StatMapCorpus contains 23,549 English-language statistical maps identified in MapPool (Schnürer 2024), a collection of 75.9 million map-like images drawn from CommonPool/DataComp. Candidates were narrowed by a multilingual keyword filter over alt-text, an SVM-RBF classifier over CLIP ViT-L/14 embeddings trained by active learning (retained at p ≥ 0.80), and a cascade of vision-language model gates. Every record is a statistical map: values attached to administrative or statistical units, a quantitative legend where one is present, lettered in English where the map carries any text, and at least one identified cartographic method.
The deposit contains annotations and identifiers only; no images are redistributed. Each record carries the image URL and file-verification surrogates (sha256, 64-bit pHash) computed from the exact file that was annotated, and an included script retrieves the images and verifies them against the published checksums. All records were reachable on 1 August 2026; because the population comes from a web crawl, the reachable fraction can decrease from that date.
The annotation layer provides a multi-label cartographic method taxonomy (choropleth, diagrams, dot density, isolines, cartogram, flow map, heat map), a classed-legend flag, a short model-generated content description, and a CLIP ViT-L/14 embedding inherited from MapPool. Method labels are strongly skewed toward choropleth. Byte-identical copies are flagged; near-duplicates can be recovered from the published pHash, and CODEBOOK.md documents every column and explains how to group them when partitioning.
Annotations are model-generated and have not been validated against human coding; the annotation layer should be treated as a silver standard, and a human-coded validation is planned for a subsequent version. The corpus is not a random sample of statistical maps — it inherits the composition of a web crawl and the selection criteria above.
The annotations are licensed CC BY 4.0. Rights in the images remain with their owners; NOTICE.md sets out attribution, the derivation chain and the takedown procedure.
(2026-08-23)
