Data for the Fons Martí Massafont Costals collection: OpenCLIP embeddings, clusterings, and an interactive cluster tree

Data files derived from the 9,021 digitised photographs in the Fons Martí Massafont Costals collection, held by the Centre for Image Research and Dissemination (CRDI) of Girona City Council and made publicly available through Europeana under CC BY-NC-ND 4.0. The dataset contains OpenCLIP ViT-L/14 (LAION-2B) image embeddings, UMAP dimensionality reductions, HDBSCAN cluster labels, metadata for all 9,021 records, and a self-contained interactive dendrogram of the cluster hierarchy. It accompanies two articles. The Visual World of Martí Massafont: Using Multimodal AI to Analyse Commercial Street Photography in Girona, 1947–1997 (Smits, forthcoming) clusters a 10-dimensional reduction into 108 groups; code at https://github.com/tpsmi/streetphotography-visual-grammar. A Jagged Cut: Reading Visual Archives at Multiple Resolutions with HDBSCAN (Smits, forthcoming) uses a 25-dimensional reduction producing 72 clusters, and reads the full HDBSCAN hierarchy rather than the selected clusters alone; code at [second repository URL]. The photographs themselves are not included; they can be downloaded from Europeana using the code in either repository.