Add Annotations

Annotations tell the VLM what to look for or answer. For phrase grounding datasets, annotations are bounding boxes with text phrases. For visual question answering datasets, annotations are question-answer pairs. Each annotation pair consumes Data Rows from your organization's quota. You need at least 20 annotated images to start a training run, though 100 or more gives better results. For task-specific annotation walkthroughs, see the detailed guides for phrase grounding, VQA, and freeform text.

Choose the method that fits your situation.

Upload existing annotation files

Use this method if you already have labeled data in COCO, YOLO, or another supported format.

1

Open the Annotations tab

In the Dataset Explorer, click the Annotations tab to access the upload area.

Next →

You should see

Your annotations are uploaded when you see a successfully finished job.

Create annotations in the annotator

Use this method if you need to label images from scratch.

1

Open the Annotator tab

Go to the Dataset Overview page and click the Annotator tab to open the labeling interface.

Next →

You should see

Your annotations are ready when you see annotation count matching the image count, and heatmaps showing annotation patterns.

Next steps