Datasets:
saattrupdan
commited on
Commit
·
6671176
1
Parent(s):
80464a4
docs: Update readme
Browse files
README.md
CHANGED
@@ -1,60 +1,146 @@
|
|
1 |
---
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
- name: original_width
|
16 |
-
dtype: int32
|
17 |
-
- name: mime_type
|
18 |
-
dtype: string
|
19 |
-
- name: caption_attribution_description
|
20 |
-
dtype: string
|
21 |
-
- name: page_url
|
22 |
-
dtype: string
|
23 |
-
- name: attribution_passes_lang_id
|
24 |
-
dtype: bool
|
25 |
-
- name: caption_alt_text_description
|
26 |
-
dtype: string
|
27 |
-
- name: caption_reference_description
|
28 |
-
dtype: string
|
29 |
-
- name: caption_title_and_reference_description
|
30 |
-
dtype: string
|
31 |
-
- name: context_page_description
|
32 |
-
dtype: string
|
33 |
-
- name: context_section_description
|
34 |
-
dtype: string
|
35 |
-
- name: hierarchical_section_title
|
36 |
-
dtype: string
|
37 |
-
- name: is_main_image
|
38 |
-
dtype: bool
|
39 |
-
- name: page_changed_recently
|
40 |
-
dtype: bool
|
41 |
-
- name: page_title
|
42 |
-
dtype: string
|
43 |
-
- name: section_title
|
44 |
-
dtype: string
|
45 |
-
splits:
|
46 |
-
- name: test
|
47 |
-
num_bytes: 50882438.848311014
|
48 |
-
num_examples: 1024
|
49 |
-
- name: train
|
50 |
-
num_bytes: 8321067587.439611
|
51 |
-
num_examples: 167460
|
52 |
-
- name: val
|
53 |
-
num_bytes: 12720609.712077754
|
54 |
-
num_examples: 256
|
55 |
-
download_size: 8000497021
|
56 |
-
dataset_size: 8384670636.000001
|
57 |
---
|
58 |
-
# Dataset Card for "danish-wit"
|
59 |
|
60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
pretty_name: DanishWIT
|
3 |
+
language:
|
4 |
+
- da
|
5 |
+
license:
|
6 |
+
- cc-by-sa-4.0
|
7 |
+
size_categories:
|
8 |
+
- 100K<n<1M
|
9 |
+
source_datasets:
|
10 |
+
- wikimedia/wit_base
|
11 |
+
task_categories:
|
12 |
+
- image-to-text
|
13 |
+
task_ids:
|
14 |
+
- image-captioning
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
---
|
|
|
16 |
|
17 |
+
# Dataset Card for Danish WIT
|
18 |
+
|
19 |
+
### Dataset Summary
|
20 |
+
|
21 |
+
Google presented the Wikipedia Image Text (WIT) dataset in [July
|
22 |
+
2021](https://dl.acm.org/doi/abs/10.1145/3404835.3463257), a dataset which contains
|
23 |
+
scraped images from Wikipedia along with their descriptions. WikiMedia released
|
24 |
+
WIT-Base in [September
|
25 |
+
2021](https://techblog.wikimedia.org/2021/09/09/the-wikipedia-image-caption-matching-challenge-and-a-huge-release-of-image-data-for-research/),
|
26 |
+
being a modified version of WIT where they have removed the images with empty
|
27 |
+
"reference descriptions", as well as removing images where a person's face covers more
|
28 |
+
than 10% of the image surface, along with inappropriate images that are candidate for
|
29 |
+
deletion. This dataset is the Danish portion of the WIT-Base dataset, consisting of
|
30 |
+
roughly 160,000 images with associated Danish descriptions. We release the dataset
|
31 |
+
under the CC BY-SA 4.0 license, in accordance with WIT-Base's identical license.
|
32 |
+
|
33 |
+
|
34 |
+
### Supported Tasks and Leaderboards
|
35 |
+
|
36 |
+
Training machine learning models for caption generation, zero-shot image classification
|
37 |
+
and text-image search are the intended tasks for this dataset. No leaderboard is active
|
38 |
+
at this point.
|
39 |
+
|
40 |
+
|
41 |
+
### Languages
|
42 |
+
|
43 |
+
The dataset is available in Danish (`da`).
|
44 |
+
|
45 |
+
|
46 |
+
## Dataset Structure
|
47 |
+
|
48 |
+
### Data Instances
|
49 |
+
|
50 |
+
- **Size of downloaded dataset files:** 7.5 GB
|
51 |
+
- **Size of the generated dataset:** 7.8 GB
|
52 |
+
- **Total amount of disk used:** 15.3 GB
|
53 |
+
|
54 |
+
An example from the `train` split looks as follows.
|
55 |
+
```
|
56 |
+
{
|
57 |
+
"image": {
|
58 |
+
"bytes": b"\xff\xd8\xff\xe0\x00\x10JFIF...",
|
59 |
+
"path": None
|
60 |
+
},
|
61 |
+
"image_url": "https://upload.wikimedia.org/wikipedia/commons/4/45/Bispen_-_inside.jpg",
|
62 |
+
"embedding": [2.8568285, 2.9562542, 0.33794892, 8.753725, ...],
|
63 |
+
"metadata_url": "http://commons.wikimedia.org/wiki/File:Bispen_-_inside.jpg",
|
64 |
+
"original_height": 3161,
|
65 |
+
"original_width": 2316,
|
66 |
+
"mime_type": "image/jpeg",
|
67 |
+
"caption_attribution_description": "Kulturhuset Bispen set indefra. Biblioteket er til venstre",
|
68 |
+
"page_url": "https://da.wikipedia.org/wiki/Bispen",
|
69 |
+
"attribution_passes_lang_id": True,
|
70 |
+
"caption_alt_text_description": None,
|
71 |
+
"caption_reference_description": "Bispen set indefra fra 1. sal, hvor ....",
|
72 |
+
"caption_title_and_reference_description": "Bispen [SEP] Bispen set indefra ...",
|
73 |
+
"context_page_description": "Bispen er navnet på det offentlige kulturhus i ...",
|
74 |
+
"context_section_description": "Bispen er navnet på det offentlige kulturhus i ...",
|
75 |
+
"hierarchical_section_title": "Bispen",
|
76 |
+
"is_main_image": True,
|
77 |
+
"page_changed_recently": True,
|
78 |
+
"page_title": "Bispen",
|
79 |
+
"section_title": None
|
80 |
+
}
|
81 |
+
```
|
82 |
+
|
83 |
+
### Data Fields
|
84 |
+
|
85 |
+
The data fields are the same among all splits.
|
86 |
+
|
87 |
+
- `image`: a `dict` feature.
|
88 |
+
- `image_url`: a `str` feature.
|
89 |
+
- `embedding`: a `list` feature.
|
90 |
+
- `metadata_url`: a `str` feature.
|
91 |
+
- `original_height`: an `int` or `NaN` feature.
|
92 |
+
- `original_width`: an `int` or `NaN` feature.
|
93 |
+
- `mime_type`: a `str` or `None` feature.
|
94 |
+
- `caption_attribution_description`: a `str` or `None` feature.
|
95 |
+
- `page_url`: a `str` feature.
|
96 |
+
- `attribution_passes_lang_id`: a `bool` or `None` feature.
|
97 |
+
- `caption_alt_text_description`: a `str` or `None` feature.
|
98 |
+
- `caption_reference_description`: a `str` or `None` feature.
|
99 |
+
- `caption_title_and_reference_description`: a `str` or `None` feature.
|
100 |
+
- `context_page_description`: a `str` or `None` feature.
|
101 |
+
- `context_section_description`: a `str` or `None` feature.
|
102 |
+
- `hierarchical_section_title`: a `str` feature.
|
103 |
+
- `is_main_image`: a `bool` or `None` feature.
|
104 |
+
- `page_changed_recently`: a `bool` or `None` feature.
|
105 |
+
- `page_title`: a `str` feature.
|
106 |
+
- `section_title`: a `str` or `None` feature.
|
107 |
+
|
108 |
+
### Data Splits
|
109 |
+
|
110 |
+
Roughly 2.60% of the WIT-Base dataset comes from the Danish Wikipedia. We have split
|
111 |
+
the resulting 168,740 samples into a training set, validation set and testing set of
|
112 |
+
the following sizes:
|
113 |
+
|
114 |
+
| split | samples |
|
115 |
+
|---------|--------:|
|
116 |
+
| train | 167,460 |
|
117 |
+
| val | 256 |
|
118 |
+
| test | 1,024 |
|
119 |
+
|
120 |
+
|
121 |
+
## Dataset Creation
|
122 |
+
|
123 |
+
### Curation Rationale
|
124 |
+
|
125 |
+
It is quite cumbersome to extract the Danish portion of the WIT-Base dataset,
|
126 |
+
especially as the dataset takes up 333 GB of disk space, so the curation of Danish-WIT
|
127 |
+
is purely to make it easier to work with the Danish portion of it.
|
128 |
+
|
129 |
+
### Source Data
|
130 |
+
|
131 |
+
The original data was collected from WikiMedia's
|
132 |
+
[WIT-Base](https://huggingface.co/datasets/wikimedia/wit_base) dataset, which in turn
|
133 |
+
comes from Google's [WIT](https://huggingface.co/datasets/google/wit) dataset.
|
134 |
+
|
135 |
+
|
136 |
+
## Additional Information
|
137 |
+
|
138 |
+
### Dataset Curators
|
139 |
+
|
140 |
+
[Dan Saattrup Nielsen](https://saattrupdan.github.io/) from the [The Alexandra
|
141 |
+
Institute](https://alexandra.dk/) curated this dataset.
|
142 |
+
|
143 |
+
### Licensing Information
|
144 |
+
|
145 |
+
The dataset is licensed under the [CC BY-SA 4.0
|
146 |
+
license](https://creativecommons.org/licenses/by-sa/4.0/).
|