Roboflow + CVAT
Pre-label images in CVAT with a Roboflow Universe model, then bring the finished labels into Roboflow to train and deploy your own model.
How to connect CVAT to Roboflow
6 stepsPick a model on Roboflow Universe
Find a Universe project that has a trained model and copy its URL. Copy your Roboflow API key as well.
Add the model in CVAT
On CVAT's Models page, add a model, paste the Universe URL and your API key into the Roboflow integration form, and save.
Run the model on your task
Open the task, choose the Roboflow model from the AI tools or automatic annotation dialog, map its classes to your CVAT labels and run it. Review and correct the boxes it draws.
Export the task dataset
Save your work, then export the task dataset as COCO 1.0. PASCAL VOC 1.1 and YOLO 1.1 also work.
Upload to Roboflow
Create a Roboflow project and drag the exported images and annotation files onto the upload page. Roboflow reads the format and keeps your classes.
Generate a version and train
Generate a dataset version with any preprocessing and augmentation you want, then train a model such as RF-DETR and deploy it.
What teams build with CVAT
Start from a model instead of a blank image
Run a Universe model in CVAT so labelers correct predictions rather than draw every box by hand.
Keep labeling in CVAT
Your team keeps its CVAT workflow and sends finished tasks to Roboflow for training.
Train on CVAT labels without a conversion script
Upload a COCO export and Roboflow converts it to whatever format your model needs.
Questions
COCO 1.0 is a good default because it carries class names and supports boxes and polygons in one file. Roboflow also reads PASCAL VOC and YOLO exports.
CVAT lists the Roboflow integration for CVAT Online and CVAT Enterprise. CVAT's docs mark it as not supported on the self-hosted community edition.
CVAT says semi-automatic annotation with a Roboflow model is available on every plan, including Free. One-click automatic annotation of a whole task is a paid CVAT feature.
Models hosted on Roboflow Universe, including your own if you trained them in Roboflow. CVAT's blog says the integration supports classification, object detection and segmentation models.
Why are companies choosing Roboflow?
The end-to-end platform for vision AI: create datasets, train models and deploy to production.
Flexible deployment modes
Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.
Integrated workflow builder
Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.
Model monitoring and alerts
Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.
