Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -1,73 +1,153 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
sequence: int64
|
19 |
-
- name: pdf_cells
|
20 |
-
list:
|
21 |
-
list:
|
22 |
-
- name: bbox
|
23 |
-
sequence: float64
|
24 |
-
- name: font
|
25 |
-
struct:
|
26 |
-
- name: color
|
27 |
-
sequence: int64
|
28 |
-
- name: name
|
29 |
-
dtype: string
|
30 |
-
- name: size
|
31 |
-
dtype: float64
|
32 |
-
- name: text
|
33 |
-
dtype: string
|
34 |
-
- name: metadata
|
35 |
-
struct:
|
36 |
-
- name: coco_height
|
37 |
-
dtype: int64
|
38 |
-
- name: coco_width
|
39 |
-
dtype: int64
|
40 |
-
- name: collection
|
41 |
-
dtype: string
|
42 |
-
- name: doc_category
|
43 |
-
dtype: string
|
44 |
-
- name: image_id
|
45 |
-
dtype: int64
|
46 |
-
- name: num_pages
|
47 |
-
dtype: int64
|
48 |
-
- name: original_filename
|
49 |
-
dtype: string
|
50 |
-
- name: original_height
|
51 |
-
dtype: float64
|
52 |
-
- name: original_width
|
53 |
-
dtype: float64
|
54 |
-
- name: page_hash
|
55 |
-
dtype: string
|
56 |
-
- name: page_no
|
57 |
-
dtype: int64
|
58 |
-
splits:
|
59 |
-
- name: test
|
60 |
-
num_bytes: 1996722627.125
|
61 |
-
num_examples: 4999
|
62 |
-
- name: val
|
63 |
-
num_bytes: 2494706174.875
|
64 |
-
num_examples: 6489
|
65 |
-
- name: train
|
66 |
-
num_bytes: 28179797029.125
|
67 |
-
num_examples: 69375
|
68 |
-
download_size: 31783096529
|
69 |
-
dataset_size: 32671225831.125
|
70 |
---
|
71 |
-
# Dataset Card for "DocLayNet-v1.1"
|
72 |
|
73 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
annotations_creators:
|
3 |
+
- crowdsourced
|
4 |
+
license: other
|
5 |
+
pretty_name: DocLayNet
|
6 |
+
size_categories:
|
7 |
+
- 10K<n<100K
|
8 |
+
tags:
|
9 |
+
- layout-segmentation
|
10 |
+
- COCO
|
11 |
+
- document-understanding
|
12 |
+
- PDF
|
13 |
+
task_categories:
|
14 |
+
- object-detection
|
15 |
+
- image-segmentation
|
16 |
+
task_ids:
|
17 |
+
- instance-segmentation
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
---
|
|
|
19 |
|
20 |
+
# Dataset Card for DocLayNet v1.1
|
21 |
+
|
22 |
+
## Table of Contents
|
23 |
+
- [Table of Contents](#table-of-contents)
|
24 |
+
- [Dataset Description](#dataset-description)
|
25 |
+
- [Dataset Summary](#dataset-summary)
|
26 |
+
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
27 |
+
- [Dataset Structure](#dataset-structure)
|
28 |
+
- [Data Fields](#data-fields)
|
29 |
+
- [Data Splits](#data-splits)
|
30 |
+
- [Dataset Creation](#dataset-creation)
|
31 |
+
- [Annotations](#annotations)
|
32 |
+
- [Additional Information](#additional-information)
|
33 |
+
- [Dataset Curators](#dataset-curators)
|
34 |
+
- [Licensing Information](#licensing-information)
|
35 |
+
- [Citation Information](#citation-information)
|
36 |
+
- [Contributions](#contributions)
|
37 |
+
|
38 |
+
## Dataset Description
|
39 |
+
|
40 |
+
- **Homepage:** https://developer.ibm.com/exchanges/data/all/doclaynet/
|
41 |
+
- **Repository:** https://github.com/DS4SD/DocLayNet
|
42 |
+
- **Paper:** https://doi.org/10.1145/3534678.3539043
|
43 |
+
|
44 |
+
### Dataset Summary
|
45 |
+
|
46 |
+
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:
|
47 |
+
|
48 |
+
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
|
49 |
+
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
|
50 |
+
3. *Detailed label set*: DocLayNet defines 11 class labels to distinguish layout features in high detail.
|
51 |
+
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
|
52 |
+
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.
|
53 |
+
|
54 |
+
|
55 |
+
## Dataset Structure
|
56 |
+
|
57 |
+
This dataset is structured differently from the other repository [ds4sd/DocLayNet](https://huggingface.co/datasets/ds4sd/DocLayNet), as this one includes the content (PDF cells) of the detections, and abandons the COCO format.
|
58 |
+
|
59 |
+
* `image`: page PIL image.
|
60 |
+
* `bboxes`: a list of layout bounding boxes.
|
61 |
+
* `category_id`: a list of class ids corresponding to the bounding boxes.
|
62 |
+
* `segmentation`: a list of layout segmentation polygons.
|
63 |
+
* `pdf_cells`: a list of lists corresponding to `bbox`. Each list contains the PDF cells (content) inside the bbox.
|
64 |
+
* `metadata`: page and document metadetails.
|
65 |
+
|
66 |
+
Bounding boxes classes / categories:
|
67 |
+
|
68 |
+
```
|
69 |
+
1: Caption
|
70 |
+
2: Footnote
|
71 |
+
3: Formula
|
72 |
+
4: List-item
|
73 |
+
5: Page-footer
|
74 |
+
6: Page-header
|
75 |
+
7: Picture
|
76 |
+
8: Section-header
|
77 |
+
9: Table
|
78 |
+
10: Text
|
79 |
+
11: Title
|
80 |
+
```
|
81 |
+
|
82 |
+
|
83 |
+
The `["metadata"]["doc_category"]` field uses one of the following constants:
|
84 |
+
|
85 |
+
```
|
86 |
+
* financial_reports,
|
87 |
+
* scientific_articles,
|
88 |
+
* laws_and_regulations,
|
89 |
+
* government_tenders,
|
90 |
+
* manuals,
|
91 |
+
* patents
|
92 |
+
```
|
93 |
+
|
94 |
+
|
95 |
+
### Data Splits
|
96 |
+
|
97 |
+
The dataset provides three splits
|
98 |
+
- `train`
|
99 |
+
- `val`
|
100 |
+
- `test`
|
101 |
+
|
102 |
+
## Dataset Creation
|
103 |
+
|
104 |
+
### Annotations
|
105 |
+
|
106 |
+
#### Annotation process
|
107 |
+
|
108 |
+
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).
|
109 |
+
|
110 |
+
|
111 |
+
#### Who are the annotators?
|
112 |
+
|
113 |
+
Annotations are crowdsourced.
|
114 |
+
|
115 |
+
|
116 |
+
## Additional Information
|
117 |
+
|
118 |
+
### Dataset Curators
|
119 |
+
|
120 |
+
The dataset is curated by the [Deep Search team](https://ds4sd.github.io/) at IBM Research.
|
121 |
+
You can contact us at [deepsearch-core@zurich.ibm.com](mailto:deepsearch-core@zurich.ibm.com).
|
122 |
+
|
123 |
+
Curators:
|
124 |
+
- Christoph Auer, [@cau-git](https://github.com/cau-git)
|
125 |
+
- Michele Dolfi, [@dolfim-ibm](https://github.com/dolfim-ibm)
|
126 |
+
- Ahmed Nassar, [@nassarofficial](https://github.com/nassarofficial)
|
127 |
+
- Peter Staar, [@PeterStaar-IBM](https://github.com/PeterStaar-IBM)
|
128 |
+
|
129 |
+
### Licensing Information
|
130 |
+
|
131 |
+
License: [CDLA-Permissive-1.0](https://cdla.io/permissive-1-0/)
|
132 |
+
|
133 |
+
|
134 |
+
### Citation Information
|
135 |
+
|
136 |
+
|
137 |
+
```bib
|
138 |
+
@article{doclaynet2022,
|
139 |
+
title = {DocLayNet: A Large Human-Annotated Dataset for Document-Layout Segmentation},
|
140 |
+
doi = {10.1145/3534678.353904},
|
141 |
+
url = {https://doi.org/10.1145/3534678.3539043},
|
142 |
+
author = {Pfitzmann, Birgit and Auer, Christoph and Dolfi, Michele and Nassar, Ahmed S and Staar, Peter W J},
|
143 |
+
year = {2022},
|
144 |
+
isbn = {9781450393850},
|
145 |
+
publisher = {Association for Computing Machinery},
|
146 |
+
address = {New York, NY, USA},
|
147 |
+
booktitle = {Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
|
148 |
+
pages = {3743–3751},
|
149 |
+
numpages = {9},
|
150 |
+
location = {Washington DC, USA},
|
151 |
+
series = {KDD '22}
|
152 |
+
}
|
153 |
+
```
|