CVAT
Data annotation platform for labelling images, video, 3D point clouds and audio, offered as cloud, self-hosted and open source.
CVAT is a data annotation platform for building AI training datasets. CVAT is used to import images, video, 3D point clouds or audio, label them manually or with model assistance, review the results and export them in formats such as COCO or YOLO. CVAT is available as a hosted service, self-hosted software and open source.
Best for Machine learning teams that need to label visual or audio data at scale, with a choice between a hosted service and running the software themselves.
- Listed in
- Image Recognition & OCRDeveloper API
- Platforms
- Web, API
- Pricing model
- Freemium
- Free plan
- Yes
- API
- Yes
- Made by
- CVAT.ai Corporation
- Added
Last verified: . Information taken from cvat.ai.
What is CVAT?
CVAT is a data annotation platform used to prepare training datasets for computer vision and audio models. Annotators draw bounding boxes, polygons, pixel masks, keypoint skeletons and tracks on images and video, label objects in 3D point clouds, and transcribe or segment audio. Data is imported from local files or from cloud buckets such as Amazon S3, Azure Blob Storage and Google Cloud Storage, then organised into projects, tasks and jobs that are assigned to annotators and reviewers. Models including SAM 2, SAM 3, Ultralytics and Hugging Face models can pre-label data to speed up manual work. Quality is checked with review stages, ground truth jobs, honeypots and consensus. Finished datasets export in more than 20 formats. CVAT comes as a hosted service, a self-hosted enterprise edition and an open-source community edition, and the company also sells a managed labelling service.
Key Features
Multi-Format Annotation
Supports boxes, polygons, masks, skeletons, tags and tracks across images, video, 3D point clouds and audio.
Model-Assisted Labelling
Runs SAM 2, SAM 3, Ultralytics, Hugging Face or custom models to produce draft annotations.
Project and Team Management
Splits work into projects, tasks and jobs with roles, permissions, stages and assignees.
Quality Control
Measures annotation quality using ground truth jobs, honeypots, consensus and reviewer acceptance stages.
Storage Connections
Reads data from S3-compatible, Azure and Google cloud storage as well as local uploads.
Export and Automation
Exports to formats such as COCO, YOLO, KITTI and Pascal VOC, with an API, SDK and CLI for pipelines.
How CVAT works
Import data
Upload local files and existing annotations or connect a cloud storage bucket.
Organise the work
Create projects, tasks and jobs and assign annotators and reviewers.
Label
Annotate by hand or let an integrated model propose labels to correct.
Validate
Pass jobs through review and measure quality against ground truth or consensus.
Export
Download the dataset in the format your training pipeline expects.
Screenshots
Use Cases
- Driver Assistance Datasets
- Label lane markings, vehicles and LiDAR scenes for automotive perception models.
- Aerial Imagery Mapping
- Outline buildings, roads and land use zones in drone or satellite pictures.
- Infrastructure Inspection
- Mark defects on solar panels, power lines or pipelines in inspection footage.
- Warehouse Robotics
- Annotate packages, labels and obstacles seen by cameras on robots.
- Outsourced Labelling
- Hand a dataset to CVAT's own annotation team instead of staffing the work internally.
Things to know
- Plan limits on the hosted service
- CVAT Online caps projects, tasks, internal storage and monthly AI tool calls according to the plan.
- Team work needs an organisation
- Collaboration on CVAT Online requires creating an organisation and subscribing to the Team plan.
- SSO on annual plans only
- The pricing table notes that single sign-on is available only with yearly subscriptions.
- Limits may change
- The company states it can adjust limits with prior notice based on usage patterns.
FAQs
What is CVAT?
CVAT is a data annotation platform for labelling images, video, 3D point clouds and audio to create training data for AI models.
Is CVAT free?
Yes, in two forms. CVAT Community is an open-source, self-hosted edition offered at no cost, and CVAT Online includes a Free plan alongside paid Solo and Team plans.
How much does CVAT cost?
CVAT Online sells Solo and Team subscriptions with monthly or yearly billing, and Enterprise is arranged through sales. For current figures, see the official pricing page.
Can CVAT be self-hosted?
Yes. CVAT can run on your own infrastructure through the open-source Community edition or the Enterprise edition, including air-gapped deployments.
Does CVAT have an API?
Yes. CVAT provides an API, an SDK and a CLI for automating imports, task management and exports.
Which export formats does CVAT support?
CVAT exports datasets in more than 20 formats, with Pascal VOC, Cityscapes, KITTI, YOLO and COCO among them.
Who makes CVAT?
CVAT is developed by CVAT.ai Corporation, which also offers professional annotation services.
Who it's for
- Machine learning engineers
- Computer vision teams
- Annotation teams
- Researchers
- Enterprises
Comments
No comments yet. If you have used CVAT, tell other readers how it went.
Alternatives to CVAT
| Tool | Free plan | Starting price | API |
|---|---|---|---|
 Label StudioOpen source platform for labeling data and evaluating AI models with human review. |
Yes | — | Yes |
 Segments.aiData labeling platform for robotics and autonomous driving that annotates 3D point clouds and camera images together. |
— | — | Yes |
UltralyticsComputer vision platform from the makers of YOLO for annotating data, training models on cloud GPUs and deploying them. |
Yes | — | Yes |
 Digitap.aiAPI platform for banks and fintechs covering digital KYC onboarding, bank statement analysis and alternate-data credit assessment. |
— | — | Yes |
 AirbyteOpen-source data movement platform that replicates data from 700+ sources and exposes it to analytics and AI agents. |
— | — | Yes |
Listed in the same categories as CVAT. A starting price appears only where we checked it on the maker’s pricing page; a dash means not checked or not published. All Image Recognition & OCR tools