Source Code is released for the paper entitled Unsupervised Person Re-identification via Multi-domain Joint Leatning.
- Python 3.6
- GPU Memory >= 20G
- Numpy
- Pytorch 0.4+
Preparation 1: create folder for dataset.
first, download Market-1501 and DukeMTMC-reID dataset from the links below:
google drive: https://drive.google.com/file/d/0B8-rUzbwVRk0c054eEozWG9COHM/view?usp=sharing https://drive.google.com/open?id=1jjE85dRCMOgRtvJ5RQV9-Afs-2_5dY3O baidu disk: https://pan.baidu.com/s/1ntIi2Op https://pan.baidu.com/s/1jS0XM7Var5nQGcbf9xUztw
second,
mkdir data
unzip Market-1501-v15.09.15.zip
ln -s Market-1501-v15.09.15 market
unzip DukeMTMC-reID.zip
ln -s DukeMTMC-reID duke
then, get the directory structure
├── MDJL
├── data
├── market
├── Market-1501-v15.09.15
├── duke
├── DukeMTMC-reID
Preparation 2: Put the images with the same id in one folder. You may use
python prepare.py
Finally, train, test and evaluate the re-ID model with the below command:
python train.py
If you refer to this code, please cite our paper as follows: @article{CHEN2023109369, title = {Unsupervised person re-identification via multi-domain joint learning}, journal = {Pattern Recognition}, volume = {138}, pages = {109369}, year = {2023}, issn = {0031-3203}, doi = {https://doi.org/10.1016/j.patcog.2023.109369}, url = {https://www.sciencedirect.com/science/article/pii/S0031320323000705}, author = {Feng Chen and Nian Wang and Jun Tang and Pu Yan and Jun Yu}, }