agomberto commited on
Commit
d724c59
1 Parent(s): 43fa597

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +193 -68
README.md CHANGED
@@ -1,70 +1,195 @@
1
  ---
2
- dataset_info:
3
- features:
4
- - name: id
5
- dtype: string
6
- - name: texts
7
- sequence: string
8
- - name: bboxes_block
9
- sequence:
10
- sequence: int64
11
- - name: bboxes_line
12
- sequence:
13
- sequence: int64
14
- - name: categories
15
- sequence:
16
- class_label:
17
- names:
18
- '0': Caption
19
- '1': Footnote
20
- '2': Formula
21
- '3': List-item
22
- '4': Page-footer
23
- '5': Page-header
24
- '6': Picture
25
- '7': Section-header
26
- '8': Table
27
- '9': Text
28
- '10': Title
29
- - name: page_hash
30
- dtype: string
31
- - name: original_filename
32
- dtype: string
33
- - name: page_no
34
- dtype: int32
35
- - name: num_pages
36
- dtype: int32
37
- - name: original_width
38
- dtype: int32
39
- - name: original_height
40
- dtype: int32
41
- - name: coco_width
42
- dtype: int32
43
- - name: coco_height
44
- dtype: int32
45
- - name: collection
46
- dtype: string
47
- - name: doc_category
48
- dtype: string
49
- splits:
50
- - name: train
51
- num_bytes: 773564178
52
- num_examples: 69103
53
- - name: validation
54
- num_bytes: 74492844
55
- num_examples: 6480
56
- - name: test
57
- num_bytes: 58899926
58
- num_examples: 4994
59
- download_size: 196812287
60
- dataset_size: 906956948
61
- configs:
62
- - config_name: default
63
- data_files:
64
- - split: train
65
- path: data/train-*
66
- - split: validation
67
- path: data/validation-*
68
- - split: test
69
- path: data/test-*
70
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - en
4
+ - de
5
+ - fr
6
+ - ja
7
+ annotations_creators:
8
+ - crowdsourced
9
+ license: other
10
+ pretty_name: DocLayNet large
11
+ size_categories:
12
+ - 10K<n<100K
13
+ tags:
14
+ - DocLayNet
15
+ - COCO
16
+ - PDF
17
+ - IBM
18
+ - Financial-Reports
19
+ - Finance
20
+ - Manuals
21
+ - Scientific-Articles
22
+ - Science
23
+ - Laws
24
+ - Law
25
+ - Regulations
26
+ - Patents
27
+ - Government-Tenders
28
+ - object-detection
29
+ - image-segmentation
30
+ - token-classification
31
+ task_categories:
32
+ - object-detection
33
+ - image-segmentation
34
+ - token-classification
35
+ task_ids:
36
+ - instance-segmentation
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37
  ---
38
+
39
+ # Dataset Card for DocLayNet large without image
40
+
41
+ ## About this card (02/14/2024)
42
+
43
+ ### Property and license
44
+
45
+ All information from this page but the content of this paragraph "About this card (02/14/2025)" has been copied/pasted from [Dataset Card for DocLayNet](https://huggingface.co/datasets/ds4sd/DocLayNet).
46
+
47
+ DocLayNet is a dataset created by Deep Search (IBM Research) published under [license CDLA-Permissive-1.0](https://huggingface.co/datasets/ds4sd/DocLayNet#licensing-information).
48
+
49
+ I do not claim any rights to the data taken from this dataset and published on this page.
50
+
51
+ # Dataset Card for DocLayNet
52
+
53
+ ## Table of Contents
54
+ - [Table of Contents](#table-of-contents)
55
+ - [Dataset Description](#dataset-description)
56
+ - [Dataset Summary](#dataset-summary)
57
+ - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
58
+ - [Dataset Structure](#dataset-structure)
59
+ - [Data Fields](#data-fields)
60
+ - [Data Splits](#data-splits)
61
+ - [Dataset Creation](#dataset-creation)
62
+ - [Annotations](#annotations)
63
+ - [Additional Information](#additional-information)
64
+ - [Dataset Curators](#dataset-curators)
65
+ - [Licensing Information](#licensing-information)
66
+ - [Citation Information](#citation-information)
67
+ - [Contributions](#contributions)
68
+
69
+ ## Dataset Description
70
+
71
+ - **Homepage:** https://developer.ibm.com/exchanges/data/all/doclaynet/
72
+ - **Repository:** https://github.com/DS4SD/DocLayNet
73
+ - **Paper:** https://doi.org/10.1145/3534678.3539043
74
+ - **Leaderboard:**
75
+ - **Point of Contact:**
76
+
77
+ ### Dataset Summary
78
+
79
+ DocLayNet provides page-by-page layout segmentation ground-truth using bounding-boxes for 11 distinct class labels on 80863 unique pages from 6 document categories. It provides several unique features compared to related work such as PubLayNet or DocBank:
80
+
81
+ 1. *Human Annotation*: DocLayNet is hand-annotated by well-trained experts, providing a gold-standard in layout segmentation through human recognition and interpretation of each page layout
82
+ 2. *Large layout variability*: DocLayNet includes diverse and complex layouts from a large variety of public sources in Finance, Science, Patents, Tenders, Law texts and Manuals
83
+ 3. *Detailed label set*: DocLayNet defines 11 class labels to distinguish layout features in high detail.
84
+ 4. *Redundant annotations*: A fraction of the pages in DocLayNet are double- or triple-annotated, allowing to estimate annotation uncertainty and an upper-bound of achievable prediction accuracy with ML models
85
+ 5. *Pre-defined train- test- and validation-sets*: DocLayNet provides fixed sets for each to ensure proportional representation of the class-labels and avoid leakage of unique layout styles across the sets.
86
+
87
+ ### Supported Tasks and Leaderboards
88
+
89
+ We are hosting a competition in ICDAR 2023 based on the DocLayNet dataset. For more information see https://ds4sd.github.io/icdar23-doclaynet/.
90
+
91
+ ## Dataset Structure
92
+
93
+ ### Data Fields
94
+
95
+ DocLayNet provides four types of data assets:
96
+
97
+ 1. Bounding-box annotations in COCO format for each PNG image
98
+ 2. Extra: Single-page PDF files matching each PNG image
99
+ 3. Extra: JSON file matching each PDF page, which provides the digital text cells with coordinates and content
100
+
101
+ The COCO image record are defined like this example
102
+
103
+ ```js
104
+ ...
105
+ {
106
+ "id": 1,
107
+ "width": 1025,
108
+ "height": 1025,
109
+ "file_name": "132a855ee8b23533d8ae69af0049c038171a06ddfcac892c3c6d7e6b4091c642.png",
110
+
111
+ // Custom fields:
112
+ "doc_category": "financial_reports" // high-level document category
113
+ "collection": "ann_reports_00_04_fancy", // sub-collection name
114
+ "doc_name": "NASDAQ_FFIN_2002.pdf", // original document filename
115
+ "page_no": 9, // page number in original document
116
+ "precedence": 0, // Annotation order, non-zero in case of redundant double- or triple-annotation
117
+ },
118
+ ...
119
+ ```
120
+
121
+ The `doc_category` field uses one of the following constants:
122
+
123
+ ```
124
+ financial_reports,
125
+ scientific_articles,
126
+ laws_and_regulations,
127
+ government_tenders,
128
+ manuals,
129
+ patents
130
+ ```
131
+
132
+
133
+ ### Data Splits
134
+
135
+ The dataset provides three splits
136
+ - `train`
137
+ - `val`
138
+ - `test`
139
+
140
+ ## Dataset Creation
141
+
142
+ ### Annotations
143
+
144
+ #### Annotation process
145
+
146
+ The labeling guideline used for training of the annotation experts are available at [DocLayNet_Labeling_Guide_Public.pdf](https://raw.githubusercontent.com/DS4SD/DocLayNet/main/assets/DocLayNet_Labeling_Guide_Public.pdf).
147
+
148
+
149
+ #### Who are the annotators?
150
+
151
+ Annotations are crowdsourced.
152
+
153
+
154
+ ## Additional Information
155
+
156
+ ### Dataset Curators
157
+
158
+ The dataset is curated by the [Deep Search team](https://ds4sd.github.io/) at IBM Research.
159
+ You can contact us at [deepsearch-core@zurich.ibm.com](mailto:deepsearch-core@zurich.ibm.com).
160
+
161
+ Curators:
162
+ - Christoph Auer, [@cau-git](https://github.com/cau-git)
163
+ - Michele Dolfi, [@dolfim-ibm](https://github.com/dolfim-ibm)
164
+ - Ahmed Nassar, [@nassarofficial](https://github.com/nassarofficial)
165
+ - Peter Staar, [@PeterStaar-IBM](https://github.com/PeterStaar-IBM)
166
+
167
+ ### Licensing Information
168
+
169
+ License: [CDLA-Permissive-1.0](https://cdla.io/permissive-1-0/)
170
+
171
+
172
+ ### Citation Information
173
+
174
+
175
+ ```bib
176
+ @article{doclaynet2022,
177
+ title = {DocLayNet: A Large Human-Annotated Dataset for Document-Layout Segmentation},
178
+ doi = {10.1145/3534678.353904},
179
+ url = {https://doi.org/10.1145/3534678.3539043},
180
+ author = {Pfitzmann, Birgit and Auer, Christoph and Dolfi, Michele and Nassar, Ahmed S and Staar, Peter W J},
181
+ year = {2022},
182
+ isbn = {9781450393850},
183
+ publisher = {Association for Computing Machinery},
184
+ address = {New York, NY, USA},
185
+ booktitle = {Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
186
+ pages = {3743–3751},
187
+ numpages = {9},
188
+ location = {Washington DC, USA},
189
+ series = {KDD '22}
190
+ }
191
+ ```
192
+
193
+ ### Contributions
194
+
195
+ Thanks to [@dolfim-ibm](https://github.com/dolfim-ibm), [@cau-git](https://github.com/cau-git) for adding this dataset.