---
{}
---
**Description**  
The **[FUNSD](https://guillaumejaume.github.io/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](https://guillaumejaume.github.io/FUNSD/work/) (2019).