supervision

What it is and what it does

Supervision is a toolkit for building computer vision applications around detection, segmentation, and classification models. It abstracts away common tasks—loading datasets in multiple formats (YOLO, COCO, Pascal VOC), converting between them, splitting and merging datasets, and visualizing model outputs with customizable annotators. The package is model-agnostic and works with any framework; it includes connectors for popular libraries, but you can also pass detections directly if your model already returns the expected format.

The library handles the plumbing so you focus on application logic rather than format parsing or visualization boilerplate. It depends on standard scientific Python libraries (numpy, scipy, pillow, matplotlib) plus video support (av) and YAML parsing, making it straightforward to integrate into existing pipelines. The package is actively maintained, well-starred, and supports modern Python versions (3.10–3.14).

Use it for:

• Load and split YOLO or COCO datasets for training, validation, and testing without manual file parsing
• Annotate detection results with bounding boxes, masks, or labels for debugging model predictions
• Convert datasets between YOLO, COCO, and Pascal VOC formats for compatibility with different training frameworks
• Track objects across video frames and calculate metrics like dwell time or speed using zone-based analysis
• Merge multiple detection datasets with different class vocabularies into a unified training set

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Supervision provides utilities for loading, annotating, and processing computer vision datasets and model outputs—detection, segmentation, and classification results from any model framework.

Yes. Supervision is a mature, actively maintained toolkit with no known vulnerabilities, permissive licensing, and low install friction. It solves real friction points in computer vision workflows—dataset loading, format conversion, and visualization—and integrates cleanly with model frameworks. Install it if you work with detection, segmentation, or classification models and want to avoid reinventing dataset utilities.

Install

supervision on PyPI

pip install supervision

uv add supervision

poetry add supervision

Installing supervision

Before you install

Low install friction with a pure-Python wheel. Active maintenance (last commit 2026-08-14, released 2026-08-04) and a large community (49398 stars). Ten runtime dependencies are all well-established libraries.

License in practice

MIT license permits unrestricted use, modification, and distribution in commercial and private projects with minimal restrictions.

Quickstart

pip install supervision

import supervision as sv
from pillow import Image

image = Image.open("path/to/image.jpg")
detections = sv.Detections(...)
box_annotator = sv.BoxAnnotator()
annotated_frame = box_annotator.annotate(scene=image.copy(), detections=detections)

Requires Python >= 3.10

Verify before relying

• Whether the package works with models outside the listed connectors without custom adapter code
• Performance characteristics when handling large datasets or high-resolution video streams
• Whether all dataset format conversions preserve annotation fidelity across YOLO, Pascal VOC, and COCO

Package facts

Evidence: supervision-0.30.0-py3-none-any.whl

Keywords: AI, deep-learning, DL, machine-learning, ML, Roboflow, vision

Tags

More Software Development packages