Datasets:

Modalities:
Image
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:
HugoLaurencon commited on
Commit
0af5457
·
verified ·
1 Parent(s): 0c566a9

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +61 -22
README.md CHANGED
@@ -3,21 +3,9 @@ language:
3
  - en
4
  license: cc-by-4.0
5
  size_categories:
6
- - 100K<n<1M
7
  pretty_name: WebSight
8
  dataset_info:
9
- - config_name: v0.1
10
- features:
11
- - name: image
12
- dtype: image
13
- - name: text
14
- dtype: string
15
- splits:
16
- - name: train
17
- num_bytes: 35386660486.65
18
- num_examples: 822987
19
- download_size: 31394170440
20
- dataset_size: 35386660486.65
21
  - config_name: v0.2
22
  features:
23
  - name: image
@@ -32,30 +20,54 @@ dataset_info:
32
  num_examples: 1922671
33
  download_size: 144861710051
34
  dataset_size: 368943620718.125
35
- configs:
36
  - config_name: v0.1
37
- data_files:
38
- - split: train
39
- path: data/train-*
 
 
 
 
 
 
 
 
 
40
  - config_name: v0.2
41
  data_files:
42
  - split: train
43
  path: v0.2/train-*
 
 
 
 
44
  tags:
45
  - code
46
  - synthetic
47
  ---
48
  # Dataset Card for WebSight
49
 
50
- ## Dataset Version
51
-
52
- Alpha version, v0.1.
53
 
54
  ## Dataset Description
55
 
56
- This dataset consists of 823,000 HTML/CSS codes representing synthetically generated English websites, each accompanied by a corresponding screenshot (rendered with [Playwright](https://github.com/microsoft/playwright)).
 
 
57
 
58
- This dataset serves as a valuable resource for tasks such as generating UI codes from a screenshot. Notably, it is used for the training of the forthcoming Idefics-2 model, aiming at augmenting its proficiency in generating website code based on a provided screenshot.
 
 
 
 
 
 
 
 
 
 
 
 
 
59
 
60
  ## Data Fields
61
 
@@ -94,6 +106,33 @@ When you want to display an image, don't use <img> in the HTML, always display a
94
  ```
95
  Following these steps, the HTML and CSS codes were extracted from the outputs of Deepseek-Coder and formatted into the structure `'<html>\n<style>\n{css}</style>\n{body}\n</html>'`.
96
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
97
  ## Terms of Use
98
 
99
  By using the dataset, you agree to comply with the original licenses of the source content as well as the dataset license (CC-BY-4.0). Additionally, if you use this dataset to train a Machine Learning model, you agree to disclose your use of the dataset when releasing the model or an ML application using the model.
 
3
  - en
4
  license: cc-by-4.0
5
  size_categories:
6
+ - 1M<n<10M
7
  pretty_name: WebSight
8
  dataset_info:
 
 
 
 
 
 
 
 
 
 
 
 
9
  - config_name: v0.2
10
  features:
11
  - name: image
 
20
  num_examples: 1922671
21
  download_size: 144861710051
22
  dataset_size: 368943620718.125
 
23
  - config_name: v0.1
24
+ features:
25
+ - name: image
26
+ dtype: image
27
+ - name: text
28
+ dtype: string
29
+ splits:
30
+ - name: train
31
+ num_bytes: 35386660486.65
32
+ num_examples: 822987
33
+ download_size: 31394170440
34
+ dataset_size: 35386660486.65
35
+ configs:
36
  - config_name: v0.2
37
  data_files:
38
  - split: train
39
  path: v0.2/train-*
40
+ - config_name: v0.1
41
+ data_files:
42
+ - split: train
43
+ path: data/train-*
44
  tags:
45
  - code
46
  - synthetic
47
  ---
48
  # Dataset Card for WebSight
49
 
 
 
 
50
 
51
  ## Dataset Description
52
 
53
+ WebSight is a large synthetic dataset containing HTML/CSS codes representing synthetically generated English websites, each accompanied by a corresponding screenshot.
54
+
55
+ This dataset serves as a valuable resource for tasks such as generating UI codes from a screenshot. It is for example used to train the model [Sightseer](https://huggingface.co/HuggingFaceM4/Sightseer).
56
 
57
+ It comes in two versions:
58
+ - v0.1: Websites are coded with HTML + CSS. They do not include real images.
59
+ - v0.2: Websites are coded with HTML + Tailwind CSS. They do include real images.
60
+
61
+ Essentially, here are the improvements in version v0.2, compared to version v0.1:
62
+ - Websites include real images (related to the context of the website!!)
63
+ - Usage of Tailwind CSS instead of traditional CSS
64
+ - Contains 2x more examples
65
+ - Contains more tables
66
+ - Better resolution for the screenshots
67
+ - Presence of a column indicating the LLM-generated idea to create the websites
68
+
69
+ <details>
70
+ <summary>Details for WebSight-v0.1 (HTML + CSS)</summary>
71
 
72
  ## Data Fields
73
 
 
106
  ```
107
  Following these steps, the HTML and CSS codes were extracted from the outputs of Deepseek-Coder and formatted into the structure `'<html>\n<style>\n{css}</style>\n{body}\n</html>'`.
108
 
109
+ </details>
110
+
111
+ <details>
112
+ <summary>Details for WebSight-v0.2 (HTML + Tailwind CSS)</summary>
113
+
114
+ ## Data Fields
115
+
116
+ An example of a sample appears as follows:
117
+ ```
118
+ {
119
+ 'images': PIL.Image,
120
+ 'text': '<html>\n<link href="https://cdn.jsdelivr.net/npm/tailwindcss@2.2.19/dist/tailwind.min.css" rel="stylesheet">\n{body}\n</html>',
121
+ }
122
+ ```
123
+ where `body` is the body of the HTML code, containing directly the Tailwind CSS code and facilitating the straightforward training of a model.
124
+
125
+ ## Data Splits
126
+
127
+ There is only one split, `train`, that contains TO DO images and codes.
128
+
129
+ ## Dataset Creation
130
+
131
+ TO DO.
132
+
133
+ For now, the creation of the dataset is documented in the technical report.
134
+ </details>
135
+
136
  ## Terms of Use
137
 
138
  By using the dataset, you agree to comply with the original licenses of the source content as well as the dataset license (CC-BY-4.0). Additionally, if you use this dataset to train a Machine Learning model, you agree to disclose your use of the dataset when releasing the model or an ML application using the model.