Document layout annotation for document AI

It labels the positions and semantic classes of page elements such as titles, text blocks, tables, figures, captions, and references so models can learn document structure.

Yes. Mark figure and caption regions with separate classes so their positions and semantic roles remain explicit.

Yes. Annotate table regions in the original page view. Define row or field labels where your training task requires more detail.

Yes. Selectable text can be used to create annotation regions. Visual tools remain available for scanned pages and graphical content.

Yes. A shared region taxonomy can be applied to varied PDFs, including reports, papers, forms, and business documents. Include representative layouts in your annotation guidelines.

Define the region taxonomy before labeling, include difficult layouts in the guidelines, and review ambiguous boundaries or overlapping content.

The native document workflow supports PDF files. Each PDF remains one work item, with annotations attached to their original page numbers as teams navigate the document.

Reviewers inspect page regions and their labels, resolve comments, and correct or reject work through the configured workflow. Shared classes and properties keep document labeling consistent.

Yes. Unitlab Unified Export Format preserves document metadata and page numbers alongside labels, properties, and supported relationships for downstream document AI pipelines.