|
|
--- |
|
|
license: cc-by-4.0 |
|
|
task_categories: |
|
|
- image-to-text |
|
|
language: |
|
|
- de |
|
|
- en |
|
|
task_ids: |
|
|
- image-captioning |
|
|
pretty_name: G400M |
|
|
size_categories: |
|
|
- 100M<n<1B |
|
|
source_datasets: |
|
|
- mlfoundations/datacomp_xlarge |
|
|
--- |
|
|
|
|
|
# Dataset Card for G400M |
|
|
|
|
|
G400M is a German language image-text dataset with 400M image-text pairs extracted from the [xlarge pool of DataComp](https://huggingface.co/datasets/mlfoundations/datacomp_xlarge). |
|
|
The data is filtered and balanced using the algorithm applied by [MetaCLIP](https://github.com/facebookresearch/MetaCLIP), that is: |
|
|
1. Build a collection of 500k strings (namely the metadata) from the German Wikipedia. |
|
|
2. Filter the data pool for German and English data using [fastText](https://github.com/facebookresearch/fastText). |
|
|
3. Apply substring matching to the captions with the metadata. |
|
|
4. Sample the image-text pairs using the algorithm by MetaCLIP with the (magic) target number per metadata entry of 20k. |
|
|
|
|
|
We follow [DataComp](https://huggingface.co/datasets/mlfoundations/datacomp_xlarge) and distribute the image url-text samples and metadata under a standard Creative Common CC-BY-4.0 license. The individual images are under their own copyrights. |
