supervision
RoboflowRoboflow's model-agnostic toolkit for working with computer-vision results. It converts detections and segmentations from Ultralytics, Transformers, MMDetection, Roboflow Inference and other libraries into one format, then draws annotations, filters and tracks detections, counts objects in zones, slices large images for inference, loads and converts datasets and computes metrics.
Openness
5 high confidence- license
- MIT(OSI)
- source
- public(github.com/roboflow/supervision)
- core features withheld
- no — Roboflow's paid platform is a separate product
MIT-licensed and built from the public repository. Roboflow sells a hosted vision platform, which lists supervision as a free open-source part of its ecosystem rather than as something priced.
- https://pypi.org/pypi/supervision/json recorded 2026-09-27
PyPI record for supervision 0.30.5, uploaded 2026-09-22, license_expression MIT, Repository github.com/roboflow/supervision.
- https://raw.githubusercontent.com/roboflow/supervision/develop/LICENSE.md recorded 2026-09-27
LICENSE.md is the "MIT License", "Copyright (c) 2022 Roboflow".
- https://roboflow.com/pricing recorded 2026-09-27
Roboflow pricing lists paid platform plans; supervision appears only in the footer under the open-source ecosystem, not as a priced item.
- https://ungh.cc/repos/roboflow/supervision/files/develop recorded 2026-09-27
Repository tree with no ee, enterprise, commercial or pro directory, and no model or training directory.
Adoption
3 high confidenceMeasured on monthly PyPI downloads of the supervision package.
- https://pypistats.org/api/packages/supervision/recent recorded 2026-09-27
last_month 819,606 downloads of supervision
Capability
1 high confidencesupervision works on the output of models from other libraries and fits or runs no model itself, so it sits with the primitives. Its own README calls it model agnostic, and its quickstart imports the detector from a separate package.
- https://raw.githubusercontent.com/roboflow/supervision/develop/README.md recorded 2026-09-27
README: "Supervision was designed to be model agnostic. Just plug in any classification, detection, or segmentation model. For your convenience, we have created connectors for the most popular libraries".
Verified 2026-09-27