icfhr-2016 / README.md
SSamDav's picture
Update README.md
b3444ac verified
metadata
dataset_info:
  features:
    - name: image
      dtype: image
    - name: alto
      dtype: string
    - name: filename
      dtype: string
  splits:
    - name: train
      num_bytes: 440129470
      num_examples: 350
    - name: val
      num_bytes: 58958626
      num_examples: 50
    - name: test
      num_bytes: 59181301
      num_examples: 50
  download_size: 554065486
  dataset_size: 558269397
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: val
        path: data/val-*
      - split: test
        path: data/test-*
license: cc-by-4.0
language:
  - de
tags:
  - handwriting-text-recognition
  - htr
  - early-modern-german
  - historical-documents
pretty_name: HTR Dataset ICFHR 2016

HTR Dataset ICFHR 2016

Dataset Description

The HTR Dataset ICFHR 2016 is a historical document image dataset primarily intended for Handwriting Text Recognition (HTR) research and benchmarking.

This dataset was created as part of the European Commission's READ project (Recognition and Enrichment of Archival Documents) and consists of a subset of documents from the Ratsprotokolle collection, which contains minutes of council meetings held from 1470 to 1805.

The documents are written in Early Modern German and feature complex handwriting, making this a challenging resource for HTR algorithms. It was notably used in the HTR competition held at the ICFHR 2016 conference.

Data Structure

The dataset is distributed across two main archives: Train-And-Val-ICFHR-2016.tgz and Test-ICFHR-2016.tgz.

The training and validation sets comprise 400 pages. The ground-truth transcription is provided at the line level within the standard PAGE XML format (not plain text transcripts). Each data instance generally includes a page image and the corresponding PAGE XML file.

Dataset Creation

Source Data

The data originates from the Ratsprotokolle collection (minutes of council meetings from 1470 to 1805).

Creators

The primary creators of this specific dataset compilation are: A.H. Toselli, V. Romero, M. Villegas, E. Vidal, and J.A. Sánchez.

License

The dataset is licensed under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license.

Citation

When using this dataset, please cite the original Zenodo record as follows:

@inproceedings{ToselliRomeroVillegasVidalSanchez2018,
  author    = {Toselli, A. H. and Romero, V. and Villegas, M. and Vidal, E. and S\'{a}nchez, J. A.},
  title     = {HTR Dataset ICFHR 2016},
  publisher = {Zenodo},
  year      = {2018},
  doi       = {10.5281/zenodo.1164045},
  url       = {[https://zenodo.org/records/1164045](https://zenodo.org/records/1164045)}
}