UCR

Implementation of paper "Unsupervised Lifelong Person Re-identification via Contrastive Rehearsal".

Installation

conda create -n env_ucr python=3.6
source activate env_ucr 
pip install numpy torch==1.4.0 torchvision==0.5.0 h5py six Pillow scipy sklearn metric-learn tqdm faiss-gpu==1.6.3
python setup.py develop

Prepare Datasets

cd examples && mkdir data

Download the raw datasets Market-1501, Cuhk-Sysu, MSMT17, VIPeR, PRID2011, GRID, iLIDS, CUHK01, CUHK02, SenseReID, CUHK03 and 3DPeS, and then unzip them under the directory like

UCR/examples/data
├── market1501
│   ├── bounding_box_train/
│   ├── bounding_box_test/
│   └── query/
├── cuhk-sysu
│   └── CUHK-SYSU
│       ├── Image/
│       └── annotation/
├── msmt17
│   └── MSMT17_V2
├── viper
│   └── VIPeR
├── prid2011
│   └── prid_2011
├── grid
│   └── underground_reid
├── ilids
│   └── i-LIDS_Pedestrian
├── cuhk01
│   └── campus
├── cuhk02
│   └── Dataset
├── sensereid
│   └── SenseReID
├── cuhk03
│   └── cuhk03_release
└── 3dpes
    └── 3DPeS

Train:

Train UCR on default order (Market to Cuhk-Sysu to MSMT17). The results reported in the paper were obtained with 4 GPUs.

Unsupervised lifelong training

sh unsupervised_lifelong.sh

Supervised lifelong training

sh supervised_lifelong.sh

Test:

python examples/test.py --init examples/logs/step3.pth.tar

Citation

If you find this project useful, please kindly star our project and cite our paper.

@article{chen2022unsupervised,
  title={Unsupervised Lifelong Person Re-identification via Contrastive Rehearsal},
  author={Chen, Hao and Lagadec, Benoit and Bremond, Francois},
  journal={arXiv preprint arXiv:2203.06468},
  year={2022}
}