Files changed (1) hide show
  1. README.md +82 -2
README.md CHANGED
@@ -1,4 +1,22 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  dataset_info:
3
  features:
4
  - name: dataset
@@ -15,13 +33,75 @@ dataset_info:
15
  dtype: string
16
  splits:
17
  - name: test
18
- num_bytes: 178249701.0
19
  num_examples: 123
20
  download_size: 176996271
21
- dataset_size: 178249701.0
22
  configs:
23
  - config_name: default
24
  data_files:
25
  - split: test
26
  path: data/test-*
27
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ task_categories:
3
+ - image-to-text
4
+ pretty_name: ATR benchmark
5
+ size_categories:
6
+ - n<1K
7
+ language:
8
+ - fr
9
+ - la
10
+ - en
11
+ - no
12
+ - ar
13
+ - zh
14
+ - de
15
+ - nl
16
+ tags:
17
+ - atr
18
+ - htr
19
+ - ocr
20
  dataset_info:
21
  features:
22
  - name: dataset
 
33
  dtype: string
34
  splits:
35
  - name: test
36
+ num_bytes: 178249701
37
  num_examples: 123
38
  download_size: 176996271
39
+ dataset_size: 178249701
40
  configs:
41
  - config_name: default
42
  data_files:
43
  - split: test
44
  path: data/test-*
45
  ---
46
+
47
+ # ATR benchmark - Page/paragraph level
48
+
49
+ ## Dataset Description
50
+
51
+ - **Homepage:** [ATR benchmark](https://huggingface.co/datasets/Teklia/ATR-benchmark)
52
+ - **Point of Contact:** [TEKLIA](https://teklia.com)
53
+
54
+ ## Dataset Summary
55
+
56
+ The ATR benchmark dataset is a multilingual dataset that includes 83 document images, at page or paragraph level. This dataset has been designed to test ATR models and combines data from several public datasets:
57
+ - [BnL Historical Newspapers](https://data.bnl.lu/data/historical-newspapers/)
58
+ - [CASIA-HWDB2](https://nlpr.ia.ac.cn/databases/handwriting/Offline_database.html)
59
+ - [DIY History - Social Justice](http://diyhistory.lib.uiowa.edu/)
60
+ - FINLAM - Historical Newspapers
61
+ - [Horae - Books of hours](https://github.com/oriflamms/HORAE)
62
+ - [IAM](https://fki.tic.heia-fr.ch/databases/iam-handwriting-database)
63
+ - [NorHand v3](https://zenodo.org/records/10255840)
64
+ - [Marius PELLET](https://europeana.transcribathon.eu/documents/story/?story=121795)
65
+ - [RASM](http://www.primaresearch.org/RASM2019/)
66
+ - [READ-2016](https://zenodo.org/records/218236)
67
+ - [RIMES](https://teklia.com/research/rimes-database/)
68
+ - [ScribbleLens](https://openslr.org/84/)
69
+
70
+ Images are in their original size.
71
+
72
+ ### Split
73
+
74
+ | dataset | images | language |
75
+ | ------------------------------ | ------:| ---------------- |
76
+ | BnL Historical Newspapers | 3 | German |
77
+ | CASIA-HWDB2 | 10 | Chinese |
78
+ | DIY History - Social Justice | 20 | English |
79
+ | FINLAM - Historical Newspapers | 10 | English / French |
80
+ | Horae - Books of hours | 10 | Latin |
81
+ | IAM | 10 | English |
82
+ | NorHand v3 | 10 | Norwegian |
83
+ | Marius PELLET | 10 | French |
84
+ | RASM | 10 | Arabic |
85
+ | READ-2016 | 10 | German |
86
+ | RIMES | 10 | French |
87
+ | ScribbleLens | 10 | Dutch |
88
+
89
+ ## Dataset Structure
90
+
91
+ ### Data Instances
92
+
93
+ ```
94
+ {
95
+ 'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size={img size} at 0x1A800E8E190,
96
+ 'text': '{transcription}'
97
+ }
98
+ ```
99
+
100
+ ### Data Fields
101
+
102
+ - `image`: a PIL.Image.Image object containing the image. Note that when accessing the image column (using dataset[0]["image"]), the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the "image" column, i.e. dataset[0]["image"] should always be preferred over dataset["image"][0].
103
+ - `dataset`: the name of the original dataset.
104
+ - `arkindex_id`: Arkindex element id corresponding to the current page or paragraph.
105
+ - `text`: the label transcription of the image.
106
+ - `language`: language of text on the image.
107
+ - `level`: full document page or a single paragraph.