AI-Assisted Tools

IntelliScribe is Datature Vi's AI-assisted annotation tool for phrase grounding and freeform image annotation. It reduces manual annotation work by generating text automatically: for phrase grounding, it creates captions and links phrases to bounding boxes; for freeform image annotation, it generates text content that you can edit to match your schema. You still review and refine the output, but the repetitive text work goes much faster.

IntelliScribe is available in the phrase grounding and freeform image annotators. It does not apply to VQA or video annotations.

What IntelliScribe does

IntelliScribe provides AI features inside both the phrase grounding and freeform image annotators.

IntelliScribe Caption

When you press C in the annotator, IntelliScribe analyzes the image and generates text automatically.

• In phrase grounding: The caption appears in the Phrase Grounding panel describing the main objects. You can accept it as-is or edit it with the Text tool (press T).
• In freeform image annotation: The generated text appears in the Freeform panel. Edit it to match your annotation schema before saving.

The caption generation works well on common objects in clear images. For specialized equipment or domain-specific vocabulary, the AI caption gives you a starting point. Review and edit the terminology before continuing.

IntelliScribe Phrases

When you press P, IntelliScribe reads the current caption and your drawn bounding boxes, then creates phrase-box links automatically. Each phrase in the caption gets highlighted and connected to the box that most closely matches it.

You need at least one bounding box and a caption before running IntelliScribe Phrases. Review the output: the AI is accurate on clear, distinct objects, but may mismatch phrases when objects overlap or captions are ambiguous.

IntelliScribe Phrases is available in phrase grounding only, since freeform text annotations do not use bounding boxes.

How it fits into annotation

The AI-assisted workflow runs after you draw your boxes:

1

Open the Annotator tab

From the Dataset Overview page, click the Annotator tab to open the labeling interface.

Next →

You should see

Your AI-assisted annotation is complete when each phrase in the caption is linked to the correct bounding box and all connections are reviewed.

Manual phrase grounding annotation walkthrough

When to use AI assistance

Manual vs AI-assisted annotation

The recommended approach for most projects: use IntelliScribe for speed, then edit the caption for domain accuracy before running IntelliScribe Phrases.

Tips for better AI output

The quality of IntelliScribe's output depends on the image quality and how you use it.

Image quality:

• Use clear, well-lit photos. Low-contrast or blurry images produce weaker captions.
• Front-facing angles with minimal occlusion work best.
• Higher resolution images generally yield more detailed captions.

Caption editing:

• Replace generic terms with your domain vocabulary before drawing boxes. A caption that says "integrated circuit" when you need "FPGA chip" will produce incorrect phrase links.
• Keep captions structured so phrases are distinctly separated. Overlapping phrases cannot be linked simultaneously.

After running IntelliScribe Phrases:

• Check each link. Click any highlighted phrase to confirm it connects to the right box.
• For incorrect links, press D and click the phrase to unlink it, then manually highlight the correct phrase.

Frequently asked questions

Do this with the Vi SDK

import vi

client = vi.Client(
    secret_key="your-secret-key",
    organization_id="your-organization-id"
)

result = client.annotations.upload(
    dataset_id="your-dataset-id",
    paths="annotations.jsonl",
    wait_until_done=True
)
print(f"Imported: {result.total_annotations}")

For more details, see the full SDK reference.

Related resources