Add Annotations

📍

Step 3 of 3: Add Annotations

Part of the dataset preparation quickstart. Next: Train your model.

Annotations teach your VLM what to detect or understand. Upload existing annotations or create new ones using our visual annotator.

⏱️ Time: ~5-10 minutes (varies by dataset)

💡

Need detailed guidance?

This is a quickstart overview. For comprehensive annotation guides:

• Upload annotations guide — All formats and requirements
• Annotate for Phrase Grounding — Object detection
• Annotate for VQA — Question answering

• AI-assisted tools — Speed up annotation

Choose your method

Upload existing annotations

Best for: When you already have annotated data in COCO, YOLO, or other formats

1. From your dataset Explorer page, click the Annotations tab

2. Click Upload in the Upload Annotations section

3. Select your annotation format

Supported formats:

• Phrase Grounding: Vi JSONL, COCO, Pascal VOC, YOLO Darknet, YOLO Keras PyTorch, CSV Four Corner, CSV Width Height
• Visual Question Answering: Vi JSONL only
• Freeform: Vi JSONL (coming soon)

View format specifications →

4. Upload your annotation files

📘

Important

Annotation files must reference images by filename. Ensure filenames match exactly.

Detailed format guide →

Common questions

What's next?

✅

Dataset ready!

Your images are annotated and ready for training.

Train Your VLM →

Create a training workflow, configure your VLM, and start fine-tuning your model.