---
language:
- ks
- ur
task_categories:
- image-to-text
tags:
- ocr
- synthetic
- kashmiri
- urdu-script
- nastaliq
pretty_name: Koshur SynthOCR
configs:
- config_name: mixed
  data_files:
  - split: train
    path: data/mixed-*.parquet
---

# Koshur SynthOCR — mixed split first publish

This is the mixed split of Koshur SynthOCR, published first so the Hugging Face Dataset Viewer can show image + label rows while the larger sentence, paragraph, page, and full configs continue processing.

Published/repacked: **2026-06-17**
HF repo: `Omarrran/koshur-synthocr`

## Mixed split

- Expected rows after Narqalam removal: **119,114**
- Removed Narqalam rows: **35,110**
- Expected Parquet shards: **60**
- Kept fonts:
  - Afan Koshur Naksh: **35,236**
  - Gulmarg Nastaleeq: **83,878**
- Excluded font: **Nakash Narqalam**

## Columns

- `image`: rendered OCR image as a Hugging Face `Image` feature.
- `text`: exact Unicode OCR label / target transcription.
- `sample_type`: `mixed`.
- `font_family`: `Afan Koshur Naksh` or `Gulmarg Nastaleeq`.
- `source_run_id`: original generation run id.
- `image_path`: original relative image path.
- `width`, `height`: image dimensions in pixels.
- `text_chars`: Unicode character count of the label.
- `line_count`: line count of the label.

## Loading

```python
from datasets import load_dataset

ds = load_dataset("Omarrran/koshur-synthocr", "mixed", split="train")
print(len(ds))
print(ds[0]["image"], ds[0]["text"])
```

The complete multi-config README will be replaced after all splits finish.