Hugging Face's logo

Datasets:
Sigurdur
/
timarit-ocr

Tasks:
Image-to-Text
Modalities:
Image
Text
Formats:
parquet
Languages:
Icelandic
Size:
1K - 10K
Tags:
ocr
icelandic
historical-documents
newspapers
document-layout
timarit
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
Community
1
timarit-ocr / README.md
Sigurdur's picture
Sigurdur
Update README.md
d4cffaf verified
preview code
|
raw
history blame contribute delete
7.22 kB
metadata 
dataset_info:
  features:
    - name: image
      dtype: image
    - name: issue_id
      dtype: string
    - name: page_name
      dtype: string
    - name: publication_id
      dtype: string
    - name: year
      dtype: int32
    - name: ocr
      dtype: string
  splits:
    - name: train
      num_bytes: 327307277
      num_examples: 1000
  download_size: 325933814
  dataset_size: 327307277
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: apache-2.0
task_categories:
  - image-to-text
language:
  - is
tags:
  - ocr
  - icelandic
  - historical-documents
  - newspapers
  - document-layout
  - timarit
  - research
size_categories:
  - n<1K
pretty_name: Timarit OCR

Timarit OCR

Dataset Description

This dataset contains page-level images extracted from historical Icelandic newspapers and periodicals sourced from timarit.is, Iceland's national digital archive of newspapers and magazines. Each sample pairs a page image with structured OCR output, providing ground-truth data for Icelandic historical documents.

OCR transcriptions were generated using rednote-hilab/dots.ocr, a state-of-the-art document OCR model.

Dataset Summary

  • Language: Icelandic (is)
  • Source: timarit.is — Iceland's National and University Library digital archive
  • OCR Model: rednote-hilab/dots.ocr
  • Format: Images (PNG) with structured JSON OCR output
  • Primary Use Case: OCR training and evaluation for historical Icelandic text

Supported Tasks

  • OCR / Image-to-Text: Pairs of document images and their transcribed text for training or evaluating OCR models on historical Icelandic typography.
  • Document Layout Analysis: Page-level images suitable for detecting text regions, columns, and other layout elements typical of historical Nordic newspaper printing.

Dataset Structure

Data Fields

Field Type Description
image Image Page-level scan of the source document (505px–3370px wide)
imagewidth int Width of the image in pixels
issue_id string Identifier for the specific issue (1–6 chars)
page_name string Name of the page within the issue (178 unique values)
publication_id string Identifier for the publication (1–4 chars)
year int32 Publication year
ocr string Structured JSON OCR output from dots.ocr, containing detected text regions

The ocr column contains a JSON-serialized list of detected text blocks. Each block has the following structure:

Key Type Description
box list[int] Bounding box of the text region as [x1, y1, x2, y2]
text string Transcribed text content of the region
category string Layout category (e.g. "Title", "Text")
confidence float Model confidence score for the detection (0.0–1.0)

Example ocr entry:

[
  {"box": [366, 136, 1060, 502], "text": "FJÁRMÁLA FTIRLITIÐ", "category": "Title", "confidence": 1.0},
  {"box": [496, 721, 868, 760], "text": "Leiðbeinandi tilmæli", "category": "Text", "confidence": 1.0}
]

To parse the ocr column in Python:

import json
from datasets import load_dataset

ds = load_dataset("Sigurdur/timarit-ocr")
ocr_blocks = json.loads(ds["train"][0]["ocr"])
for block in ocr_blocks:
    print(block["category"], block["text"])

Data Splits

Split Size
train 1000

Source Data

About timarit.is

timarit.is is the digital newspaper and periodical archive maintained by the National and University Library of Iceland (Landsbókasafn Íslands – Háskólabókasafn). It contains digitized issues of Icelandic newspapers and magazines spanning from the 19th century to the present, making it the primary resource for historical Icelandic printed text.

Collection Process

Pages were sourced from publicly available scans on timarit.is. Images were extracted at the page level and OCR transcriptions were generated using rednote-hilab/dots.ocr.

Copyright & Licensing

This dataset is released under the Apache 2.0 License. Attribution is required — if you use this dataset, you must cite the author (see citation below).

Source Material Copyright

The source material in this dataset is drawn from timarit.is, the digital archive of the National and University Library of Iceland. Under Icelandic copyright law, a work enters the public domain 70 years after the death of the author. Not all documents in this dataset can be guaranteed to meet this threshold.

This dataset is intended solely for research purposes. Users must not use this dataset or its contents for commercial purposes. It is the responsibility of the user to verify the copyright status of any specific document before use beyond non-commercial academic research. The dataset author assumes no liability for copyright infringement arising from downstream use.

Usage 

import json
from datasets import load_dataset

ds = load_dataset("Sigurdur/timarit-ocr")
sample = ds["train"][0]

# View image metadata
print(sample["publication_id"], sample["issue_id"], sample["year"])

# Parse OCR blocks
ocr_blocks = json.loads(sample["ocr"])
for block in ocr_blocks:
    print(block["category"], block["text"])

Intended Use

This dataset is intended for:

  • Training and fine-tuning OCR models on historical Icelandic text
  • Evaluating OCR performance on low-resource Nordic language documents
  • Research into historical document understanding and layout analysis for Icelandic

Limitations & Biases

  • Coverage is limited to publications available on timarit.is and may not represent all historical Icelandic print media.
  • Historical documents may contain archaic spelling, ligatures, and fonts that differ significantly from modern Icelandic text.
  • Scan quality varies across publications and time periods, which may affect model performance.
  • OCR transcriptions were generated automatically by dots.ocr and may contain errors, particularly on degraded or low-resolution scans.

Citation

If you use this dataset, please cite:

@dataset{sigurdur2025timaritocr,
  author    = {Sigurdur Sigurdarson},
  title     = {Timarit OCR: Historical Icelandic Newspaper OCR Dataset},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Sigurdur/timarit-ocr}
}

You may also want to cite the OCR model used to generate transcriptions:

@misc{rednote2025dotsocr,
  author    = {rednote-hilab},
  title     = {dots.ocr},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/rednote-hilab/dots.ocr}
}

Related Datasets