How Do I Choose a Dataset Type?
Pick the right dataset type for your task in Datature Vi. Compare phrase grounding, VQA, and freeform text by output format, annotation effort, and use case.
Your dataset type determines what your model outputs and how you annotate your data. Pick the wrong type and you will spend time annotating in a format that does not match your goal. This page helps you decide between the three types available in Datature Vi: phrase grounding, visual question answering, and freeform text.
Start with your output
The fastest way to choose a dataset type is to ask: what does my application need the model to return?
If your task could fit more than one type, start with VQA. It is the most flexible of the three. You can switch later by creating a new dataset; your images stay in your workspace.
Side-by-side comparison
Match your industry
Different industries tend to favor specific dataset types. Use these patterns as a starting point, not a rule.
Frequently asked questions
Related resources
Updated 4 months ago
