CVAT

What CVAT does

CVAT (Computer Vision Annotation Tool) is a leading platform for building high-quality visual datasets for vision AI. It offers open-source, cloud, and enterprise products, as well as labeling services, for image, video, and 3D annotation with AI-assisted labeling, quality assurance, team collaboration, analytics, and developer APIs. CVAT helps companies build future-proof computer vision models by transforming raw, unstructured visual data into high-quality datasets that enable AI to solve real human problems and drive digital transformation. It began as an internal tool at Intel in 2017, was released as open source in 2018, and spun out into an independent company in 2022.

Products

• CVAT OnlineCloud-based data annotation platform for solo labelers and teams, offering various plans including Free, Solo, Team, and Enterprise. It supports image, video, and 3D data types, project management, team collaboration, data management, API access, data export, quality control, reporting, and automation features.
• CVAT for EnterprisesA self-hosted solution for organizations prioritizing enhanced security, compliance, and control over their data. It can be deployed within the customer's own infrastructure and offers advanced features like SSO, role-based access controls, audit logs, custom feature development, and priority expert support.
• Professional Data Annotation & Labeling ServicesManaged labeling services for high-volume audio transcription, speech labeling, and visual data annotation. It includes a free pilot project, detailed proposals, production labeling, quality assurance, and dataset delivery, leveraging a team of 300+ annotators.

Key capabilities

• Import data from local files and cloud storage (Amazon S3, Azure Blob Storage, Google Cloud Storage, S3-compatible buckets)
• Manage projects, tasks, jobs, and team access (roles, permissions, stages, statuses, task ownership)
• Label data with manual and automated tools (boxes, polygons, masks, skeletons, tags, tracks)
• Speed up labeling with SAM 2, SAM 3, Ultralytics, Hugging Face models, or custom models
• Support detection, segmentation, tracking, pose estimation, and other CV workflows
• Validate annotations and control quality (validation, acceptance stages, ground truth, honeypots, consensus workflows)
• Monitor task progress and team workload (progress tracking, workload monitoring, throughput measurement)
• Export datasets in 20+ formats (COCO, YOLO, KITTI, Cityscapes, Pascal VOC)
• Automate import, task management, and export through API, SDK, and CLI
• Build custom integrations around CVAT data labeling and ML workflows
• Team collaboration with organization creation and project sharing
• Internal storage and cloud storage integration
• API Access for programmatic functionality
• Data export of annotations only or annotations & images
• Manual review and QA
• Ground truth job and Honey Pot for quality control
• Reports & analytics for performance & monitoring
• SSO (Single Sign-On) support (SAML, OIDC)
• Role-based access controls
• Audit logs
• Webhooks
• Internal AI agent calls (AI Tools)
• External AI agent calls (Hugging Face, Roboflow)
• Automatic annotation on a batch of images
• Integration with Hugging Face & Roboflow models
• SAM 2 and SAM 3 for image segmentation
• SAM 2 for video segmentation and tracking
• Community support on GitHub and Discord
• Download quality report as a confusion matrix (CSV)
• More reliable uploads and task creation (TUS uploads, manifest files with videos)
• Faster imports (Ultralytics YOLO Classification) and stable exports (3D annotation download)
• Smoother annotation UI (reduced redundant calls)
• CLI agent resilience (robust handling of queue-related HTTP requests)
• Security fixes (XSS vulnerabilities, OPA upgrade, staff permissions)
• GDPR & CCPA compliant
• Snap to Point for Polygon and Polyline Editing
• Join Tool for Polygons
• Use Existing Labels as Text Prompts for SAM 3 (class-guided segmentation)
• Docker Compose support for SAM 2 Agent and YOLO
• Cloud Storage as Backing Storage for Local Tasks
• Download Projects, Tasks, and Jobs Lists as CSV
• Simplify polygons and polylines faster
• Bounding boxes as default prompt for SAM models
• Default backing cloud storage for new tasks
• Public visibility for native functions in the CLI
• More practical tools for task migration
• Annotation layers for dense scenes (Layer stack view, sorting, grouping, filtering)
• Flexible label and skeleton editing (delete attributes, add/edit/delete skeleton element attributes)
• New filters for skeleton annotations (by sub-label, element attributes, element occlusion)
• Safer annotation import options (replace or append)
• LDAP support for user management
• Blurring images for task previews
• Image, video, and 3D point clouds annotations
• Custom feature development for Enterprise plans
• Deployment options: on-prem, in VPC, or fully air-gapped
• Security updates and reports for Enterprise plans
• Private helpdesk support for critical issues
• Access to experts and training/consultation for Enterprise plans
• EU AI Act compliant

Use cases

• Building high-quality visual datasets for vision AI
• Image annotation
• Video annotation
• 3D annotation
• Object detection
• Object segmentation
• AI-assisted labeling
• Quality assurance for datasets
• Team collaboration on annotation projects
• Analytics for annotation progress and workload
• Developer APIs for integration
• Model training for AI-powered inventory counting solutions
• Instance segmentation workflows
• Supporting data annotation and model training for AI-powered intelligence platforms (e.g., SIYTE)
• Tracking attention-critical areas in video ads for marketing effectiveness (e.g., Dragonfly AI)
• Reviewing computer vision detections of animals in camera trap image datasets (e.g., e2m)
• Structuring large-scale annotations for equipment detection on inspection photos (e.g., tbmaestro)
• Transforming CCTV footage from maritime environments into training data for AI detection systems (e.g., for vessel safety)
• Labeling sports images to build high-quality ground truth for AI analysis of performance
• Model-assisted annotation using YOLO and SAM2 pre-annotation
• Analyzing misclassifications in spreadsheets with confusion matrix CSV
• Automated detection and counting of asbestos fibers in scanning-electron-microscope images
• Training perception models for autonomous trucking (e.g., LiDAR point cloud annotation)
• Early detection of invasive species in freshwater lakes (e.g., zebra mussels)
• High-volume audio transcription
• Speech labeling for voice AI
• Multi-language transcription
• Multi-speaker transcription with speaker-separated tracks
• Transcriptions for separate letters, words or whole phrases
• Voice activity detection (VAD) style segmentation
• Speaker turn segmentation for diarization
• Sound event segmentation (SED) for non-speech audio events
• Assigning categorical labels to speaker tracks or segments (single-label, multi-label tagging, attributes)
• Speech recognition datasets
• Call analytics
• Voice-enabled devices
• Accessibility workflows
• Captions, transcripts, and speaker separation datasets for meeting search
• In-cabin speech data labeling for noisy environments
• Subtitles and searchable audio archives
• Speech datasets for dictation and clinical audio
• Audio event segmentation and classification for underwater monitoring

AI approach

CVAT.AI provides a platform for building high-quality visual datasets for vision AI. It offers tools for data annotation, labeling, and quality assurance, with AI-assisted labeling features. The platform integrates with various AI models and allows users to build custom integrations. It supports detection, segmentation, tracking, and pose estimation workflows. CVAT.AI also offers professional labeling services for image, video, 3D, and audio data.

Tech named: SAM 2, SAM 3, Ultralytics, Hugging Face models, YOLO, OpenCV, Roboflow, OPA, LiDAR

Industries served

• Software Development
• Computer Vision
• Machine Learning
• Data Science
• AI
• Robotics
• Autonomous Driving
• Industrial AI
• Lab automation
• Environmental AI
• Underwater computer vision
• Autonomous Trucking
• Logistics
• Marketing
• Sports
• Healthcare
• Marine Acoustics
• Bioacoustics
• Contact Centers & Customer Support
• Communication & Conferencing Platforms
• Consumer Devices & IoT
• Automotive & Mobility

What it says sets it apart

• Open-source foundation with a strong global community
• Comprehensive platform for image, video, and 3D annotation
• AI-assisted labeling tools (SAM 2, SAM 3, Ultralytics, Hugging Face models)
• Robust quality assurance features (ground truth, honeypots, consensus workflows)
• Flexible deployment options (open-source, cloud, enterprise self-hosted)
• Extensive data import and export formats (20+ formats)
• Developer-friendly with API, SDK, and CLI for automation and integration
• Scalability for teams of all sizes, from solo labelers to large enterprises
• Strong focus on security and compliance (SSO, RBAC, audit logs, GDPR, CCPA, EU AI Act)
• Continuous product updates and feature enhancements based on user feedback and industry needs
• Professional data annotation services available for high-volume tasks
• Originating from Intel's internal tools, built by computer vision engineers
• Deep collaboration with customers to develop specific features (e.g., FERNRIDE's 3D annotation tools)
• User-friendly UI compared to competitors mentioned by customers
• Responsive and supportive team behind the product, as highlighted by customers

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.