Keypoint Message Passing

This repository hosts the code for our paper Keypoint Message Passing for Video-based Person Re-Identification.

Working in progress...

Build Environment

1. Make sure conda is installed.

2. Create environment from file:

conda env create -f environment.yml

3. Install torchreid

git clone https://github.com/KaiyangZhou/deep-person-reid.git
cd deep-person-reid/
pip install -r requirements.txt
python setup.py develop

Prepare Dataset

1. Download MARS dataset and keypoints.

2. Organize the file tree as below:

KeypointMessagePassing
└── data
    └── mars
        ├── info/
        ├── bbox_train/
        ├── bbox_test/
        ├── bbox_train_keypoints/
        ├── bbox_test_keypoints/

Run

# training
CUDA_VISIBLE_DEVICES=0 python scripts/trainval_vginteract.py --cfg_file <prefix>/cfg.yaml --data.save_dir logs/<version_number>/ --data.sources ['marspose'] --data.targets ['marspose'] --train.max_epoch <epoch_number>

# testing
CUDA_VISIBLE_DEVICES=0 python scripts/trainval_vginteract.py --cfg_file logs/<version_number>/<time_stamp_and_machine_name>/cfg.yaml --model.resume logs/<version_number>/<time_stamp_and_machine_name>/model/model.pth.tar-<epoch_number> --test.evaluate

Note: cfg.yaml contains the default hyper-parameters. The following flags override the defaut hyper-params.

Citation

@inproceedings{chen2021keypoint,
  title={Keypoint Message Passing for Video-based Person Re-Identification},
  author={Chen, Di and Doering, Andreas and Zhang, Shanshan and Yang, Jian and Gall, Juergen and Schiele, Bernt},
  booktitle={AAAI},
  year={2022}
}