Quickstart: Annota Studio

This quickstart takes you from an empty organization to an exported, annotated image dataset. It takes about 15 minutes with a handful of images. You need an account and an organization: see Create your account.

1. Create an image dataset

In Annota Studio, select Datasets in the sidebar.
Select Create dataset.
Enter a Dataset name, for example Product photos.
Under Dataset type, choose Images.
Select Create.

2. Upload images

In the new dataset, select Upload assets.
Drag your images onto the page, or click the drop area to browse. PNG, JPEG, WebP, and GIF images are accepted.
Select Upload followed by the number of files, for example Upload 12 files.
When the uploads finish, select View dataset.

For formats and size limits, see Upload images and documents.

3. Add labels

Labels are the categories you want to find, such as person or car. Use category names, not proper names: Magic Select and Magic Annotate use label names to know what to look for.

On the dataset page, find the Labels card and select Add labels.
In Manage labels, choose a color, type a Label name, and select Add. Repeat for each label.
Close the dialog.

See Labels.

4. Annotate your images

On the dataset page, select Annotate > Editor. The editor opens on the first image.
Choose Bounding Box in the toolbar and drag from one corner of an object to the opposite corner. For an exact outline, choose Polygon and click around the object, then click the first point again to close it.
Pick a label for the shape. Annotations are saved automatically.
When the image is done, set its review status to Approved (shortcut 3).
Move to the next image with the next asset arrow at the top of the editor, and repeat.

To speed things up, use Magic Select to outline an object with one click, or Magic Annotate to find every instance of the selected labels in an image or in the whole dataset. These tools use credits. See Annotate images and Automatic annotation.

5. Save a dataset version

A version is an immutable copy of the dataset that exports and training can rely on, even after you keep editing.

Select Back to dataset in the editor, then open the Versions tab.
Select Create version, then Create version again in the dialog.
Wait until the version’s status is Ready. It is created in the background, so you can leave the page.

See Dataset versions.

6. Export the dataset

Open the Exports tab and select Create export.
Under Source, choose the version you just created.
Under Format, choose COCO.
Select Start export.
When the status is Ready, select the download button on the row. Download the archive before the expiration date shown on the row.

Exports do not use credits. See Export datasets.

Optional: train a model Beta

If your organization has access to the models private beta, you can train an object detection model on the version you saved. Training uses only approved images.

Select Models in the sidebar, then Create model.
Enter a Model name, choose Object detection, and select Create model.
Open the model, select Train new version, choose your Image dataset and Dataset version, and start the run.
Follow progress with View training. A successful run adds a model version you can download.

See Models and Train a model.