SemArt+: Multimodal Benchmark for European Paintings from 3rd to 19th Centuries

SemArt+ is a multi-relational multimodal benchmark for art understanding. 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. The dataset was built by combining the original SemArt dataset with per-sentence aspect annotations from Explain Me the Painting and is curated by Antonio Purificato.

Use Cases

• Train multimodal models for art understanding based on the described image-text-graph structure.
• Benchmark semantic relationship extraction models using the 151,430 typed semantic edges.
• Analyze connections between visual art and descriptive text across centuries of European paintings.

Strengths

• Contains 34,770 images of European paintings, providing a substantial visual corpus.
• Includes 151,430 typed semantic edges connecting images and texts, enabling relational analysis.
• Covers a long temporal range from the 3rd to the 19th century.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic and temporal bias inherent to its source of European paintings.

Provenance

Source

Combination of the original SemArt dataset and the Explain Me the Painting dataset.

Collection Method

Built by combining datasets and adding per-sentence aspect annotations.

Time Range

3rd to 19th centuries

Freshness

Last updated 2026-07-21 13:44:40; freshness should be verified.

Geography

European paintings