PIDMOT

Official implementation for TCSVT paper: “One-shot Multiple Object Tracking with Robust ID Preservation”

Installation

PIDMOT is built upon codebase of FairMOT. We use python 3.7 and pytorch >= 1.2.0

Step1. Install PIDMOT

git clone https://github.com/Kroery/PIDMOT.git
cd PIDMOT
pip3 install -r requirements.txt

Step2. Install DCNv2. We use DCNv2 in our backbone network and more details can be found in their repo.

git clone https://github.com/CharlesShang/DCNv2
cd DCNv2
./make.sh

Training

• Download the training data
• Change the dataset root directory 'root' in src/lib/cfg/data.json and 'data_dir' in src/lib/opts.py
• Pretrain on MOTSynth and finetuned by CrowdHuman:

sh experiments/motsynth_saca_idm_clip.sh
sh experiments/crowdhuman_motsynth_saca_idm_clip.sh

• Train on MOT17:

sh experiments/mix_mot17_ch60_synth_saca_idm_clip.sh

• Train on MOT20:

sh experiments/mix_mot20_ch60_synth_saca_idm_clip.sh

Tracking

• Tracking on MOT17 test set:

cd src
python track.py mot --arch dlaSACAidm_34 --load_model $model_path$ --test_mot17 True --match_thres 0.4 --conf_thres 0.25

• Tracking on MOT20 test set:

cd src
python track.py mot --arch dlaSACAidm_34 --load_model $model_path$ --test_mot20 True --match_thres 0.4 --conf_thres 0.25

Citation

@ARTICLE{10342840,
  author={Lv, Weiyi and Zhang, Ning and Zhang, Junjie and Zeng, Dan},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={One-shot Multiple Object Tracking with Robust ID Preservation}, 
  year={2023},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TCSVT.2023.3339609}}