/TASTR

The implementation code and models of TASTR.

Primary LanguagePythonMIT LicenseMIT

Progressive Unsupervised Person Re-identification by Tracklet Association with Spatio-Temporal Regularization [link]

The implementation code and models of TASTR.

Environment

  • Ubuntu 16.04
  • Python 3.7
  • Pytorch 1.0.1
  • CUDA 9.0
  • NVIDIA GTX 1080ti x 2

Installation

git clone https://github.com/xieqk/TASTR.git

Prepare dataset

Download DukeMTMC-reID dataset then extract file to your data directory ( denoted as ${data_root} ). The file structure is as follows

${data_root}
└── dukemtmc-reid/
    └── DukeMTMC-reID
        ├── bounding_box_test
        ├── bounding_box_train
        ├── CITATION.txt
        ├── LICENSE_DukeMTMC-reID.txt
        ├── LICENSE_DukeMTMC.txt
        ├── query
        └── README.md

Training

modify the ${data_root} in ./scripts/train_dukemtmcreid.sh, then run

./scripts/train_dukemtmcreid.sh

Citation

@article{xie2020progressive,
  title={Progressive Unsupervised Person Re-identification by Tracklet Association with Spatio-Temporal Regularization},
  author={Xie, Qiaokang and Zhou, Wengang and Qi, Guo-Jun and Tian, Qi and Li, Houqiang},
  journal={IEEE Transactions on Multimedia (TMM)},
  year={2020},
  publisher={IEEE}
}

Request the Campus4K dataset from xieqiaok [at] mail.ustc.edu.cn.