README.md

---
configs:
  - config_name: task1
    data_files:
      - split: train
        path: data/train-*
      - split: test_small
        path: data/test_small-*
      - split: test_large
        path: data/test_large-*
  - config_name: task2
    data_files:
      - split: train
        path: task2/train-*
      - split: test_small
        path: task2/test_small-*
      - split: test_large
        path: task2/test_large-*
language:
- fra
task_categories:
- visual-question-answering
size_categories:
- 100K<n<1M
---


## Description

This dataset is a processed version of [worldcuisines/vqa](https://huggingface.co/datasets/worldcuisines/vqa) to make it easier to use.    
Specifically, the original dataset has been processed to provide the image directly as a PIL rather than an URL path in an `image` column.  

Note that this dataset contains only the French part of the original dataset for `task1` (Dish Name Prediction) and `task2` (Location Prediction).  
For further details, please consult the [worldcuisines/vqa](https://huggingface.co/datasets/worldcuisines/vqa) dataset card.  

Finally, the `dataset` column is for internal use by the person who processed the dataset, and can therefore be ignored by the user.


## Citation
```
@article{winata2024worldcuisines,
  title={WorldCuisines: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines},
  author={Winata, Genta Indra and Hudi, Frederikus and Irawan, Patrick Amadeus and Anugraha, David and Putri, Rifki Afina and Wang, Yutong and Nohejl, Adam and Prathama, Ubaidillah Ariq and Ousidhoum, Nedjma and Amriani, Afifa and others},
  journal={arXiv preprint arXiv:2410.12705},
  year={2024}
}
```