Hugging Face's logo

Datasets:
Process-Venue
/
Marathi_Handwritten

Tasks:
Image Classification
Image-to-Text
Image Feature Extraction
Modalities:
Image
Text
Formats:
parquet
Languages:
Marathi
Size:
1K - 10K
Tags:
handwriting-recognition
OCR
Marathi
image-processing
deep-learning
computer-vision
Libraries:
Datasets
pandas
License:
Dataset card Data Studio Files Community
Marathi_Handwritten / README.md
processvenue's picture
processvenue
Update README.md
4f1a718 verified
preview code
|
raw
history blame contribute delete
3.63 kB
metadata
license: mit
task_categories:
  - image-classification
  - image-to-text
  - image-feature-extraction
tags:
  - handwriting-recognition
  - OCR
  - Marathi
  - image-processing
  - deep-learning
  - computer-vision
pretty_name: Marathi Handwritten OCR
size_categories:
  - 1K<n<10K
language:
  - mr

Dataset Card for Marathi Handwritten OCR Dataset

Dataset Description

  • Repository: processvenue/Marathi_Handwritten
  • Total Examples: 2520
  • Splits:
    • train: 2016 examples
    • validation: 504 examples
  • Features: image, id, filename, label, type
  • Size: 2520 images
  • Language: mr
  • License: mit

Dataset Summary

The Marathi Handwritten Text Dataset is a collection of handwritten text images in Marathi (देवनागरी लिपी), aimed at supporting the development of Optical Character Recognition (OCR) systems, handwriting analysis tools, and language research.The dataset was curated from native Marathi speakers to ensure a variety of handwriting styles and character variations. The dataset contains 2520 images with two categories:

  • बाराखडी (Barakhadi/syllables): 560 characters
  • शब्द (Words): 1949 words
  • Image Format: PNG
  • Label Format: Text file
  • Image Dimensions: 115x57 OR 114x58 pixels
  • Storage Format: Parquet (Hugging Face)

Language Statistics:

  • Marathi is the third most spoken language in India
  • Approximately 83 million speakers (7% of India's population)
  • Official language of Maharashtra
  • One of the 22 scheduled languages of India

Applications

  1. OCR Development
  • Handwriting recognition systems
  • Document digitization
  • Text extraction tools
  1. Educational Tools
  • Language learning applications
  • Writing practice systems
  • Digital literacy programs
  1. Research Applications
  • Script analysis
  • Language processing
  • Pattern recognition studies

Data Fields

  • image: Image feature (PIL Image)
  • id: Feature field
  • filename: Feature field
  • label: Feature field
  • type: Feature field

Dataset Creation

The dataset was created through careful curation of handwritten samples from native Marathi speakers. The dataset includes both character-level (syllables) and word-level annotations. Contributors were provided with pre-written text designed to capture:

  • Common character variations
  • Diverse writing styles
  • Standard word formations
  • Typical punctuation usage

The images were carefully processed and stored in PNG format, while metadata and labels were structured in a Parquet file format.

Usage Example 

from datasets import load_dataset
import matplotlib.pyplot as plt

# Load the dataset
dataset = load_dataset("processvenue/Marathi_Handwritten")

# Get an example from train split
example = dataset['train'][0]

# Display the image
plt.figure(figsize=(5, 5))
plt.imshow(example['image'])
plt.title(f"Sample Image")
plt.axis('off')
plt.show()

Citation

If you use this dataset in your research, please cite:

@misc{sanskrit-ocr-dataset,
  title     = {Sanskrit OCR Dataset},
  author    = {ProcessVenue},
  year      = {2024},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/datasets/processvenue/Sanskrit-OCR-Dataset},
  organization = {ProcessVenue},
  contact   = {sales@processvenue.com},
  Website  = {https://www.processvenue.com/}
}

References 

@misc{Sarode_Marathi_Handwritten_Text,
author = {Hrushikesh Sarode},
title = {Marathi Handwritten Text Dataset},
year = {n.d.},
url = {https://www.kaggle.com/datasets/hrushikeshsarode/marathi-handwritten-text-dataset},
note = {Accessed: February 12, 2025}
}