Supervision
Open-source Python computer vision library for annotating images, tracking objects and managing datasets. MIT licensed, with RF-DETR and Ultralytics support.
Supervision is an MIT-licensed Python library from Roboflow for developers building computer vision applications around their own models. It handles the work around predictions: drawing results on images and video, following objects across frames, and turning detections into counts. It can work with images and datasets on your machine.
Its shared Detections format lets applications use outputs from different detection and segmentation models through the same tools. RF-DETR returns this format directly. Connectors also support Ultralytics, Transformers, SAM, Detectron2 and MMDetection, along with Roboflow Inference and Azure AI Vision. The Roboflow Inference integration requires a Roboflow API key.
Customizable annotators draw bounding boxes, segmentation masks and labels. Object tracking assigns persistent IDs across video frames, while zone tools count or filter detections within polygons. It also supports counting objects that cross a line and detecting small objects in images. These capabilities suit applications that need to measure activity in video as well as display model predictions.
Dataset tools load, split, merge and save collections, with conversion between YOLO, COCO and Pascal VOC formats. For developers comparing models, Supervision provides mean average precision (mAP) metrics and confusion matrices to assess prediction quality.
