|
--- |
|
size_categories: |
|
- 100K<n<1M |
|
license: mit |
|
language: |
|
- fi |
|
tags: |
|
- HTR |
|
- OCR |
|
configs: |
|
- config_name: default |
|
data_files: |
|
- split: train |
|
path: "final_rec_data.zip" |
|
--- |
|
|
|
# OCR training data from AIDA-project |
|
|
|
<img src='kuvat/Kuva12.png' width='500'> |
|
|
|
### Dataset Summary |
|
|
|
The zip file contains textlines and their annotations from AIDA-project. There are ~ 166k textlines that are mainly in Finnish language, but contain a little Swedish and |
|
English and little French and German textlines. The textlines contains typewritten and also handwritten lines. Roughly 24 % of the annotated lines are handwritten and the rest are |
|
typewritten. The dataset also contains 120 000 synthetic images. |
|
|
|
### Supported Tasks |
|
|
|
The dataset was created mainly for text recognition task. |
|
|
|
### Languages |
|
|
|
The majority of the textlines are in Finnish, but some are in Swedish and English. In addition to this there are few French and German textlines. |
|
|
|
## Dataset structure |
|
|
|
### Data Instances |
|
|
|
The zip file contains two folders. Folder called text_lines contains all the text lines. The other folder called annotations contain the annotations in PaddleOCR format. |
|
The annotations are divided into train, validation and test sets. In addition to this, the annotations are divided into handwritten, typewritten and ship, which contains |
|
annotations of ship records that are mainly handwritten. Handwritten and typewritten annotations are also divided into "best" and "semi" files. "Best" means that the annotator has |
|
understood every letter in the line as "semi" means that some character are not understood. |
|
|
|
### Data Fields |
|
|
|
PaddleOCR format means that the annotations are saved into a txt file containing multiple annotations. One annotations is placed per line in the file. First, the format |
|
contains a path to an image, then a separating "\t" character and then the transcription. An example of the format is shown below. |
|
|
|
``` |
|
/path/to/0001.jpg\tHello World |
|
/path/to/0002.jpg\tThis is PaddleOCR format. |
|
... |
|
``` |
|
|
|
### Data Splits |
|
|
|
Below is how the annotated data is split. The number in parantheses shows the amount of "semi" textlines. |
|
|
|
| Dataset Split | Typewritten | Handwritten | Ship Registry | |
|
| ------------- | ----------- | ----------- | ------------- | |
|
| Train | 22253 (248) | 6943 (424) | 3796 | |
|
| Validation | 4744 (9) | 1151 (25) | 469 | |
|
| Test | 4272 (3) | 1270 (16) | 472 | |
|
|
|
## Dataset Creation |
|
|
|
### Source Data |
|
|
|
The data is collected from Central Archives for Finnish Business (ELKA). It consists of various document types including letters, ship records, business publications etc. It |
|
includes correspondence between companies, organizations and the public. |
|
|
|
### Who are the source language producers? |
|
|
|
Given the various types of archival material used in annotation, the scope of producers of the original texts is broad. It includes private individuals and employees of |
|
different companies. |
|
|
|
### Annotations |
|
|
|
The textlines were first cropped out of the original image and then transcribed. If the transcription was unclear, the annotator marked it as either "somewhat unclear" or |
|
"unclear". Unclear images were discarded, but the "somewhat discarded" images are presented here as in the "semi" annotation files. The rough estimate for "somewhat |
|
unclear" class is that less than 100% and more than 50% of the characters are unclear. |
|
|
|
### Who are the annotators? |
|
|
|
Annotators were employees of National Archives of Finland and ELKA. |
|
|
|
### Synthetic data |
|
|
|
As a way to increase the amount of training data, we created synthetic data by using this library https://github.com/Belval/TextRecognitionDataGenerator. We collected Finnish books from https://www.gutenberg.org/ and Finnish magazines from https://archive.org/ and created different kinds of textlines. The different kinds include normal textlines, rotated textlines, textlines following a sinosoidal curve and textlines where characters are subjected to noise. |
|
|
|
### Personal and Sensitive Information |
|
|
|
The dataset is not anonymized, so individuals' names can be found in the dataset. |
