Your AI annotation platform, hosted in France
The computer vision annotation suite powered by CVAT, installed and maintained by DINAO. Your images, videos, and datasets stay on our French servers — AI-assisted annotation, on a sovereign infrastructure.
What is CVAT?
CVAT (Computer Vision Annotation Tool) is an open-source web platform for annotating images, videos, and 3D point clouds, designed to produce training datasets for computer vision. It covers all annotation types — bounding boxes, polygons, polylines, segmentation masks, keypoints/skeletons, cuboids, ellipses, tags — and handles tracking and interpolation for video.
Designed for teams, CVAT provides project and task management, quality control (ground-truth, confusion matrix), analytics, role management (RBAC), as well as AI-assisted annotation via models like SAM/SAM 2, YOLO, or Mask R-CNN. Export covers over 20 standard formats (COCO, YOLO, Pascal VOC…).
Technically, CVAT is built on Django and React, with PostgreSQL and Redis, deployed as a multi-container Docker stack. Auto-annotation relies on Nuclio serverless, and an NVIDIA GPU is recommended for the heaviest models. A REST API, Python SDK, and CLI complete the suite.
Host CVAT at DINAO
Resource tiers compatible with CVAT prerequisites (minimum 2 vCPU / 4 Go / 20 Go). Hosted in France, fully managed.
This application uses an unofficial/unverified image or requires custom resources. We review and deploy it upon request.
Request a quoteThis application uses AI
The container hosts the application, not the AI engine (which requires dedicated GPUs) : inference is handled externally, with your own provider key. Prioritize a sovereign engine — Mistral AI, NumSpot, Scaleway, or OVHcloud AI Endpoints (France, GDPR) ; an international provider (OpenAI, Anthropic…) only if a specific capability requires it. Inference subscriptions are not included in hosting.
Technical details
You might be wondering…
What is CVAT used for?
CVAT is a computer vision annotation suite. You annotate images, videos, and 3D point clouds to produce the training datasets required for AI detection, segmentation, or classification models.
Does AI-assisted annotation run on your servers or an external cloud?
In the self-hosted edition, auto-annotation models (SAM, YOLO…) run locally within your DINAO instance, without calling any external cloud service. Your images never leave the server — see the sovereignty note below.
What export formats are supported?
Over 20 standard formats : COCO, YOLO, Pascal VOC, and many others. You retrieve your datasets in the format expected by your training framework, via the interface, API, or Python SDK.
Do I need a GPU?
Manual annotation works without a GPU. For auto-annotation by large models (especially SAM), an NVIDIA GPU is strongly recommended : it is available as an option (Team AI plan) or dedicated (Production plan).
Can I change plans or export my data?
Yes. You can upgrade or downgrade at any time, and your datasets remain exportable in standard formats — no proprietary lock-in.
