metadata

license: apache-2.0
language:
  - ru
tags:
  - ocr
  - htr
  - handwritten text recognition
  - near duplicate detection
  - reuse detection
pretty_name: HWR200
size_categories:
  - 10K<n<100K

HWR200: New open access dataset of handwritten texts images in Russian

This is a dataset of handwritten texts images in Russian created by 200 writers with different handwriting and photographed in different environment.

How to download

pip install huggingface_hub
apt-get install git-lfs
git clone https://huggingface.co/datasets/AntiplagiatCompany/HWR200

Description

Total size is 44G
Total number of images with text is 30030
Number of writers is 200
Every handwritten text is photographed in three different ways: scanned, in poor light, in good light
Different authors could write the same texts
Some texts are "reuses" . they have copies of sentences from other texts

Annotation example

// for original texts:
{
    sentences: [{id: <id>, text: <sentence>}, ...],
    words_count: <word count>,
    full_text: <full text>
}

// for reuse texts:
{
    reuse_0: {
        sentences: [{id: <id>, text: <sentence>}, ...],
        id: <original text file name>
        intersection_score: <intersection_score>
    }
    reuse_1: {  // if exists
        sentences: [{id: <id>, text: <sentence>}, ...],
        id: <original text file name>
        intersection_score: <intersection_score>
    }
    start clear sentences: [<sentence>, <sentence>, ...]  // if exists
    end clear sentences: [<sentence>, <sentence>, ...]  // if exists
    words_count: <word count>
    full_text: <full text>
}

// for fpr texts:
{
    sentences: [{id: <id>, text: <sentence>}, ...],
    words_count: <word count>,
    full_text: <full text>
}