supervisionWe write your reusable computer vision tools. 💜

supervision

supervision is a Python library that provides reusable, model-agnostic computer vision tools for building production-ready applications. It streamlines the entire computer vision pipeline—from loading datasets to running inference with any detection, classification, or segmentation model, and visualizing results with highly customizable annotators.

Key Features

• Model-Agnostic Framework: Works seamlessly with popular libraries like Ultralytics, Transformers, MMDetection, and Inference without lock-in
• Customizable Annotators: Draw detections on images and videos with flexible, composable visualization tools for detection boxes, masks, labels, and more
• Dataset Management: Load, split, merge, and convert datasets across YOLO, COCO, and Pascal VOC formats with built-in utilities

Use Cases

• Object Detection and Tracking: Detect and track objects in images and video streams with zone-based counting and analysis
• Data Pipeline Automation: Load and prepare datasets from multiple sources, split for training, and convert between annotation formats
• Real-Time Video Analysis: Process video streams with detections, apply custom visualizations, and extract metrics like dwell time or speed estimation

Who Is It For

supervision is designed for computer vision engineers, data scientists, and developers who need reliable, production-grade tools for building detection and segmentation pipelines. Whether you're prototyping with notebooks or deploying at scale, supervision provides the foundational utilities to accelerate development.