README.md

rupindersingh1313/30_8_2025_dataset

Dataset Description

This dataset contains Punjabi OCR data with page images and their corresponding text annotations, ready for machine learning applications.

Dataset Summary

• Language: Punjabi (pa-IN)
• Script: Gurmukhi
• Total Pages: 769
• Source: Generated using Punjabi OCR annotation pipeline
• Format: Image-annotation pairs with original JSON annotations

Dataset Splits

• Train: 615 samples
• Validation: 76 samples
• Test: 78 samples

The dataset is split into train/validation/test sets with an 80/10/10 ratio by default:

• Training set for model training
• Validation set for hyperparameter tuning and model selection
• Test set for final evaluation

Dataset Structure

Each row contains:

• image: The page image (PNG format, high resolution)
• annotation: Complete OCR annotation in JSON format (as string)

The annotation JSON contains the original structure with:

• Document metadata (language, script, image dimensions)
• Text hierarchy (regions, lines, words)
• Bounding box coordinates for all text elements
• Complete text transcription

Usage

from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("rupindersingh1313/30_8_2025_dataset")

# Access different splits
train_data = dataset["train"]
val_data = dataset["validation"] 
test_data = dataset["test"]

# Iterate through training data
for sample in train_data:
    image = sample["image"]
    annotation = json.loads(sample["annotation"])  # Parse JSON annotation
    print(f"Image shape: {image.size}")
    print(f"Annotation keys: {list(annotation.keys())}")

Annotation Format

The annotation field contains JSON with this structure:

{
  "document": {
    "id": "doc_001",
    "language": "pa-IN",
    "script": "Gurmukhi",
    "image": {"width": 2481, "height": 3507, "dpi": 300}
  },
  "hierarchy": {
    "regions": [
      {
        "region_id": 1,
        "type": "text_block",
        "polygon": [x1, y1, x2, y2, ...],
        "lines": [
          {
            "line_id": 1,
            "polygon": [x1, y1, x2, y2, ...],
            "words": [
              {
                "word_id": 1,
                "text": "ਪੰਜਾਬੀ",
                "polygon": [x1, y1, x2, y2, ...]
              }
            ]
          }
        ]
      }
    ]
  }
}

Use Cases

This dataset is suitable for:

• OCR Model Training: Train custom OCR models for Punjabi text
• Text Detection: Develop text region detection algorithms
• Document Layout Analysis: Analyze document structure and layout
• Multilingual NLP: Include Punjabi in multilingual language models
• Research: Academic research in OCR and document processing

Data Quality

• High-resolution images (300 DPI)
• Accurate text transcriptions
• Precise bounding box annotations
• Consistent formatting and structure
• Quality-controlled annotation process

License

Please ensure proper attribution when using this dataset. Contact the dataset creators for commercial use permissions.

Citation

If you use this dataset, please cite:

@dataset{punjabi_ocr_dataset,
  title={Punjabi OCR Dataset - rupindersingh1313/30_8_2025_dataset},
  author={Generated using Punjabi OCR Pipeline},
  year={2025},
  url={https://huggingface.co/datasets/rupindersingh1313/30_8_2025_dataset},
  note={High-quality Punjabi OCR dataset with images and annotations}
}

Contact

For questions, issues, or contributions, please contact the dataset maintainers.