LPT

Official implementation of the CVPR2022 paper "Learning of Global Objective for Network Flow in Multi-Object Tracking"

Setup

# Install cvxpy and qpth
pip install cvxpy
pip install qpth

# Install pytorch_geometric(Version < 2.0.0) and gurobipy, please check if gurobipy is correctly installed
import gurobipy as gp
gurobi_solver = gp.Model()

# Install torchreid from: https://github.com/KaiyangZhou/deep-person-reid, and put it inside ./lib folder.

Data

Download pre-processed detections&&appearance features (~1.4GB): [Google Drive] We aslo provide the tracking results on MOT17/20 test set in txt format, which are the results reported in the paper.

Training

In order to replicate the results, you need to adjust the MOT17/20 data path accordingly.

Execute run_train.ipynb and monitor all relevant metrics, in general the loss converges at around 8 epochs.

Inference

Execute run_test.ipynb. Note that for one specific MOT20 sequence, it may take more time for the inference due to the large number of objects to track.

If you have any questions using this code, please open an issue. I'll respond ASAP.

Citing

If you find this code useful in your research, please consider citing:

@inproceedings{li2022learning,
  title={Learning of Global Objective for Network Flow in Multi-Object Tracking},
  author={Li, Shuai and Kong, Yu and Rezatofighi, Hamid},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={8855--8865},
  year={2022}
}