supervision
A Python toolkit of reusable building blocks for computer vision: loading datasets, drawing annotations, track
- Stars
- 51.5k
- Forks
- 4.9k
- Licence
- MIT
- Language
- Python
- Last update
- 4 days ago
- Type
- Library
What it is
Supervision is a Python library that provides ready-made pieces for computer vision projects. It handles common chores around a detection model: loading and converting datasets in formats like COCO, YOLO and Pascal VOC, drawing boxes and labels on images or video frames, and counting objects in zones. It is designed to work with any detection or segmentation model rather than being tied to one.
- Draw bounding boxes and labels over detections from a model using the built-in annotators
- Load a COCO, YOLO or Pascal VOC dataset, split it into train and test sets, and merge datasets
- Convert a YOLO dataset to Pascal VOC format
- Count objects entering a defined zone in a video stream
Great for
- Python developers building apps on top of object detection or segmentation models
- Teams that already use YOLO, Ultralytics, Transformers, or MMDetection and want shared visualization code
- People preparing or converting computer vision datasets between COCO, YOLO and Pascal VOC formats
- Builders of video analytics such as zone counting or tracking demos
Not ideal for
- Non-programmers looking for a point-and-click app, since it is a code library with Python examples only
- Projects that need a trained model included, because the README covers connecting models rather than providing one
- Users who cannot run Python 3.10 or newer, which the README lists as required
How to get it
- 1
Install with a package manager
You'll need: Python 3.10 or newer
Install the package
pip install supervision - 2
Run it
- Install Python 3.10 or newer and create a virtual environment
- Run pip install supervision
- Load an image with OpenCV and create an sv.BoxAnnotator
- Pass your model's detections to annotate() to get an annotated frame
- 3
Get the most out of it
- Use the model connectors (Ultralytics, Transformers, MMDetection, Inference, rfdetr) so detections come back in the sv.Detections format
- Use sv.DetectionDataset to load, split and merge datasets in COCO, YOLO or Pascal VOC formats
- Explore the linked cookbooks and end-to-end examples for tracking and zone-based counting
Downloads
Strengths
- Model agnostic: the README says it connects to popular libraries and works with any classification, detection or segmentation model
- Covers several steps of a vision pipeline (datasets, annotation, zones) in one package under the MIT license
- Install is a single pip command and the README includes runnable examples
Trade-offs
- Scope is limited to the building-block layer; you still need your own model and application logic
- Some integrations, such as the Roboflow inference example, need an API key from Roboflow
- The README provided is truncated, so details of features beyond the quickstart were not verified here
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| supervision (this) | 51.5k | Library | intermediate | 4 days ago | Python developers building apps on top of object detection or segmentation models |
| albumentations | 15.3k | Library | beginner | 1.4 years ago | Machine learning engineers training image classification or segmentation models with limited data |
| ai-engineering-from-scratch | 76.3k | Learning resource | intermediate | 4 days ago | Students who want to understand how LLMs, retrieval, and agents work internally |
| AI-For-Beginners | 70.5k | Learning resource | beginner | 13 days ago | Beginners who want a structured, self-paced introduction to AI |
| carla | 14.6k | App | advanced | 4 days ago | Autonomous driving researchers testing perception or control software |
| facenet | 14.4k | AI model / agent | advanced | 3.3 years ago | Researchers reproducing the FaceNet paper with TensorFlow |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.