cvat-ai/cvat

CVAT is a leading data annotation platform for image, video, audio, and 3D datasets. It offers open-source, cloud, and enterprise products, as well as labeling services, with AI-assisted labeling, quality assurance, team collaboration, analytics, and developer APIs.

What it solves

CVAT is a professional data annotation platform designed to streamline the creation of high-quality datasets for computer vision and other AI models. It eliminates the manual effort of labeling images, videos, audio, and 3D point clouds by providing a centralized, collaborative environment for teams to manage and review annotations.

How it works

CVAT operates as a self-hosted server (deployed via Docker) that provides a web-based interface for annotators. Users can manually label data using various tools (bounding boxes, polygons, masks, etc.) or use "auto-labeling" by connecting ML models (like SAM or YOLO) via a serverless infrastructure (Nuclio) to automatically suggest labels.

Who it’s for

It is built for AI research and production teams who need to create large-scale labeled datasets, as well as organizations that want full control over their data infrastructure to ensure privacy and security.

Highlights

  • Multi-modal support: Supports annotation for images, video, audio, and 3D point clouds.
  • AI-powered labeling: Integrates with serverless ML models for automatic detection, segmentation, and tracking.
  • Enterprise-grade collaboration: Includes multi-user roles, task assignments, and quality control workflows (consensus and ground truth checks).
  • Extensive format support: Imports and exports data in 20+ common formats, including COCO, YOLO, and Pascal VOC.
  • Developer-friendly: Provides a Python SDK, CLI, and REST API for automating dataset pipelines.

Related

  • Project
    l3p-cv/lost

    A flexible, web-based framework for collaborative image annotation that supports semi-automatic pipelines to speed up the labeling of machine learning datasets.

  • Project
    open-edge-platform/datumaro

    A dataset management framework and CLI tool used to build, transform, and analyze AI datasets, specifically focusing on converting between various annotation formats.

  • Project
    wkentaro/labelme

    Labelme is a graphical image and video annotation tool that helps users create labeled datasets for computer vision tasks like segmentation and object detection, featuring AI-assisted labeling capabilities.

  • Project
    davidjurgens/potato

    Potato is a free, self‑hosted web tool for building annotation projects across text, audio, video, images, documents, and LLM/agent traces. Configured via YAML, it offers multimodal labeling, AI‑assisted suggestions, active learning, rich quality‑control metrics, QDA features, and one‑click deployment to many cloud providers.

  • Project
    opendatalab/labelU

    A multimodal data annotation platform that provides tools for labeling images, videos, and audio, featuring AI-assisted auto-annotation to accelerate dataset creation.