metadata
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
dataset_info:
features:
- name: image
dtype: image
- name: text
dtype: string
- name: link
dtype: string
splits:
- name: train
num_bytes: 19444314153
num_examples: 115818
- name: validation
num_bytes: 4201166538
num_examples: 25701
- name: test
num_bytes: 1547637231
num_examples: 9911
download_size: 24003659385
dataset_size: 25193117922
license: apache-2.0
task_categories:
- image-text-to-image
- visual-document-retrieval
pretty_name: SemArtPlus
size_categories:
- 100K<n<1M
Dataset Card for Dataset Name
SemArt+ is a multi-relational multimodal benchmark for art understanding, built by combining the original SemArt dataset with per-sentence aspect annotations from Explain Me the Painting. It contains 34,770 images of European paintings from the 3rd to 19th centuries, connected to 62,289 texts through 151,430 typed semantic edges.
- Curated by: - Antonio Purificato
- Language(s) (NLP): English
- License: Apache 2.0
Dataset Sources
- Repository: https://github.com/antoniopurificato/artistic_sheaf/blob/stable
- Paper: https://arxiv.org/abs/2607.16321v1
Dataset Structure
Image: the image from the WikiArt dataset.Link: type of relation presented in the text.Text: Description related to the given image.
Citation
@misc{schaerf2026artsemanticssheafinformedcontrastive,
title={Art Beyond Semantics: Sheaf-Informed Contrastive Learning for Multi-Relational Representations},
author={Ludovica Schaerf and Antonio Purificato and Piera Riccio and Fabrizio Silvestri and Noa Garcia},
year={2026},
eprint={2607.16321},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2607.16321},
}