README.md

---
license: cc-by-sa-4.0
task_categories:
- visual-question-answering
- question-answering
- text-generation
language:
- en
pretty_name: GIMMICK QA Base Dataset
size_categories:
- 1K<n<10K
---

![GIMMICK Overview](https://raw.githubusercontent.com/floschne/gimmick/refs/heads/main/assets/gimmick_fig1.png "GIMMICK Overview")

# GIMMICK QA Base Dataset

This is the **base** dataset for the GIMMICK Cultural Origin QA (COQA) **and** GIMMICK Cultural Knowledge QA datasets introduced by our paper [`GIMMICK - Globally Inclusive Multimodal Multitask Cultural Knowledge Benchmarking`](https://huggingface.co/papers/2502.13766).

## Dataset Details

The dataset comprises 728 cultural events or facets from 144 countries in six global regions depicted by 6857 images from the [UNESCO ICH project](https://ich.unesco.org/). More details are reported in the paper.

For information on how to use the dataset for the COQA and CKQA tasks, please refer to the [GitHub Repository](https://www.github.com/floschne/gimmick) :-)

## Citation

```bibtex
@article{schneider2025gimmick,
      title={GIMMICK -- Globally Inclusive Multimodal Multitask Cultural Knowledge Benchmarking}, 
      author={Florian Schneider and Carolin Holtermann and Chris Biemann and Anne Lauscher},
      year={2025},
      eprint={2502.13766},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.13766}, 
}
```