Dataset Card for OCR-Google_Books

A line-to-text dataset for Tibetan OCR.

Dataset Details

Dataset Structure

• Features:

  • id: Image file identifier
  • label: Text transcription
  • image: Image of a line of Tibetan text

• Splits:

  • Train: 601,152 samples (37.3M characters)
  • Eval: 75,136 samples (4.7M characters)
  • Test: 75,168 samples (4.7M characters)

Uses

Direct Use

• Training and evaluation of Tibetan OCR models
• Multi-script OCR development
• Comparative analysis of modern vs. traditional printing methods
• Large-scale OCR model pretraining

Out-of-Scope Use

• Not be suitable for handwritten Tibetan texts

Dataset Creation

Curation Rationale and Process

This dataset was created to support the development of robust OCR systems for Tibetan literature, encompassing both modern typography and traditional woodblock printing methods. The inclusion of multiple scripts and printing techniques makes it valuable for training models that can handle diverse Tibetan textual sources.

The dataset is constructed from Google Books scans of Tibetan texts, with Line-level image-text pairs extracted from scanned pages

Usage

from datasets import load_dataset

# Load training split
dataset = load_dataset("openpecha/OCR-Google_Books", split="train")

# Example features
print(dataset[0])
# {'id': 'I1KG1163750042_0025',
#'label':'ཡིན་པས་ཆབ་སྲིད་དང་འབྲེལ་བ་བྱུང་བ་ཙམ་ལ་ངོ་མཚར་དགོས་དོན་གང་',
#'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=860x45>}

Dataset Contact

BDRC - help@bdrc.org