The official repository for MDReID: Modality-Decoupled Learning for Any-to-Any Multi-Modal Object Re-Identification [pdf]

mkdir data

Download the person datasets RGBNT201, RGBNT100 (code：rjin), and the MSVR310.

pip install -r requirements.txt

You need to download the pretrained CLIP model: ViT-B-16 (Code：52fu)

You can train the MDReID with:

python train_net.py --config_file configs/RGBNT201/MDReID.yml

Some examples:

python train_net.py --config_file configs/RGBNT201/MDReID.yml

1. The device ID to be used can be set in config/defaults.py

2. If you need to train on the RGBNT100 and MSVR310 datasets, please ensure the corresponding path is modified accordingly.

python test_net.py --config_file 'choose which config to test' --model_path 'your path of trained checkpoints'

Some examples:

python test_net.py --config_file configs/MSVR310/MDReID.yml --model_path MSVR310_MDReIDbest.pth

Please kindly cite this paper in your publications if it helps your research:

@inproceedings{feng2025mdreid,
  title={MDReID: Modality-Decoupled Learning for Any-to-Any Multi-Modal Object Re-Identification},
  author={Yingying, Feng and Jie, Li and Jie, Hu and Yukang, Zhang and Lei, Tan and Jiayi, Ji},
  booktitle={Advances in Neural Information Processing Systems},
  year={2025}
}

Our code is based on TOP-ReID[1]

[1]Wang Yuhao, Liu Xuehu, Zhang Pingping, Lu Hu, Tu Zhengzheng, Lu Huchuan. 2024. TOP-ReID: Multi-Spectral Object Re-identification with Token Permutation. Proceedings of the AAAI Conference on Artificial Intelligence. 38. 5758-5766.

If you have any questions, please feel free to contact us. E-mail: tanlei@stu.xmu.edu.cn