image-text_kurrent-xix-lines

Line-level variant of dh-unibe/image-text_kurrent-xix at revision 20743e9f6982460b243e014091626da43e2b57f7. One row per transcribed line — the cropped line image and its transcription — so the material can be read without a PageXML parser or a line segmenter.

Lines	622298
Characters	26008983
Source	dh-unibe/image-text_kurrent-xix
Source revision	20743e9f6982460b243e014091626da43e2b57f7

What the source reported while being read: pool: 23808 pages, 622303 transcribed lines, 26009031 chars, 275 distinct characters, 11596 pages skipped, 76862.1 MB, 606 over-wide lines (0.10 %), worst aspect 83:1

The crops are cut by the same code that feeds this project's training runs (cropping.write_crops in thodel/training-atr-models), so a line here is byte-identical to the line a trainer sees.

Columns

image — the line crop, JPEG bytes. text — its transcription. page — the PageXML the line came from. line_index — position in the cropping order. project — the source project, where the source records one. source_repo, source_revision — the pinned origin of every row.

Deliberately narrower than dh-unibe/data-towerbooks-textlines, which also carries line_id, region_id and two reading orders: those live in the PageXML and this project's sample does not keep them, so publishing them would mean inventing them.

---

The source dataset's own description

Carried over unchanged from dh-unibe/image-text_kurrent-xix:

Dataset Card for image-text_kurrent-xix

This dataset was created using pagexml-hf converter from Transkribus PageXML data.

Dataset Summary

This dataset contains 158525 samples across 1 split(s).

Projects Included

• MM_1_001
• MM_1_002
• MM_1_003
• MM_1_004
• MM_1_005
• MM_1_006
• MM_1_007
• MM_1_008
• MM_1_009
• MM_1_010
• MM_1_011
• MM_1_012
• TEST_CITlab_Bassermann_0_4
• TEST_CITlab_Bassermann_Manuscripts
• TEST_CITlab_Bassermann_Manuscripts_0_2
• TEST_CITlab_Binder_Kochbuch_2
• TEST_CITlab_Escher_M1
• TEST_CITlab_Gusbeth
• TEST_CITlab_Handschriftliche_Archivquellen
• TEST_CITlab_Konzilsprotkolle_B_Schwartz_5_2018
• TEST_CITlab_MargareteSick
• TEST_CITlab_MargareteSick20180731
• TEST_CITlab_MargareteSick20180731a
• TEST_CITlab_MÜLLER
• TEST_CITlab_Protokoll_Hoftheater_1806
• TEST_CITlab_Rehlen_1834_a
• TEST_CITlab_RvE_Barlaam_HS_D_
• TEST_CITlab_Steiner
• TEST_CITlab_Suppes
• TEST_CITlab_Suppes_3500
• TEST_CITlab_Tagebuch_Arnold_v1
• TEST_CITlab_umkc_Roland_M1
• TEST_CITlab_umkc_Roland_M2
• TRAIN_CITlab_Bassermann_0_4
• TRAIN_CITlab_Bassermann_Manuscripts
• TRAIN_CITlab_Bassermann_Manuscripts_0_2
• TRAIN_CITlab_Binder_Kochbuch_2
• TRAIN_CITlab_Escher_M1
• TRAIN_CITlab_Gusbeth
• TRAIN_CITlab_Handschriftliche_Archivquellen
• TRAIN_CITlab_Konzilsprotkolle_B_Schwartz_5_2018
• TRAIN_CITlab_MargareteSick
• TRAIN_CITlab_MargareteSick20180731
• TRAIN_CITlab_MargareteSick20180731a
• TRAIN_CITlab_MÜLLER
• TRAIN_CITlab_Protokoll_Hoftheater_1806
• TRAIN_CITlab_Rehlen_1834_a
• TRAIN_CITlab_RvE_Barlaam_HS_D_
• TRAIN_CITlab_Steiner
• TRAIN_CITlab_Suppes
• TRAIN_CITlab_Suppes_3500
• TRAIN_CITlab_Tagebuch_Arnold_v1
• TRAIN_CITlab_umkc_Roland_M1
• TRAIN_CITlab_umkc_Roland_M2
• hufeland_privatbesitz_1829
• nn_msgermqu2124_1827
• nn_msgermqu2345_1827
• parthey

Dataset Structure

Data Splits

• train: 158525 samples

Dataset Size

• Approximate total size: 14843467.55 MB
• Total samples: 158525

Features

• image: Image(mode=None, decode=False)
• xml_content: Value('string')
• filename: Value('string')
• project_name: Value('string')

Data Organization

Data is organized as parquet shards by split and project:

data/
├── <split>/
│   └── <project_name>/
│       └── <timestamp>-<shard>.parquet

The HuggingFace Hub automatically merges all parquet files when loading the dataset.

Usage

from datasets import load_dataset

# Load entire dataset
dataset = load_dataset("dh-unibe/image-text_kurrent-xix") 

# Load specific split
train_dataset = load_dataset("dh-unibe/image-text_kurrent-xix", split="train")