lbourdois/caption-wit_base_french

---
task_categories:
- visual-question-answering
language:
- fra
license: cc-by-sa-4.0
---

## Description

This dataset is a processed version of [wikimedia/wit_base](https://huggingface.co/datasets/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](https://fr.wikipedia.org/wiki/Tahsin_Yaz%C4%B1c%C4%B1#/media/Fichier:WaltonWalker&TahsinYazici.jpg), [Walton Harris Walker](https://fr.wikipedia.org/wiki/Walton_Walker#/media/Fichier:General_Walker_and_Lt._Gen_William_Dean.jpg), [La brigade turque](https://fr.wikipedia.org/wiki/Brigade_turque#/media/Fichier:WaltonWalker&TahsinYazici.jpg)).
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}
}
```