Datasets:
license: mit
dataset_info:
features:
- name: image
dtype: image
- name: art_style
dtype: string
- name: painting
dtype: string
- name: emotion
dtype: string
- name: language
dtype: string
- name: text
dtype: string
- name: split
dtype: string
splits:
- name: train
num_bytes: 18587167692.616
num_examples: 62989
- name: validation
num_bytes: 965978050.797
num_examples: 3191
- name: test
num_bytes: 2330046601.416
num_examples: 6402
download_size: 4565327615
dataset_size: 21883192344.829002
task_categories:
- image-to-text
- text-classification
- image-classification
- text-to-image
- text-generation
language:
- en
- ar
- zh
tags:
- Affective Captioning
- Emotions
- Prediction
- Art
- ArtELingo
pretty_name: ArtELingo
size_categories:
- 100K<n<1M
ArtELingo is a benchmark and dataset introduced in a research paper aimed at promoting work on diversity across languages and cultures. It is an extension of ArtEmis, which is a collection of 80,000 artworks from WikiArt with 450,000 emotion labels and English-only captions. ArtELingo expands this dataset by adding 790,000 annotations in Arabic and Chinese. The purpose of these additional annotations is to evaluate the performance of "cultural-transfer" in AI systems.
The dataset in ArtELingo contains many artworks with multiple annotations in three languages, providing a diverse set of data that enables the study of similarities and differences across languages and cultures. The researchers investigate captioning tasks and find that diversity in annotations improves the performance of baseline models.
The goal of ArtELingo is to encourage research on multilinguality and culturally-aware AI. By including annotations in multiple languages and considering cultural differences, the dataset aims to build more human-compatible AI that is sensitive to emotional nuances across various cultural contexts. The researchers believe that studying emotions in this way is crucial to understanding a significant aspect of human intelligence.
In summary, ArtELingo is a dataset that extends ArtEmis by providing annotations in multiple languages and cultures, facilitating research on diversity in AI systems and improving their performance in emotion-related tasks like label prediction and affective caption generation. The dataset is publicly available, and the researchers hope that it will facilitate future studies in multilingual and culturally-aware artificial intelligence.