---
configs:
- config_name: default
  data_files: data/*.parquet
  features:
  - name: image
    dtype: image
  - name: label
    dtype: string
language: [fa]
task_categories: [image-to-text]
tags: [ocr, persian, printed-text]
pretty_name: Persian Printed OCR 3.5M
---

# Persian Printed OCR 3.5M

A unified corpus of **3,517,974 Persian printed OCR image/text pairs**, selected from five public
datasets using GlotLID v3. Only the `accept` bucket is included; 232,317 ambiguous and 190,733
rejected rows are excluded. The viewer exposes exactly `image` and `label`.

## Sources

- `AliShafiee2003/persian-ocr-garshasp-70c` — pinned revision `36bfdcdeac20c02231f4ee08472f80db2fc467bb` (CC-BY-4.0)
- `hezarai/parsynth-ocr-200k` — pinned revision `e55cca662250585aa0278303901ef30b9b0612f9` (no license declared on its HF card)
- `mohajesmaeili/Persian_Arabic_TextLine_Image_Ocr_Medium` — pinned revision `b99becc3cadd733998c53ab1a599eb94ac7dfb99` (no license declared on its HF card)
- `mohajesmaeili/Persian_Arabic_TextLine_Image_Ocr_Small` — pinned revision `7ca7a34389bdf805465876eb54ebd3e93c5a48b1` (no license declared on its HF card)
- `ordaktaktak/Persian-OCR-230k` — pinned revision `05d248b30988b41bb905f7919c0bac39cab38d70` (MIT)

Users remain responsible for complying with each upstream source's terms. This aggregation does not
replace or broaden upstream licenses.

## Persian filter

Accepted when GlotLID predicted `fas_Arab` with probability at least 0.65, or predicted `fas_Arab`
and the text contained Persian-specific characters. The complete intake comprised 3,941,024 rows.