Multiple Object Tracking as ID Prediction

Ruopeng Gao,  Ji Qi,  Limin Wang, 
Nanjing University
📧 Primary Contact: ruopenggao@gmail.com

:mag: Overview

TL; DR. We propose a novel perspective to regard the multiple object tracking task as an in-context ID prediction problem. Given a set of trajectories carried with ID information, MOTIP directly decodes the ID labels for current detections, which is straightforward and effective.

:fire: News

• 2025.07.10: Thanks everyone—we’ve hit 300 stars :tada:! A new TUTORIAL (is updating gradually) is live to help understand and work with the model.
• 2025.04.11: Support loading previous MOTIP checkpoints from the prev-engine to inference :floppy_disk:. See MODEL_ZOO for details.
• 2025.04.06: Now, you can use the video demo to perform nearly real-time tracking on your videos :joystick:.
• 2025.04.05: We support FP16 for faster inference :racing_car:.
• 2025.04.03: The new codebase is released :tada:. Compared to the previous version, it is more concise and efficient :rocket:. Feel free to enjoy it!
• 2025.03.25: Our revised paper is released at arXiv:2403.16848. The latest codebase is being organized :construction:.
• 2025.02.27: Our paper is accepted by CVPR 2025 :tada: :tada:. The revised paper and a more efficient codebase will be released in March. Almost there :nerd_face: ~
• 2024.03.26: The first version of our paper is released at arXiv:2403.16848v1 :pushpin:. The corresponding codebase is stored in the prev-engine branch (No longer maintained starting April 2025 :no_entry:).

:dash: Quick Start

• See INSTALL.md for instructions of installing required components.
• See DATASET.md for datasets download and preparation.
• See GET_STARTED.md for how to get started with our MOTIP, including pre-training, training, and inference.
• See MODEL_ZOO.md for well-trained models.
• See MISCELLANEOUS.md for other miscellaneous settings unrelated to the model structure, such as logging.
• See TUTORIAL.md to understand and better develop our models.

:bouquet: Acknowledgements

This project is built upon Deformable DETR, MOTR, TrackEval. Thanks to the contributors of these great codebases.

:pencil2: Citation

If you think this project is helpful, please feel free to leave a :star: and cite our paper:

@InProceedings{{MOTIP},
    author    = {Gao, Ruopeng and Qi, Ji and Wang, Limin},
    title     = {Multiple Object Tracking as ID Prediction},
    booktitle = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {27883-27893}
}

:star2: Stars