Handika et al. Dataset for BioCosmos: computer vision and large-language models for large-scale image-based biodiversity exploration and trait discovery

BioCosmos image occurrences This dataset accompanies Handika et al., in prep. It lists every museum specimen image used by BioCosmos, served by the butterfly portal LepiVerse.com. The dataset contains 619,739 images (dorsal and ventral) of 515,141 specimens, covering 8,401 butterfly species (Lepidoptera: Papilionoidea). A total of 65 institutions host the specimens, aggregated from GBIF, SCAN, and Ecdysis. Files: image_occurrences.parquet and image_occurrences.csv.gz: image occurrence tables image_occurrences_columns.csv: column description image_occurrences_manifest.json: row counts, reference data versions and file checksums MethodsNames were matched to the Catalogue of Life (COL26.10 XR; https://doi.org/10.48580/dgzlf). We validated coordinates against GADM 4.1 (https://gadm.org). Dorsal and ventral views were classified following Lürig et al. 2025 (https://doi.org/10.1111/ele.14495). Each URI points to the provider's original file. LicensesThe table contains metadata only, including the image licenses. The GBIF dataset was downloaded via https://doi.org/10.15468/dl.a8vkwd.