Zero-shot Extraction

A technique that relies on the vision-language capabilities of foundation models to directly extract corresponding values based solely on field definitions or schema descriptions provided by the user, without requiring any labeled samples.