OCLReID

This project is for target person tracking based on mmtrack framework. For running this code with robot/rosbag, please refer to OCL-RPF

Install

For Video Running Only

Create a conda environment and install OCLReID (based on mmtrack), worked in RTX3090

git clone https://github.com/MedlarTea/OCLReID
cd OCLReID
conda create -n oclreid python=3.7
conda activate oclreid
conda install pytorch=1.11 cudatoolkit=11.3 torchvision=0.12.0 -c pytorch
pip install mmcv-full==1.5.3 -f https://download.openmmlab.com/mmcv/dist/cu113/torch1.11.0/index.html
pip install mmdet==2.26.0
pip install -r requirements.txt
pip install -r requirements/build.txt
pip install -v -e .

# install orientation estimation method
cd mmtrack/models/orientation
pip install -r requirements.txt
pip install -v -e .

Download pre-trained weights for OCLReID

• Download 2d joint detection models: Google drive and put the checkpoints to OCLReID/mmtrack/models/pose/Models/sppe.
• Download ReID models: Google drive, then make directory OCLReID/checkpoints/reid and put the checkpoints to it.

Run It!

Video Running

cd OCLReID
python run_video.py --show_result

This would run the ./demo.mp4.

Run on the customized dataset

Our customized dataset is provided in dataset directory with four scenarios: corridor1, corridor2, lab_corridor and room. We provide raw_video.mp4 and labels.txt for each scenario. Specifically, bbox annotations in the label.txt are represented as x1,y1,w,h.

Note: the annotations are rough, but should be enough for evaluating the ReID performance of algorithms.

Citation

@article{ye2024oclrpf,
  title={Person re-identification for robot person following with online continual learning},
  author={Ye, Hanjing and Zhao, Jieting and Zhan, Yu and Chen, Weinan and He, Li and Zhang, Hong},
  journal={IEEE Robotics and Automation Letters},
  year={2024},
  publisher={IEEE}
}