Skip to content
Branch: master
Go to file

Latest commit


Failed to load latest commit information.
Latest commit message
Commit time

OCR Evaluation Data

Authors: Dan Garrette and Hannah Alpert-Abrams

This is data that can be used to evaluate the Ocular historical document OCR system, which can be found here: It contains train/dev/test splits for several books, each with pre-extracted lines, and gold transcriptions for the dev and test sets.

Some subset of these documents were used for testing in the following publications:

Unsupervised Code-Switching for Multilingual Historical Document Transcription [pdf]
Dan Garrette, Hannah Alpert-Abrams, Taylor Berg-Kirkpatrick, and Dan Klein
NAACL 2015

An Unsupervised Model of Orthographic Variation for Historical Document Transcription [pdf]
Dan Garrette and Hannah Alpert-Abrams
NAACL 2016

If you find any mistakes in the gold transcriptions found in this repository, please let us know. We would like for the transcriptions to be as accurate as possible.

Running Ocular with the data

To train and evaluate with this data, use the following options during font training:

-inputDocPath            documents/BOOK/train
-extractedLinesPath      extractions/BOOK
-evalInputDocPath        documents/BOOK/test
-evalExtractedLinesPath  extractions/BOOK

Retrieving image files

The actual page images are not stored in this repository. Instead, the pre-extracted lines are. However, all image files in documents can be found on the Primeros Libros website. The urls for the images are formulaic, and can be determined from the image filename. The filename template


corresponds to the url template

For example, the image for filename


can be found at


This repository contains the data used to evaluate Ocular ( in our publications.



No releases published
You can’t perform that action at this time.