How to use Supervision: what you can do with it

3 use cases

• Detect and annotate objects in a few lines of Python

  Run a detector, wrap its output as sv.Detections and draw clean boxes with an annotator.

  • REPO roboflow/supervision
  • Claude Code
  • Codex
  • Cursor
  • Data and analytics
  • Beginner
  View prompt

  Using roboflow/supervision with RF-DETR, write a script that runs detection on every image in <folder> at threshold 0.5, draws boxes with sv.BoxAnnotator and saves annotated copies to <output folder>, printing the number of detections per image.

  Share Email me this

• Convert and split vision datasets between YOLO, COCO and Pascal VOC

  Load a labeled dataset in one format, split it into train, test and validation, and save it in another.

  • REPO roboflow/supervision
  • Claude Code
  • Codex
  • Cursor
  • Data and analytics
  • Beginner
  View prompt

  With roboflow/supervision, load my COCO dataset from <images path> and <annotations path>, split it 70/15/15 into train, test and validation, and save each split in YOLO format under <output folder>. Print the class list and the size of each split.

  Share Email me this

• Measure how long people or vehicles stay in a zone

  Combine detection, tracking and zones to time how long objects spend in an area of a video stream.

  • REPO roboflow/supervision
  • Claude Code
  • Codex
  • Cursor
  • Data and analytics
  • Advanced
  View prompt

  Using roboflow/supervision, build a script that reads <video file>, detects and tracks people, counts how long each one stays inside a zone I define by coordinates, and writes an annotated video plus a CSV of track id and seconds in zone. Follow the README's dwell time tutorial approach.

  Share Email me this

Quick start with Supervision

1. Check you have what it needs: Python 3.10 or later; A detection or segmentation model; the quickstart uses RF-DETR via `pip install pillow rfdetr`; A Roboflow API key only if you use Inference or download datasets from Roboflow.
2. Open roboflow/supervision on GitHub and read the README and license.
3. Run pip install supervision
4. Install the example dependencies with pip install pillow rfdetr
5. Load an image and run RFDETRSmall() predict with a threshold to get detections
6. Create sv.BoxAnnotator() and call annotate on a copy of the image with the detections

These steps come from our top use case, Detect and annotate objects in a few lines of Python, which also has a copy-paste agent prompt.

Using the Deduped MCP, get the setup steps for roboflow/supervision

claude mcp add --transport http deduped https://dedupedai.com/mcp

Best AI repos for Data and analytics

About this repo

• Who it is forPython developers and ML engineers building vision apps who want to stop rewriting the glue code around their models.
• Why it is hot51k stars, talked about in 2 AI sources we track
• LanguagePython
• Sourcegithub.com/roboflow/supervision · Back to Hot repos

Requirements

• Python 3.10 or later
• A detection or segmentation model; the quickstart uses RF-DETR via `pip install pillow rfdetr`
• A Roboflow API key only if you use Inference or download datasets from Roboflow

Related repos

• Trended earlier REPO ultralytics/ultralytics Ultralytics Ultralytics is the package behind the YOLO family of computer vision models, up to YOLO26. One `pip install ultralytics` gives you a `yolo` CLI and a Python API for detection, segmentation and more: predict on images, train on your own dataset, validate, and export to formats like ONNX. It is AGPL-3.0 with an enterprise license for commercial use.
• Trended earlier REPO fengzhizi715/YoloWebAgentStarter Yolo Web Agent Starter A local, single-user web workbench for YOLO datasets that covers image and video import, manual annotation, validation, YOLO and COCO exchange, local training, native YOLO evaluation, and an evidence-based AI assistant.
• Trended earlier REPO lucidrains/robo_ttt Robo TTT robo_ttt is a PyTorch implementation of RoboTTT (Jiang et al., Stanford and Nvidia), providing test-time training wrappers and memory modules for robot policy models, including optional next-latent forward dynamics.