Lizagrin/wikiart_captions
WikiArt Captions Subset — Multimodal Art Retrieval Dataset This dataset is a curated subset of 6,000 paintings from the WikiArt collection.It was created as part of a project on multimodal art retrieval, combining visual, textual, and semantic information. Each record represents one artwork and includes: Field Description image_row Row index in the source subset (integer) caption Automatically generated textual description (caption) using the BLIP model… See the full description on the dataset page: https://huggingface.co/datasets/Lizagrin/wikiart_captions.
WikiArt Captions Subset — Multimodal Art Retrieval Dataset
This dataset is a curated subset of 6,000 paintings from the WikiArt collection. It was created as part of a project on multimodal art retrieval, combining visual, textual, and semantic information.
Each record represents one artwork and includes: | Field | Description | |-------|--------------| | image_row | Row index in the source subset (integer) | | caption | Automatically generated textual description (caption) using the BLIP model (Salesforce/blip-image-captioning-base) | | artist | Name of the artist (decoded from WikiArt labels) | | genre | Artwork genre (e.g., portrait, landscape, religious painting, etc.) | | style | Artistic style or movement (e.g., Impressionism, Romanticism, Realism, etc.) |
Generation Details
Captions were generated with BLIP-base, an image captioning model trained on large-scale web image–text pairs. All paintings were processed in batches using PyTorch, and each image’s caption was limited to 30 tokens to preserve clarity.
Example of generated captions:
a painting of a vase with flowers in ita painting of a man with a beard and a red baga painting of a river with boats and buildings
License
The dataset is derived from WikiArt artworks and is shared for non-commercial, educational, and research purposes only.