Datasets:
janvanlooy
commited on
Commit
•
5d22c29
1
Parent(s):
f5edb1a
Various changes
Browse files
README.md
CHANGED
@@ -119,17 +119,18 @@ python extract_images.py --parquet_file <Path to the Parquet file or folder cont
|
|
119 |
If you want to contribute to the dataset, the best way is to help us develop pipeline components for further processing. Components
|
120 |
we are currently looking to add are the following ([GitHub issues](https://github.com/ml6team/fondant/issues?q=is%3Aissue+is%3Aopen+label%3A%22Component+Contribution%22)):
|
121 |
- Image-based deduplication
|
122 |
-
- Visual quality / aesthetic quality estimation
|
123 |
- Automatic captioning
|
124 |
-
-
|
125 |
- Watermark detection
|
|
|
|
|
126 |
- Face detection
|
127 |
- Personal Identifiable Information (PII) detection
|
128 |
- Text detection
|
129 |
- AI generated image detection
|
130 |
-
- CLIP embedding generation
|
131 |
- Image-text CLIP similarity
|
132 |
-
-
|
|
|
133 |
We are also looking for core framework contributors and users who are willing to give feedback on usability and suggest potential improvements
|
134 |
|
135 |
|
@@ -165,11 +166,11 @@ fondant-cc-25m is built from CommonCrawl dumps. These dumps are constructed from
|
|
165 |
Permissive licenses have minimal restrictions on how the image can be copied, modified, and redistributed.
|
166 |
The full list of licenses can be found [here](https://creativecommons.org/about/cclicenses/).
|
167 |
We examined HTML tags of the webpages for the presence of Creative Commons license URLs. A webpage was marked permissive only when a license URL was found in
|
168 |
-
its footer, aside or sidebar. This was the case only in around 0.164% of a 100k sample
|
169 |
-
the public internet may be trained on up to 99.836% images
|
170 |
|
171 |
-
Subsequently, all the image URLs present on the web page were collected together with the license information.
|
172 |
-
|
173 |
False positives could be due to parsing errors but also incorrect attributions: images indicated by the publisher to be CC which are not.
|
174 |
More information on our approach can be found in [this blogpost](https://blog.ml6.eu/ai-image-generation-without-copyright-infringement-a9901b64541c).
|
175 |
|
|
|
119 |
If you want to contribute to the dataset, the best way is to help us develop pipeline components for further processing. Components
|
120 |
we are currently looking to add are the following ([GitHub issues](https://github.com/ml6team/fondant/issues?q=is%3Aissue+is%3Aopen+label%3A%22Component+Contribution%22)):
|
121 |
- Image-based deduplication
|
|
|
122 |
- Automatic captioning
|
123 |
+
- Visual quality / aesthetic quality estimation
|
124 |
- Watermark detection
|
125 |
+
- Not safe for work (NSFW) content detection
|
126 |
+
- CLIP embedding generation
|
127 |
- Face detection
|
128 |
- Personal Identifiable Information (PII) detection
|
129 |
- Text detection
|
130 |
- AI generated image detection
|
|
|
131 |
- Image-text CLIP similarity
|
132 |
+
- Any components that you propose to develop\
|
133 |
+
|
134 |
We are also looking for core framework contributors and users who are willing to give feedback on usability and suggest potential improvements
|
135 |
|
136 |
|
|
|
166 |
Permissive licenses have minimal restrictions on how the image can be copied, modified, and redistributed.
|
167 |
The full list of licenses can be found [here](https://creativecommons.org/about/cclicenses/).
|
168 |
We examined HTML tags of the webpages for the presence of Creative Commons license URLs. A webpage was marked permissive only when a license URL was found in
|
169 |
+
its footer, aside or sidebar. This was the case only in around 0.164% of a 100k random sample from Common Crawl. This suggests that image generation models
|
170 |
+
trained on a random sample from the public internet may be trained on up to 99.836% copyrighted images.
|
171 |
|
172 |
+
Subsequently, all the image URLs present on the web page were collected together with the license information. A manual check of a random
|
173 |
+
sample of 1032 images showed that 96.32% were attributed the correct license whil 3.68% were not.
|
174 |
False positives could be due to parsing errors but also incorrect attributions: images indicated by the publisher to be CC which are not.
|
175 |
More information on our approach can be found in [this blogpost](https://blog.ml6.eu/ai-image-generation-without-copyright-infringement-a9901b64541c).
|
176 |
|