Dataset Types
Explore the different dataset types available in Datature Vi for training vision-language models.
Dataset types define the structure and purpose of your vision AI training data. Datature Vi supports multiple dataset types to accommodate different vision-language model tasks, from object localization to conversational image understanding.
Available dataset types
Localize objects in images using natural language descriptions. Perfect for flexible object detection without pre-defined categories.
Answer questions about images in natural language. Ideal for conversational image understanding and analysis.
🚧 Coming soon — Define custom annotation schemas for specialized use cases and research projects.
Choosing the right dataset type
Select your dataset type based on your application requirements:
| Dataset Type | Best For | Output | Flexibility |
|---|---|---|---|
| Phrase Grounding | Object localization with natural language | Bounding boxes with locations | Medium - predefined structure |
| VQA | Question-answering about images | Natural language answers | Medium - Q&A format |
| Freeform | Custom annotation requirements | User-defined formats | High - fully customizable |
Common use cases by type
Phrase Grounding applications
- Robotics — Locate objects using natural descriptions
- Image editing — Select regions with text commands
- Autonomous vehicles — Identify objects with flexible queries
- Warehouse automation — Find items using natural language
Visual Question Answering applications
- Quality inspection — Ask questions about defects
- Accessibility — Describe images for visually impaired users
- Content moderation — Query image content
- Inventory management — Get information through questions
Freeform applications
🚧 Coming soon
- Research projects — Novel computer vision tasks
- Medical imaging — Custom diagnostic annotations
- Scientific imaging — Domain-specific labels
- Hybrid requirements — Complex multi-modal annotations
Getting started
Ready to create your dataset? Follow these steps:
Select the dataset type that matches your use case
Upload images and annotations
Fine-tune a VLM on your dataset
Learn more
- Phrase Grounding explained — Deep dive into visual grounding
- Visual Question Answering explained — Complete VQA guide
- Freeform explained — Custom annotation schemas (coming soon)
- Create a dataset — Step-by-step dataset creation
- Concepts overview — Back to core VLM concepts
Need help?
We're here to support your VLMOps journey. Reach out through any of these channels:
Updated about 1 month ago