Dual-Semantic Consistency Learning for Visible-Infrared Person Re-Identification (TIFS 2022)

This repository is an official implementation of DSCNet, a strong baseline for VI-ReID

📈 News

2022.10.24 DSCNet has been formally accepted by IEEE Transactions on Information Forensics & Security.
2022.10.15 Code Release.

🚀 Highlight

1. Insights: This paper derives the modality discrepancy from the channel-level semantic inconsistency. It is the FIRST method to address the limitations on the channel-level representation.
2. A strong baseline: Faster Convergence and Outstanding Performance for VI-ReID.

Model	Training Epochs	Rank-1 (%)	mAP(%)	Training Time
MCLNet	200	65.30	61.59	24 hours
DSCNet	50	73.89	69.47	5 hours

⚙️ Setup environment

• Clone this repo:

git clone https://github.com/bitreidgroup/DSCNet.git && cd DSCNet

• Create a conda environment and activate the environment.

conda env create -f environment.yaml &&  conda activate dsc

We recommend Python = 3.6, CUDA = 10.0, Cudnn = 7.6.5, Pytorch = 1.2, and CudaToolkit = 10.0.130 for the environment.

🔧 Preparing dataset

• SYSU-MM01 Dataset : The SYSU-MM01 dataset can be downloaded from this website.

• We preprocess the SYSU-MM01 dataset to speed up the training process. The identities of cameras will be also stored in ".npy" format.

python utils/pre_process_sysu.py

• RegDB Dataset : The RegDB dataset can be downloaded from this website by submitting a copyright form.

  (Named: "Dongguk Body-based Person Recognition Database (DBPerson-Recog-DB1)" on their website).

⏳ Training

You may need manually define the data path first. More details are in the config files.

• SYSU-MM01 Dataset (all-search)

bash scripts/train_sysu_all.sh

• SYSU-MM01 Dataset (indoor-search)

bash scripts/train_sysu_indoor.sh

💽 Testing

PYTHONPATH="$(dirname $0)/..":$PYTHONPATH python scripts/test.py --ckpt [CKPT_PATH] --config [CONFIT] 

For example : You can test the checkpoints by running the commands below.

bash scripts/eval_sysu.sh

⏰ Reproduce our experimental results

DSCNet: We provide some experimental results on the SYSU-MM01 datasets with pretrained models. These model are trained on 1x 2080ti

config	Rank-1(%)	Rank-10(%)	mAP(%)	Training Log	Pretrained
SYSU-MM01(all-search)	73.89	96.27	69.47	log(TBA)	Weights
SYSU-MM01(indoor-search)	79.35	95.74	82.68	log(TBA)	Weights(TBA)

Before running the commands below, please update the config files on the setting of resume.

python scripts/reproduce.sh

💾GPUs

All our experiments were performed on a single NVIDIA GeForce 2080 Ti GPU

Training Datasets	Approximate GPU memory	Approximate training time
SYSU-MM01	9GB	5 hours
RegDB	6GB	3 hours

Citation

If this repository helps your research, please cite :

@article{zhang2022dual,
  title={Dual-Semantic Consistency Learning for Visible-Infrared Person Re-Identification},
  author={Zhang, Yiyuan and Kang, Yuhao and Zhao, Sanyuan and Shen, Jianbing},
  journal={IEEE Transactions on Information Forensics and Security},
  year={2022},
  publisher={IEEE}
}

📄 References.

1. Y. Zhang, Y. Kang, S. Zhao, and J. Shen. Dual-Semantic Consistency Learning for Visible-Infrared Person Re-Identification. IEEE Transactions on Information Forensics & Security, 2022.
2. M. Ye, W. Ruan, B. Du, and M. Shou. Channel Augmented Joint Learning for Visible-Infrared Recognition. IEEE International Conference on Computer Vision (ICCV), 2021.

✉️ Contact.

If you have some questions, feel free to contact me. [email protected]