CVAT Computer Vision Annotation Platform

CVAT began in 2017 as an internal annotation tool at Intel, was released as open source on GitHub in 2018 and spun out in 2022 as the independent company CVAT.ai, with offices in Wilmington, Delaware and Paphos, Cyprus. It gives computer vision teams a widely used tool for labeling images, video and now audio.

CVAT comes in three editions: CVAT Online, a managed cloud service with free and paid plans; CVAT Enterprise, a self-hosted edition with SSO, role-based access and support; and the open-source CVAT Community. Annotators can use model-assisted labeling with SAM 2, SAM 3, Ultralytics, Hugging Face or custom models, review workflows with ground truth and honeypot jobs, quality reports, webhooks, cloud storage connections and a Python SDK and API. CVAT.ai also runs labeling services with 300+ annotators across 12 time zones.

• Image and video labeling: web editor for computer vision data in automotive, geospatial, infrastructure and logistics projects.
• Audio annotation: multimodal support added in 2026.
• Auto-labeling: SAM 2, SAM 3, Ultralytics, Hugging Face, Roboflow or custom models.
• Quality control: manual review, ground truth and honeypot jobs with quality reports.
• Developer tools: REST API, Python SDK, webhooks and cloud storage connectors.
• Flexible deployment: managed cloud, self-hosted enterprise or open-source community edition.

Vietnamese annotation outsourcing companies can run client projects on self-hosted CVAT Enterprise with SSO and audit controls. Traffic and smart city integrators can label camera footage for vehicle and license plate models. Manufacturers and agritech firms can build defect or crop datasets with SAM-assisted labeling at low cost.

CVAT is backed by a large community, with 15K+ GitHub stars, 2.8M+ Docker Hub downloads and 84K+ monthly SDK installs, which reduces vendor lock-in risk. Teams can start free, move to paid online plans from USD 23 per user per month billed yearly, or self-host on-premise or air-gapped to keep sensitive images in-house. Model-assisted labeling with SAM and custom models cuts manual polygon work.

Open-source tooling dominates computer vision labeling as companies seek to avoid lock-in and keep data on their own servers, while foundation segmentation models make semi-automatic labeling standard. Vietnam's large annotation outsourcing workforce and growing in-house AI teams in manufacturing, traffic and retail favor low-cost, self-hostable tools.

Started at Intel in 2017, open-sourced in 2018 and spun out as CVAT.ai in 2022. Offices in Wilmington, Delaware (US) and Paphos (Cyprus). Labeling services with 300+ annotators. States GDPR, CCPA and EU AI Act compliance, with encryption at rest and in transit.