roboflow/trackers

Trackers gives you clean, modular re-implementations of leading multi-object tracking algorithms released under the permissive Apache 2.0 license. You combine them with any detection model you already use.

What it solves

trackers is a Python library that provides a standardized, plug-and-play way to implement multi-object tracking (MOT) for any object detection model. It eliminates the need to write custom glue code or implement complex tracking algorithms from scratch to handle challenges like occlusions, fast motion, and moving cameras.

How it works

The library provides clean-room implementations of several industry-standard tracking algorithms, including SORT, ByteTrack, OC-SORT, BoT-SORT, C-BIoU, and McByte. It is designed to be detector-agnostic, meaning it can work with any model that produces bounding boxes (such as YOLO, DETR, or RT-DETR). It integrates natively with the supervision.Detections format, allowing users to pass detections in and receive tracked detections back via a consistent update() interface.

Who it’s for

  • Engineers shipping production pipelines who need a reliable, Apache 2.0 licensed tracking library.
  • Researchers comparing different tracking algorithms across standardized benchmarks.
  • Hobbyists building computer vision projects that require object tracking.

Highlights

  • Detector-Agnostic: Works with any model that produce bounding boxes without requiring a specific inference library.
  • Clean-room Implementations: Algorithms are re-implemented from original papers for transparency and modification.
  • Comprehensive Algorithm Suite: Includes a variety of trackers like ByteTrack and BoT-SORT for different scene types.
  • Camera Motion Compensation: Native support in BoT-SORT and McByte to maintain stable IDs during frame shifts.
  • Built-in Evaluation and Tuning: Includes CLI tools for tracking, downloading benchmark datasets (MOT17, SportsMOT), and Optuna-based hyperparameter tuning.
  • Native Integration: Seamlessly plugs into the supervision ecosystem.

Related

  • Project
    mikel-brostrom/boxmot

    A pluggable multi-object tracking framework that provides a unified interface for integrating various detectors and trackers for both axis-aligned and oriented bounding boxes.

  • Project
    adipandas/multi-object-tracker

    A Python library providing easy-to-use implementations of various multi-object tracking algorithms and OpenCV-based object detectors for video analysis.

  • Project
    noahcao/OC_SORT

    A pure motion-model-based multi-object tracker that improves robustness in crowded scenes and non-linear motion by rethinking the SORT algorithm.

  • Project
    cheind/py-motmetrics

    A Python library for evaluating multiple object tracking results using standard metrics like MOTA, MOTP, IDF1, and HOTA.

  • Project
    yzfzzz/depth-detect

    A high-performance C++ and TensorRT framework that fuses YOLO object detection with monocular/binocular depth estimation to track objects and estimate their motion states.