Datasets:
task_categories:
- visual-question-answering
language:
- fra
license: cc-by-sa-4.0
Description
This dataset is a processed version of wikimedia/wit_base.
We converted the images to PIL format and kept only the lines containing French.
The French texts come either from the caption_attribution_description column of the original dataset which we have renamed caption here, or from the wit_features column which we have renamed descriptions.
The descriptions column is a list because it can contain several texts itself.
For example, line 8 of the dataset contains three texts in the descriptions column because the image is present in three different Wikipedia articles (Tahsin Yazıcı, Walton Harris Walker, La brigade turque).
Furthermore, given that we propose viewing this dataset as a visual question answering dataset, we have added a column question containing 58 different questions written manually. It is therefore up to the user to use columns descriptions and caption as answers.
Finally, note that we have left the image_url column as a key so that the user can differentiate in the original dataset which ones are in French and which ones are not (if there is a desire to augment the dataset by translating from language A into French while not wanting to translate images already available in French).
Citation
@article{srinivasan2021wit,
title={WIT: Wikipedia-based Image Text Dataset for Multimodal Multilingual Machine Learning},
author={Srinivasan, Krishna and Raman, Karthik and Chen, Jiecao and Bendersky, Michael and Najork, Marc},
journal={arXiv preprint arXiv:2103.01913},
year={2021}
}