Repolode
Submit
roboflow

roboflow/supervision

Python toolkit of reusable computer vision utilities for loading datasets, running detections, tracking objects, and annotating frames.

View on GitHubWebsite
Stars
51,250
Forks
4,846
Open issues
80
Last commit
6 hours ago

What is it?

Supervision is a Python library of building blocks for computer vision applications. It works with the output of any classification, detection, or segmentation model, and provides connectors for libraries such as Ultralytics, Transformers, MMDetection, and Roboflow Inference.

The library covers much of the code that surrounds a model: drawing bounding boxes, masks, labels and traces through a family of customizable annotators, tracking objects across video frames, counting detections inside polygon zones or across lines, and computing evaluation metrics. It also loads, splits, merges, and converts datasets in COCO, YOLO, and Pascal VOC formats.

Install it with pip on Python 3.10 or newer. Its API centers on the sv.Detections and sv.DetectionDataset objects, so components can be combined without tying the pipeline to one model.

Key features

  • 01Annotators for drawing boxes, masks, labels, and traces on frames
  • 02Object tracking across video frames
  • 03Counting detections in polygon zones or across lines
  • 04Loading, splitting, merging, and converting COCO, YOLO, and Pascal VOC datasets
  • 05Connectors for Ultralytics, Transformers, MMDetection, and Roboflow Inference

Use cases

  • Visualizing model predictions on images or video during development
  • Counting people or vehicles entering a zone in retail or traffic footage
  • Preparing and converting object detection datasets between formats
  • Building custom computer vision pipelines around any detection model

Quick start

  1. 1

    Install the package with pip (Python 3.10 or newer)

    pip install supervision
  2. 2

    Install optional dependencies for the model example

    pip install pillow rfdetr

Things to know

  • Requires Python 3.10 or newer.
  • Some model connectors, such as Inference, need a Roboflow API key.

Topics

#computer-vision#object-detection#object-tracking#image-annotation#dataset-conversion#yolo#python#classification#coco#deep-learning

FAQ

What is supervision?
Python toolkit of reusable computer vision utilities for loading datasets, running detections, tracking objects, and annotating frames. Supervision is a Python library of building blocks for computer vision applications. It works with the output of any classification, detection, or segmentation model, and provides connectors for libraries such as Ultralytics, Transformers, MMDetection, and Roboflow Inference.
Is supervision free to use? What license does it have?
supervision is open source under the MIT license. Check the license terms before using it in commercial products.
Is supervision still maintained?
Actively maintained. The last commit to supervision was on Oct 20, 2026.
What can I use supervision for?
Visualizing model predictions on images or video during development Counting people or vehicles entering a zone in retail or traffic footage Preparing and converting object detection datasets between formats Building custom computer vision pipelines around any detection model
How do I install supervision?
Install the package with pip (Python 3.10 or newer): pip install supervision Install optional dependencies for the model example: pip install pillow rfdetr
Where can I get supervision?
The source code and installation instructions are on GitHub: https://github.com/roboflow/supervision

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects