trackers

Hello

trackers is a unified library offering clean room re-implementations of leading multi-object tracking algorithms. Its modular design allows you to easily swap trackers and integrate them with object detectors from various libraries like inference, ultralytics, or transformers.

Tracker	Paper	Year	Status	Colab
SORT		2016	✅
ByteTrack		2021	✅
OC-SORT		2022	🚧	🚧
BoT-SORT		2022	🚧	🚧

trackers-2.0.0-promo.mp4

Installation

Pip install the trackers package in a Python>=3.9 environment.

pip install trackers

install from source

By installing trackers from source, you can explore the most recent features and enhancements that have not yet been officially released. Please note that these updates are still in development and may not be as stable as the latest published release.

pip install git+https://github.com/roboflow/trackers.git

Benchmarks

Performance of the trackers on test splits from tracking datasets.

MOT17

Tracker	HOTA	IDF1	MOTA
SORT	58.4	69.9	67.2
ByteTrack	60.1	73.2	74.1

SportsMOT

Tracker	HOTA	IDF1	MOTA
SORT	70.9	68.9	95.7
ByteTrack	73.0	72.5	96.4

SoccerNet-tracking

Tracker	HOTA	IDF1	MOTA
SORT	81.6	76.2	95.1
ByteTrack	84.0	78.1	97.8

Quickstart

With a modular design, trackers lets you combine object detectors from different libraries with the tracker of your choice. Here's how you can use SORTTracker with various detectors:

import supervision as sv
from trackers import SORTTracker
from rfdetr import RFDETRMedium

tracker = SORTTracker()
model = RFDETRMedium(device="cuda")
annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)

def callback(frame, _):
    detections = model.predict(frame, threshold=NMS_THRESHOLD)
    detections = tracker.update(detections)
    return annotator.annotate(frame, detections, labels=detections.tracker_id)

sv.process_video(
    source_path="<INPUT_VIDEO_PATH>",
    target_path="<OUTPUT_VIDEO_PATH>",
    callback=callback,
)

run with inference

import supervision as sv
from trackers import SORTTracker
from inference import get_model

tracker = SORTTracker()
model = get_model(model_id="rfdetr-medium")
annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)

def callback(frame, _):
    result = model.infer(frame)[0]
    detections = sv.Detections.from_inference(result)
    detections = tracker.update(detections)
    return annotator.annotate(frame, detections, labels=detections.tracker_id)

sv.process_video(
    source_path="<INPUT_VIDEO_PATH>",
    target_path="<OUTPUT_VIDEO_PATH>",
    callback=callback,
)

run with ultralytics

import supervision as sv
from trackers import SORTTracker
from ultralytics import YOLO

tracker = SORTTracker()
model = YOLO("yolo11m.pt")
annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)

def callback(frame, _):
    result = model(frame)[0]
    detections = sv.Detections.from_ultralytics(result)
    detections = tracker.update(detections)
    return annotator.annotate(frame, detections, labels=detections.tracker_id)

sv.process_video(
    source_path="<INPUT_VIDEO_PATH>",
    target_path="<OUTPUT_VIDEO_PATH>",
    callback=callback,
)

run with transformers

import torch
import supervision as sv
from trackers import SORTTracker
from transformers import RTDetrV2ForObjectDetection, RTDetrImageProcessor

tracker = SORTTracker()
image_processor = RTDetrImageProcessor.from_pretrained("PekingU/rtdetr_v2_r18vd")
model = RTDetrV2ForObjectDetection.from_pretrained("PekingU/rtdetr_v2_r18vd")
annotator = sv.LabelAnnotator(text_position=sv.Position.CENTER)

def callback(frame, _):
    inputs = image_processor(images=frame, return_tensors="pt")
    with torch.no_grad():
        outputs = model(**inputs)

    h, w, _ = frame.shape
    results = image_processor.post_process_object_detection(
        outputs,
        target_sizes=torch.tensor([(h, w)]),
        threshold=0.5
    )[0]

    detections = sv.Detections.from_transformers(
        transformers_results=results,
        id2label=model.config.id2label
    )

    detections = tracker.update(detections)
    return annotator.annotate(frame, detections, labels=detections.tracker_id)

sv.process_video(
    source_path="<INPUT_VIDEO_PATH>",
    target_path="<OUTPUT_VIDEO_PATH>",
    callback=callback,
)

License

The code is released under the Apache 2.0 license.

Contribution

We welcome all contributions—whether it’s reporting issues, suggesting features, or submitting pull requests. Please read our contributor guidelines to learn about our processes and best practices.