Awesome

Optimal Transport for Label-Efficient Visible-Infrared Person Re-Identification (OTLA-ReID)

This is Official Repository for "Optimal Transport for Label-Efficient Visible-Infrared Person Re-Identification" (PDF, Supplementary Material), which is accepted by ECCV 2022. This work is done at the DMCV Laboratory of East China Normal University. You can link at DMCV-Lab to find DMCV Laboratory website page.

main_figure

Update:

[2022-7-17] Semi-supervised setting and supervised setting can be run with current code. Unsupervised setting will be updated with a few of days.

[2022-7-21] Update some critical informtion of REAMDE.md.

[2022-9-22] Update the code of SpCL-master, which can be used to generator pseudo labels of visible modality for unsupervised setting.

[2022-10-28] Update the paper link.

Requirements

python 3.7.11
numpy 1.21.4
torch 1.10.0
torchvision 0.11.0
easydict 1.9
PyYAML 6.0
tensorboardX 2.2

Prepare Datasets

Download the VI-ReID datasets SYSU-MM01 (Email the author to get it) and RegDB (Submit a copyright form). follow the link of DDAG to obtain more information of VI-ReID datasets. Download visible ReID datasets Market-1501, MSMT17 (Email the author to get it), DukeMTMC-reID if you want to run unsupervised setting. Please follow the link of OpenUnReID to obtain more information of visible ReID datasets.

Training

You need to firstly choose the setting: of config file corresponding VI-ReID dataset.

For semi-supervised / supervised setting, if you want to train the model(s) in the paper, run following command:

cd OTLA-ReID/
python main_train.py --config config/config_sysu.yaml

For unsupervised setting, you should write the right path of train_visible_image_path: and train_visible_label_path: , which are the produced visible data and pseudo label path of VI-ReID datasets by well-established UDA-ReID or USL-ReID methods (e.g. SpCL). Then run following command:

cd OTLA-ReID/
python main_train.py --config config/config_sysu.yaml

Here, we give an example of running SpCL to generate visible pseudo label in SpCL-master. However, you firstly need to install environment which can be found in SpCL:

For SYSU-MM01:

cd OTLA-ReID/SpCL-master/
CUDA_VISIBLE_DEVICES=0,1,2,3 python examples/spcl_train_uda.py -ds market1501 -dt sysumm01_rgb --logs-dir logs/spcl_uda/market1501TOsysumm01_rgb_resnet50 --epochs 51 --iters 800

For RegDB:

cd OTLA-ReID/SpCL-master/
CUDA_VISIBLE_DEVICES=0,1,2,3 python examples/spcl_train_uda.py -ds market1501 -dt regdb_rgb --logs-dir logs/spcl_uda/regdbTOsysumm01_rgb_resnet50 --epochs 51 --iters 50

The generated visible images and visible pseudo labels are both saved under the dataset directory.

Testing

If you want to test the trained model(s), run following command:

cd OTLA-ReID/
python main_test.py --config config/config_sysu.yaml --resume --resume_path ./sysu_semi-supervised_otla-reid/sysu_save_model/best_checkpoint.pth

Citation

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

@inproceedings{wang2022optimal,
  title={Optimal Transport for Label-Efficient Visible-Infrared Person Re-Identification},
  author={Wang, Jiangming and Zhang, Zhizhong and Chen, Mingang and Zhang, Yi and Wang, Cong and Sheng, Bin and Qu, Yanyun and Xie, Yuan},
  booktitle={European Conference on Computer Vision},
  pages={93--109},
  year={2022},
  organization={Springer}
}

Acknowledgements

This work is developed based on repositories of SeLa(ICLR 2020), DDAG(ECCV 2020), SpCL(NIPS 2020), MMT(ICLR 2020), HCD(ICCV 2021). We sincerely thanks all developers of these high-quality repositories.