metadata

license: apache-2.0
task_categories:
  - text-to-image
  - image-to-text
language:
  - th
tags:
  - handwriting-recognition
  - ocr
pretty_name: Thai Handwriting Dataset
size_categories:
  - 10K<n<100K
maintainer: Kobkrit Viriyayudhakorn ([email protected])
dataset_info:
  features:
    - name: image
      dtype: image
    - name: text
      dtype: string
    - name: label_file
      dtype: string

Thai Handwriting Dataset

This dataset combines two major Thai handwriting datasets:

BEST 2019 Thai Handwriting Recognition dataset (train-0000.parquet)
Thai Handwritten Free Dataset by Wang (train-0001.parquet onwards)

Dataset Description

BEST 2019 Dataset

Contains handwritten Thai text images along with their ground truth transcriptions. The images have been processed and standardized for machine learning tasks.

Wang Dataset

Exclusively focuses on handwritten sentences in Thai language
Contains 4,920 unique sentences covering various topics and themes
Created from contributions by 2,026 users, ensuring diverse handwriting styles
Encompasses various linguistic patterns, vocabulary, and sentence structures

Dataset Structure

The dataset is provided in parquet file format with the following columns:

image: Image data (Image type)
text: Ground truth transcription of the handwritten text (String)
label_file: Source label file name (String)

Usage

This dataset is ideal for:

Handwriting recognition
Optical character recognition (OCR)
Natural language processing (NLP)
Language generation

Researchers, developers, and enthusiasts can utilize this dataset to:

Develop and benchmark algorithms
Train machine learning models
Explore innovative techniques in Thai language analysis and handwriting recognition

Original Datasets

BEST 2019 Thai Handwriting Recognition competition: https://thailang.nectec.or.th/best/best2019-handwrittenrecognition-trainingset/
Thai Handwritten Free Dataset by Wang: Data Market https://www.wang.in.th/dataset/64abb3e951752d79380663c2