1. Introduction

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.

2. Citation

@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}
}

3. Requirements

• Docker
• Nvidia-Docker
• Docker-Compose

4. Quickstart

1. Download this repo

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

2. Build docker image

sudo docker-compose build

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

4. Download MOT Challenge dataset and place into a sub directory of OTCD. If placed outside of the OTCD directory, edit the volume mapping in docker-compose.yml.

5. When finished start docker container

sudo docker-compose up -d

6. Enter the container

sudo docker exec -it otcd bash

7. Start tracker

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

5. Code for training

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