# AI use cases for Supervision (roboflow/supervision) Repo: [roboflow/supervision](https://github.com/roboflow/supervision) | Use cases: 3 | Page: https://dedupedai.com/use-cases/roboflow-supervision/ Supervision is Roboflow's model-agnostic Python toolkit for computer vision. It wraps the output of classification, detection and segmentation models from Ultralytics, Transformers, MMDetection, Inference or RF-DETR into one Detections object, then gives you annotators to draw results, zone counting and tracking tools, and dataset utilities to load, split, merge and convert YOLO, COCO and Pascal VOC datasets. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Detect and annotate objects in a few lines of Python: https://dedupedai.com/use-cases/roboflow-supervision/detect-and-annotate-objects-in-images/ - Convert and split vision datasets between YOLO, COCO and Pascal VOC: https://dedupedai.com/use-cases/roboflow-supervision/convert-and-split-vision-datasets/ - Measure how long people or vehicles stay in a zone: https://dedupedai.com/use-cases/roboflow-supervision/measure-dwell-time-in-zones-from-video/ ## How to detect and annotate objects in a few lines of Python with Supervision Repo: [roboflow/supervision](https://github.com/roboflow/supervision) | Difficulty: Beginner | Time: 20 minutes | Basis: From the README Run a detector, wrap its output as sv.Detections and draw clean boxes with an annotator. **Who it is for:** A developer prototyping a vision feature who wants to see detections on real images fast. **The problem:** Every model returns results in a different format, so drawing and filtering them means custom code each time. **The outcome:** Annotated images with boxes from any supported model, using one consistent Detections object. ### Steps 1. Run `pip install supervision` 2. Install the example dependencies with `pip install pillow rfdetr` 3. Load an image and run `RFDETRSmall()` predict with a threshold to get detections 4. Create `sv.BoxAnnotator()` and call annotate on a copy of the image with the detections 5. Swap in another model through the connectors for Ultralytics, Transformers, MMDetection or Inference ### Prompt to paste into your agent ```text Using roboflow/supervision with RF-DETR, write a script that runs detection on every image in at threshold 0.5, draws boxes with sv.BoxAnnotator and saves annotated copies to , printing the number of detections per image. ``` ### Caveat Supervision does not include its own detector; accuracy depends entirely on the model you plug in. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/roboflow-supervision/detect-and-annotate-objects-in-images/ ## How to convert and split vision datasets between YOLO, COCO and Pascal VOC with Supervision Repo: [roboflow/supervision](https://github.com/roboflow/supervision) | Difficulty: Beginner | Time: 20 minutes | Basis: From the README Load a labeled dataset in one format, split it into train, test and validation, and save it in another. **Who it is for:** An ML engineer whose labels came out of one tool in COCO but whose training code expects YOLO. **The problem:** Converting annotation formats and splitting datasets by hand is tedious and easy to get wrong. **The outcome:** Clean train, test and validation splits saved in the format your training pipeline needs. ### Steps 1. Install with `pip install supervision` 2. Load your data with `sv.DetectionDataset.from_coco`, from_yolo or from_pascal_voc 3. Split it with `dataset.split(split_ratio=0.7)` and split the remainder again for test and validation 4. Merge datasets from different sources with `sv.DetectionDataset.merge` if needed; classes are combined 5. Save each split with as_yolo, as_coco or as_pascal_voc ### Prompt to paste into your agent ```text With roboflow/supervision, load my COCO dataset from and , split it 70/15/15 into train, test and validation, and save each split in YOLO format under . Print the class list and the size of each split. ``` ### Caveat Check a few converted labels visually before training, since format conversions can hide class or coordinate mistakes. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/roboflow-supervision/convert-and-split-vision-datasets/ ## How to measure how long people or vehicles stay in a zone with Supervision Repo: [roboflow/supervision](https://github.com/roboflow/supervision) | Difficulty: Advanced | Time: 2 hours | Basis: From the README Combine detection, tracking and zones to time how long objects spend in an area of a video stream. **Who it is for:** An operations or retail analyst who wants queue wait times or traffic dwell times from existing cameras. **The problem:** Manually watching footage to estimate wait times is slow and inaccurate. **The outcome:** Per-object time spent in each zone you draw, from recorded or live video. ### Steps 1. Install with `pip install supervision` plus the detector you will use 2. Follow the Dwell Time Analysis with Computer Vision tutorial linked in the README 3. Detect objects per frame, track them across frames and filter detections to the classes you care about 4. Define the zones on the frame and calculate time spent in designated zones for each tracked object 5. Use annotators to draw zones, tracks and timers so you can check the results by eye ### Prompt to paste into your agent ```text Using roboflow/supervision, build a script that reads