aware Person Re-identification},
  author={Ding, Zefeng and Ding, Changxing and Shao, Zhiyin and Tao, Dacheng},
  journal={arXiv preprint arXiv:2107.12666},
  year={2021}
}