Supervision

What it does

Supervision is an open-source Python library for computer vision workflows. It takes in images, video, and detection outputs from object detection or segmentation models, along with dataset files in formats such as YOLO, Pascal VOC, and COCO. It produces processed detections, annotated visuals, merged or split datasets, and format conversions.

People use Supervision to load and convert datasets, split or merge them for training and validation, and add annotation and tracking utilities on top of model outputs. Its distinct focus is on reusable building blocks that sit between a detection model and the final application, rather than a hosted end-user app.

Supervision suits developers and researchers building computer vision applications in Python. It is installed as a library and runs in local Python environments. Its source is MIT-licensed and free to use, though any models or compute you run with it are separate costs. Its documentation and repository are the primary references for setup and usage.

What it can do

• Load datasets in YOLO, Pascal VOC, and COCO formats
• Split datasets into train, test, and validation subsets
• Merge multiple datasets and combine their class lists
• Export datasets back to YOLO, Pascal VOC, and COCO formats
• Process and annotate object detection and segmentation outputs
• Support for tracking, zone-based analysis, and metrics

Example workflows

Split and convert a labeled dataset

Illustrative example. Input: a YOLO-format dataset folder. Steps: load it with DetectionDataset.from_yolo, split into train and test subsets, and export with as_coco. Output: COCO-format train and test annotation files.

Merge datasets before training

Illustrative example. Input: two DetectionDataset objects with different class sets. Steps: combine them with DetectionDataset.merge and inspect the merged class list. Output: one dataset containing images from both sources and a unified class list.

Before you choose

• Supervision supplies utilities around model outputs; choose a compatible model and inference runtime separately.
• Model licenses, compute and hosted services have their own terms and costs.
• Test annotations, tracking and metrics against your data and expected output format.

Listed under

• Coding
• Computer vision
• Dataset tools
• Developers
• Model benchmarking
• Object detection
• Object tracking

Get started

Follow the installation and quickstart sections in the documentation.

1. Install the supervision package using the installation instructions in the documentation.
2. Work through the quickstart section to confirm a basic detection and annotation run.
3. Load a dataset with the from_yolo, from_pascal_voc, or from_coco method for your data format.

Supervision is an MIT-licensed Python library. The library has no subscription fee; any model inference, hardware or hosted services used alongside it have their own costs.

Supervision is free to use. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Supervision ↗supervision.roboflow.com
• GitHub - roboflow/supervision: We write your reusable computer vision tools. 💜 · GitHub ↗github.com

Company website

supervision.roboflow.com: product, pricing and policy pages

Hacker News

Launch posts and discussion

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.