Datasets:
Create README.md
Browse filesrussian_docs_bypage is a collection of scanned Russian‑language documents split into individual pages.
The dataset contains 9,995 page images (around 3.1 GB in total) and is stored in the imagefolder format, which is directly compatible with the Hugging Face datasets library.
Sources
All documents were collected from openly accessible Russian‑language PDF sources on the web.
The pages in this dataset primarily come from:
Scientific preprints and technical reports hosted on Russian preprint servers (for example, the preprint archive of the Lebedev Physical Institute and similar portals).
Research papers and academic materials made publicly available by Russian research institutions and universities.
Official publications, reports, and methodological documents from Russian organizations that publish their materials as PDF files.
Only Russian‑language PDFs were used when rendering pages into images; no English‑language documents are included in this dataset.
Data format
Modality: document page images (PNG/JPEG, depending on the original PDF and renderer settings).
Granularity: one example corresponds to one page; each page is stored as a separate image file.
Split: a single train split with 9,995 examples.
Storage format: imagefolder on the Hub (internally converted to Parquet for efficient loading).
Example usage:
python
from datasets import load_dataset
ds = load_dataset("zimble/russian_docs_bypage", split="train")
image = ds[0]["image"] # PIL.Image page
Intended use
This dataset is intended for:
Training and evaluation of OCR models and Cyrillic text recognition systems.
Document understanding and layout‑aware vision models on Russian‑language content.
Building RAG and document‑retrieval pipelines where text is first extracted from scanned pages.
The dataset does not contain any text annotations or labels; only rendered page images are provided.