Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,158 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- multilingual
|
| 5 |
+
- af
|
| 6 |
+
- am
|
| 7 |
+
- ar
|
| 8 |
+
- as
|
| 9 |
+
- azb
|
| 10 |
+
- be
|
| 11 |
+
- bg
|
| 12 |
+
- bm
|
| 13 |
+
- bn
|
| 14 |
+
- bo
|
| 15 |
+
- bs
|
| 16 |
+
- ca
|
| 17 |
+
- ceb
|
| 18 |
+
- cs
|
| 19 |
+
- cy
|
| 20 |
+
- da
|
| 21 |
+
- de
|
| 22 |
+
- du
|
| 23 |
+
- el
|
| 24 |
+
- en
|
| 25 |
+
- eo
|
| 26 |
+
- es
|
| 27 |
+
- et
|
| 28 |
+
- eu
|
| 29 |
+
- fa
|
| 30 |
+
- fi
|
| 31 |
+
- fr
|
| 32 |
+
- ga
|
| 33 |
+
- gd
|
| 34 |
+
- gl
|
| 35 |
+
- ha
|
| 36 |
+
- hi
|
| 37 |
+
- hr
|
| 38 |
+
- ht
|
| 39 |
+
- hu
|
| 40 |
+
- id
|
| 41 |
+
- ig
|
| 42 |
+
- is
|
| 43 |
+
- it
|
| 44 |
+
- iw
|
| 45 |
+
- ja
|
| 46 |
+
- jv
|
| 47 |
+
- ka
|
| 48 |
+
- ki
|
| 49 |
+
- kk
|
| 50 |
+
- km
|
| 51 |
+
- ko
|
| 52 |
+
- la
|
| 53 |
+
- lb
|
| 54 |
+
- ln
|
| 55 |
+
- lo
|
| 56 |
+
- lt
|
| 57 |
+
- lv
|
| 58 |
+
- mi
|
| 59 |
+
- mr
|
| 60 |
+
- ms
|
| 61 |
+
- mt
|
| 62 |
+
- my
|
| 63 |
+
- 'no'
|
| 64 |
+
- oc
|
| 65 |
+
- pa
|
| 66 |
+
- pl
|
| 67 |
+
- pt
|
| 68 |
+
- qu
|
| 69 |
+
- ro
|
| 70 |
+
- ru
|
| 71 |
+
- sa
|
| 72 |
+
- sc
|
| 73 |
+
- sd
|
| 74 |
+
- sg
|
| 75 |
+
- sk
|
| 76 |
+
- sl
|
| 77 |
+
- sm
|
| 78 |
+
- so
|
| 79 |
+
- sq
|
| 80 |
+
- sr
|
| 81 |
+
- ss
|
| 82 |
+
- sv
|
| 83 |
+
- sw
|
| 84 |
+
- ta
|
| 85 |
+
- te
|
| 86 |
+
- th
|
| 87 |
+
- ti
|
| 88 |
+
- tl
|
| 89 |
+
- tn
|
| 90 |
+
- tpi
|
| 91 |
+
- tr
|
| 92 |
+
- ts
|
| 93 |
+
- tw
|
| 94 |
+
- uk
|
| 95 |
+
- ur
|
| 96 |
+
- uz
|
| 97 |
+
- vi
|
| 98 |
+
- war
|
| 99 |
+
- wo
|
| 100 |
+
- xh
|
| 101 |
+
- yo
|
| 102 |
+
- zh
|
| 103 |
+
- zu
|
| 104 |
+
task_categories:
|
| 105 |
+
- image-to-text
|
| 106 |
+
tags:
|
| 107 |
+
- ocr
|
| 108 |
+
size_categories:
|
| 109 |
+
- 1M<n<10M
|
| 110 |
+
---
|
| 111 |
+
# Synthdog Multilingual
|
| 112 |
+
|
| 113 |
+
<!-- Provide a quick summary of the dataset. -->
|
| 114 |
+
|
| 115 |
+
The Synthdog dataset created for training in [Centurio: On Drivers of Multilingual Ability of Large Vision-Language Model](https://gregor-ge.github.io/Centurio/).
|
| 116 |
+
|
| 117 |
+
Using the [official Synthdog code](https://github.com/clovaai/donut/tree/master/synthdog), we created >1 million training samples for improving OCR capabilities in Large Vision-Language Models.
|
| 118 |
+
|
| 119 |
+
## Dataset Details
|
| 120 |
+
|
| 121 |
+
We provide the images for download in two `.tar.gz` files. Download and extract them in folders of the same name (so `tar xvzf images.tar.gz -C images; tar xvzf images.tar.gz -C images_non_latin`).
|
| 122 |
+
The image path in the dataset expects images to be in those respective folders for unique identification.
|
| 123 |
+
|
| 124 |
+
|
| 125 |
+
Every language has the following amount of samples: 500,000 for English, 7,500 for non-Latin scripts, and 5,000 otherwise.
|
| 126 |
+
|
| 127 |
+
Text is taken from Wikipedia of the respective languages. Font is `GoNotoKurrent-Regular`.
|
| 128 |
+
|
| 129 |
+
|
| 130 |
+
> Note: Right-to-left written scripts (Arabic, Hebrew, ...) are unfortunatly writte correctly right-to-left but also bottom-to-top. We were not able to fix this issue. However, empirical results in Centurio suggest that this data is still helpful for improving model performance.
|
| 131 |
+
>
|
| 132 |
+
|
| 133 |
+
|
| 134 |
+
|
| 135 |
+
## Citation
|
| 136 |
+
|
| 137 |
+
<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
|
| 138 |
+
|
| 139 |
+
**BibTeX:**
|
| 140 |
+
|
| 141 |
+
```
|
| 142 |
+
@article{centurio2025,
|
| 143 |
+
author = {Gregor Geigle and
|
| 144 |
+
Florian Schneider and
|
| 145 |
+
Carolin Holtermann and
|
| 146 |
+
Chris Biemann and
|
| 147 |
+
Radu Timofte and
|
| 148 |
+
Anne Lauscher and
|
| 149 |
+
Goran Glava\v{s}},
|
| 150 |
+
title = {Centurio: On Drivers of Multilingual Ability of Large Vision-Language Model},
|
| 151 |
+
journal = {arXiv},
|
| 152 |
+
volume = {abs/2501.05122},
|
| 153 |
+
year = {2025},
|
| 154 |
+
url = {https://arxiv.org/abs/2501.05122},
|
| 155 |
+
eprinttype = {arXiv},
|
| 156 |
+
eprint = {2501.05122},
|
| 157 |
+
}
|
| 158 |
+
```
|