Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
The dataset viewer is not available for this split.
Server error while post-processing the rows. Please report the issue.
Error code:   RowsPostProcessingError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Abkhaz Cyrillic Dataset for PaddleOCR

A comprehensive ground-truth dataset for training PaddleOCR Recognition models on the Abkhaz language (ISO 639-1: ab) in Cyrillic script.

This dataset is a core component of the digital infrastructure project for the Abkhaz language, developed by Marat Ionov in collaboration with the Bagrat Shinkuba Fund (Фонд Баграта Шинкуба).

Dataset Summary

  • Total pairs: 37,514 (Image + Transcription)
  • Archive format: .zip containing .tif (300 DPI images) and .txt (plain text transcriptions)
  • Synthetic data: 25,944 samples (~69%)
  • Real data: 11,570 samples (~31%)

🌟 Key Feature: Unified Annotation Pipeline

The images and transcriptions in this dataset were directly derived from the .tif and .box files originally created for the Abkhaz Cyrillic OCR Dataset (Tesseract). This means the ground truth was manually verified and corrected only once, and then reused across different OCR engines. This guarantees exceptional annotation quality and consistency throughout the entire Abkhaz digitalization ecosystem.

Orthographic Period Handling

The dataset explicitly covers both historical periods of Abkhaz Cyrillic orthography:

  • 1954–1996: Includes legacy letters Ҧ ҧ, Ҕ ҕ, and the back apostrophe.
  • 1996–Present: Includes modern replacements Ԥ ԥ, Ӷ ӷ. Real data is distributed as evenly as possible between these two periods, reflecting the natural availability of historical archives.

Data Composition & Methodology

1. Real Data (Scans)

  • Sources: Digitized books, newspapers, and journals from apsnyteka.org (see Appendix B for the full list).
  • Characteristics: 300 DPI TIFFs. Includes single words, multi-line fragments, and paragraphs. Varies from clean grayscale to aged yellowed paper with natural scanning artifacts.

2. Synthetic Data

  • Fonts: 38 carefully selected fonts with full Abkhaz Cyrillic support (see Appendix A).
  • Purpose: Ensures the model sees every rare character combination and ligature, compensating for the long-tail distribution of historical texts.

3. Special Characters

The charset also includes Roman numerals (Latin I, V, X, L, etc.) and superscripts/subscripts. While recognized reasonably well on clean data, they remain a challenging area for noisy historical documents.

⚠️ Limitations & Recommendations for Future Work

The current dataset has a ~69% synthetic / ~31% real split. While the synthetic data provides excellent character coverage, real-world document degradation is underrepresented.

Recommendation: Future iterations of this dataset should aim for a 70/30 split in favor of real data. Specifically, curators should oversample historical documents containing Roman numerals and indices, which are currently a weak point for models trained on this distribution.

Intended Uses

  • Training or fine-tuning PaddleOCR Recognition models for Abkhaz Cyrillic.
  • Benchmarking line-level OCR quality on Northwest Caucasian languages.
  • Transfer learning for other low-resource Cyrillic scripts.

Data Fields

The archive contains pairs of files with identical base names:

  • *.tif: Grayscale or binary image of a text line or fragment.
  • *.txt: Plain UTF-8 text file containing the exact transcription of the image.

Citation

If you use this dataset in your research, please cite:

@misc{abkhaz_cyrillic_paddle_dataset_2024,
  author = {Ionov, Marat and Bagrat Shinkuba Fund},
  title = {Abkhaz Cyrillic Dataset for PaddleOCR},
  year = {2024},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/Grundik/abkhaz-cyrillic-dataset-paddle}
}

Author & Acknowledgments

Dataset compiled, augmented, and curated by Marat Ionov as part of the digital preservation initiative, supported by the Bagrat Shinkuba Fund.

License

This dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).


Appendix A: Fonts Used for Synthetic Data (38)

Andika, Carlito, Carlito Bold, Carlito Italic, Carlito Bold Italic, Cormorant Garamond, DejaVu Sans, Fira Code, Fira Mono, Fira Sans, Liberation Sans, Liberation Serif, Lora, Merriweather, Noto Sans, Noto Sans Mono, Noto Serif, Noto Serif Display, Open Sans, Open Sans Bold, Open Sans Italic, Open Sans Bold Italic, PT Mono, PT Sans, PT Sans Italic, PT Serif, PT Serif Bold, Roboto, Charis, Charis Bold, Charis Italic, Charis Bold Italic, IBM Plex Serif, IBM Plex Serif Bold, IBM Plex Serif Italic, IBM Plex Serif Bold Italic.

Appendix B: Source Materials (Real Data)

Real data samples were extracted from the following digitized archives (http://apsnyteka.org/):

  • Aguazhba_sost_Abkhazia_otchy_dom_I_2020.pdf
  • agu_gazeta_5_2017.pdf
  • Aidgylara_1989_1.pdf
  • Aidgylara_1989_2.pdf
  • Aidgylara_1989_3.pdf
  • Aidgylara_1990_1.pdf
  • Aidgylara_1990_2.pdf
  • Aidgylara_1990_3.pdf
  • Aidgylara_1990_4.pdf
  • Aidgylara_1990_5.pdf
  • Akua_Suhum_zhurnal_2004_1.pdf
  • ankvab_v_abhazskie_novelly_abh_1968.pdf
  • ankvab_v_stanovlenie_i_razvitie_abhazskoy_prozy_abh_1979.pdf
  • Aps_kapsh_№5_1956.pdf
  • Asarkial-Kolokol_2001_1_abh.pdf
  • Asarkial-Kolokol_2001_2_abh.pdf
  • Asarkial-Kolokol_2001_3_abh.pdf
  • Asarkial-Kolokol_2001_4_abh.pdf
  • Bol_kornya_2003.pdf
  • bzyb_1989_№155.pdf
  • bzyb_1994_№45.pdf
  • bzyb_1995_№3-4.pdf
  • bzyb_2002_№5.pdf
  • dzhibladze_g_posledny_iz_ushedshikh_1987.pdf
  • gogua_a_sukhum_akua_2002.pdf
  • Gunba_M_Ocherki_istorii_Abkhazii_1982_abh.pdf
  • iskusstvo_abhazii_1979_1_6.pdf
  • istoria_SSSR_1954_abh.pdf
  • kapba_r_esli_my_udostoimsia_2002.pdf
  • khagush_b_apsny_atoponimika.pdf
  • Kharania_Vitaly_Krovavye_sledy_2005.pdf
  • Kogonia_Suzar-ipa_V_Lezhachy_kamen_tyazhel_1997_abh.pdf
  • Kultura_i_zhizn_1995-1997.pdf
  • malia_e_narodnaya_odezhda_abkhazov_2003_abkh.pdf
  • Neissiakaemy_rodnik_narodnye_predaniya.pdf
  • rannie_zapisi_abhazskogo_folklora_iz_rukopisei_a_genko_2001.pdf
  • sh_tropami_k_vershinam_1975_abh.pdf
  • Vazhneyshie_Prazdniki_Pravoslavnoy_Tserkvi.pdf
  • vozvraschenie_2001.pdf
Downloads last month
53