Reference Data

The reference data includes a Ground Truth corpus and other special corpora. The Ground Truth corpus contains pages from publications printed between 1500 and 1900. The content of the corpus is based on a particular selection from the holdings of the DFG project “German Text Archive”, the Digitized Collections of the Staatsbibliothek zu Berlin and the Wolfenbüttel Digital Library of the Duke August library. The holdings of projects and digital collections of other libraries as well as additional Ground Truth data, which are compiled together with module projects, can be included in the corpus as special extensions in concertation with the OCR-D coordination project. If additional annotations or texts are necessary, these can be created in consultation with the OCR-D coordination project.

Depth of Annotation, Text Accuracy and Artifacts

The Ground Truth corpus offers three annotation depths:

Overview (The list will be extended continuously.)

The special corpora contain:

Creation of the Ground Truth

The image data were first subjected to a layout analysis (text region and line recognition) using Transkribus and then segmented automatically. The automatically recognized text as well as the lines and text regions were corrected manually. Finally the data in form of PAGE files, digital images and METS files were zipped as a BagIt file.

If you are interested in further Ground Truth data (e.g. for binarization) please contact us: hinrichsen[at]

The data are subject to a CC-BY-SA license, for the use of the image data different licenses may exist. Please contact the project and/or the owning library.

OCR-D Research Data Repository

The OCR-D research data repository collects all versions of documents and (intermediate) results created during the document analysis. It contains at least the end results of every processed document. During the ingest much metadata about the document will be extracted and made available for search/filter (e.g. identifier(s), title, classification(s), genre(s), semantic label(s), used processor(s), text). In future, there may also be export options for specific tools.

OCR-D Ground Truth Repository

Similarly, there is a publicly available OCR-D repository, which contains all ground truth data provided by OCR-D.