UROCR

Dataset Description

Urdu Optical Character Recognition (OCR) dataset compiled from multiple sources. The dataset contains 40,769 image-text pairs suitable for training and evaluating Urdu OCR models.

Languages

• Urdu (ur)

Dataset Structure

Data Instances

Each instance in the dataset contains:

• image: A PIL Image object containing Urdu text
• text: The ground truth text transcription in Urdu

Example:

{
    'image': <PIL.Image.Image>,
    'text': 'یہ اردو متن کی مثال ہے'
}

Data Fields

• image (Image): The image containing Urdu text (various resolutions)
• text (string): The corresponding Urdu text transcription

Data Splits

The dataset is split into three subsets:

Split	Number of Samples
Train	32,615 (80.0%)
Validation	4,077 (10.0%)
Test	4,077 (10.0%)
Total	40,769

Dataset Characteristics

Image Types

• Scanned documents
• Printed text images
• Synthetic text images
• Natural scene text

Text Characteristics

• Script: Urdu (Nastaliq and Naskh styles)
• Content: Various domains including literature, news, religious texts, and general text
• Text Length: Varies from single words to full sentences

Usage

Loading the Dataset

from datasets import load_dataset

# Load the entire dataset
dataset = load_dataset("mahwizzzz/urocr")

# Load specific split
train_dataset = load_dataset("mahwizzzz/urocr", split="train")
test_dataset = load_dataset("mahwizzzz/urocr", split="test")