metadata
dataset_info:
features:
- name: image
dtype: image
- name: question
dtype: string
- name: answer
dtype: string
splits:
- name: train
num_bytes: 6334040094.246579
num_examples: 380451
- name: validation
num_bytes: 1583514185.7534213
num_examples: 95113
download_size: 7871901125
dataset_size: 7917554280
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
task_categories:
- image-to-text
- visual-question-answering
language:
- dv
tags:
- thaana
- dhivehi
- ocr
pretty_name: ocr_thaa
license: apache-2.0
Dhivehi Single-Line Text-Image Dataset
A collection of synthetic Dhivehi text images for training and evaluating text-image / vision models etc.
Each image contains a single line of Dhivehi text with various visual styles and augmentations (Check the config field for more info on the row).
Note: This dataset is a subset from alakxender/dhivehi-vrd-images.