Organize your data, your way. Create as many datasets as you need without worrying about limits.

Choose from a dozen different formats when exporting your training data and refer back to previous versions to track your progress.

For each export, get a curl link that can be pasted directly into your notebooks and download from the command line.

Keep datasets securely stored and archived indefinitely. Free-Tier Users: Your datasets are retained for 90 days.

Divide your data into sets (train, validate, and test) so you have everything you need to evaluate your model performance after training.

Upload JPG, PNG, BMP, and TIF image files.

Upload and export JSON, XML, CSV, and TXT annotation formats.

Upload video files and Roboflow will extract frames based on the sample rate you specify.

Detailed statistics and charts containing helpful insights into the quality of your dataset and how to improve it.

View a breakdown of all the classes in your dataset and the number of annotations that appear in each.

A dot chart displaying all the sizes of your images along with a breakdown of all the various aspect ratios.

A heatmap showing the location of all the annotations in your dataset so you can see where they are most commonly positioned.

Auto-orient, resize, grayscale, and auto-contrast.

Static crop, tile, class modification, and filter null.

Random flip, rotate, crop, shear, brightness, exposure, blur, and noise.

Like cutout and mosaic.

Random flip, rotate, crop, shear, brightness, exposure, blur, and noise applied independently to each bounding box.

Generate up to 50 augmented versions of each source image in your dataset. Free-tier Users: Generate up to three augmented versions.

CVAT, VoTT, LabelImg, LabelMe, RectLabel, VGG Image Annotator, Scale, Labelbox, SuperAnnotate, Google Human Labeling, and Amazon SageMaker Ground Truth, and more.

Direct integrations with hosted computer vision model training services including Amazon Rekognition, Google Cloud Vision AutoML, Azure Custom Vision.

Train your model with one click using Roboflow Train. Within 24 hours, you'll get results including mAP, precision, and recall, as well as a hosted API for inference.

Share datasets with other members of your team and seamlessly collaborate on projects.

Host your dataset publicly by publishing free public versions of your datasets so others can experiment with your data.

Can't find the answers you need in our documentation or on the Roboflow Blog? Just send us an email to help@roboflow.com and we'll get you answers — fast.

Your questions and requests will be bumped to the top of the queue and answered by computer vision experts on our team.

A complete guide walking you through how to use Roboflow and how to get the most out of our tools.

Helpful tutorials on creating and training models along with interviews with leading computer vision experts.