This repo. is the PyTorch implementation of multi-object tracker OTCD. The paper is real-time online multi-object tracking in compressed domain. There maybe a slight gap between the performance obtained by this script and the performance reported in the paper.

@article{liu2019real,
  title={Real-Time Online Multi-Object Tracking in Compressed Domain},
  author={Liu, Qiankun and Liu, Bin and Wu, Yue and Li, Weihai and Yu, Nenghai},
  journal={IEEE Access},
  volume={7},
  pages={76489--76499},
  year={2019},
  publisher={IEEE}
}

PyTorch = 0.3
python >= 3.5

1. before running OTCD, CoViAR needs to be installed. Please click the URL for the usage of CoViAR.

2. down load this repo.

git clone https://github.com/liuqk3/OTCD.git
cd OTCD

2. download the pretrained model from BaiduYunPan or OneDrive. Then put all models to ./save. If you have any problems with the download process, please email me.

3. run the tracker

python tracking_on_mot.py --mot_dir path/to/MOT-dataset

The training scripts are also published in useful_scripts. You can train all the models by the given scripts.