Published September 4, 2021 | Version 1.0
Dataset Open

Vorau Abbey library Cod. 253 dataset for Document Layout Analysis

  • 1. Pattern Recognition and Human Language Technologies

Description

VORAU-253 is a music manuscript referred to as Cod. 253 of the Vorau Abbey library, which was provided by the Austrian Academy of Sciences. It is written in German Gothic notation and dated around year 1450.

This manuscript is interesting because of the complexity of its layout, where staff, text and decorations are intertwined to
compose the structure of the document.

This database is a subset of 228 pages of the archive, using 128 randomly selected pages for training/validation and 100 for test.

The database was manually annotated into the following three layout regions:

* staff: represents the regions that contains a set of horizontal lines and spaces where each one represent a different musical pitch. This region type does not contain text lines. Hence, no baselines.

* lyrics: are the words that are sung appear below their corresponding staff, and other text in the document. In all cases, text to be sung and the other text are assigned to different layout regions under the lyrics label.

* drop-capital: is a decorated letter that might appear at the beginning of a word or text line. As it is a single big letter, it contain no text lines nor baselines.

On average each page contains 12.5 [7,23] text lines distributed over an average of 10.5[7,15] ``lyrics'' regions. Moreover, each page contains 22.3[14,28] layout regions on average.

Files

Files (769.5 MB)

Name Size Download all
md5:c3603211ba2838ffc71d69d4ccc57384
769.5 MB Download