AlexandreSheva/rukopys-curated-mvp

RUKOPYS Curated MVP: Ukrainian Handwriting Recognition Dataset

Task-ready curated derivative of UkrainianCatholicUniversity/rukopys for Ukrainian handwritten document AI.

This dataset turns the raw RUKOPYS release into reproducible training artifacts for full-page vision-language fine-tuning, crop-level handwriting transcription, and layout detection. It was created to support an end-to-end HTR pipeline: curation, model training, inference, evaluation.

What This Dataset Enables

• Full-page image-to-JSON supervised fine-tuning.
• Crop-level handwritten text transcription.
• YOLO-format layout detection experiments.
• Region-level analysis and filtering.
• Reproducible model training and dataset packaging.

Contents

• metadata.jsonl: normalized page records with image metadata and region annotations.
• regions.jsonl: one row per annotated region.
• vlm_sft.jsonl: crop-level transcription examples.
• page_sft.jsonl: full-page image-to-structured-JSON examples.
• yolo/: YOLO-format layout detection dataset.
• crops/: cropped region images, if exported.
• _pack_manifest.json: tar-shard manifest for packed Hub uploads.

Stats

• Pages: 2715
• Regions: 39420
• Crop SFT examples: 38832
• Page SFT examples: 2330

Sources

• archive: 178
• dictation: 509
• school: 1782
• university: 246

Region Types

• annotation: 1145
• formula: 6174
• graph: 61
• handwritten: 31193
• image: 130
• printed: 477
• table: 240

Notes

Quality weights are assigned by annotation source:

• annotator: 1.0
• volunteer: 0.75
• auto: 0.35

Recommended Uses

• Fine-tuning VLMs for Ukrainian handwritten page parsing.
• Training layout detectors for handwritten document regions.
• Benchmarking OCR/HTR post-processing with structured outputs.
• Demonstrating a reproducible document AI pipeline from raw data to model artifacts.

Source Dataset

This dataset is derived from UkrainianCatholicUniversity/rukopys.

@dataset{rukopys_2026,
  title        = {RUKOPYS}: Ukrainian Handwritten Text Recognition Dataset,
  author       = {Dmytro Voitekh and Volodymyr Zmiivskyyi and Oleksii Molchanovskyi},
  organization = {Ukrainian Catholic University},
  year         = {2026},
  license      = {CC BY-NC-SA 4.0},
  url          = {https://huggingface.co/datasets/UkrainianCatholicUniversity/rukopys},
  note         = {First large-scale Ukrainian HTR dataset; from 1920s archival documents to 2025 school homework and exams}
}