Openaccess Embeddings Siglip2 Naflex: Image Embeddings for Public-Domain Artworks

Image embeddings for every public-domain artwork in the Metropolitan Museum of Art's Open Access collection, produced by the google/siglip2-so400m-patch16-naflex model. The dataset includes an objectID key for joining with the main image dataset and L2-normalized embeddings with a dimension of 1152. It was created by the Met Museum and last updated on HuggingFace in May 2026.

Use Cases

• Semantic search for artworks based on visual similarity using the provided embeddings.
• Training or fine-tuning multimodal models on a corpus of art images.
• Clustering or categorizing artworks by visual style or content based on embedding vectors.
• Developing recommendation systems for museum collections or art education platforms.

Strengths

• Embeddings are L2-normalized, which is a standard practice for similarity search.
• Each record includes a primary key (objectID) for reliable joins with the source image dataset.
• The source model (google/siglip2-so400m-patch16-naflex) and embedding dimension (1152) are explicitly documented.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation beyond the provided notes is absent; field semantics must be inferred after download.
• Image bytes are not stored, requiring a separate download and join operation for full use.

Provenance

Source

Metropolitan Museum of Art (Met Museum) Open Access collection.

Collection Method

Images from the openaccess dataset were processed by the google/siglip2-so400m-patch16-naflex model to generate embeddings.

Freshness

Last updated 2026-05-07 00:33:01; freshness should be verified.