metadata
{}
Description
The FUNSD (Form Understanding in Noisy Scanned Documents) dataset is a benchmark for form understanding.
Content
- 199 scanned forms
- Entities annotated with 4 categories: header, question, answer, other
- OCR text + layout information available
- Forms are noisy and semi-structured
- Tests ability to capture hierarchical relationships between entities
Intended Use
- NER/KIE for form documents
License
Provided by FUNSD dataset authors (2019).