Butterfly Phenotype–Taxonomy Clustering Benchmark

This dataset supports the experiments reported in “Self-Supervised Vision Embeddings Reveal Phenotype–Taxonomy Structure in Digitized Butterfly Collections.” It contains benchmark metadata, source database labels, specimen or occurrence identifiers, source image URLs where available, relative image paths, run-level result tables, paper-facing figures, qualitative visualizations, and precomputed DINOv3 embeddings for the ten-genus Nymphalid butterfly benchmark. Image files are redistributed only where permitted by source licenses; otherwise records are linked through available source URLs or occurrence identifiers. DINOv3-derived embeddings are provided as model-derived artifacts and remain subject to the applicable DINOv3 license and attribution terms. Precomputed embeddings were generated using DINOv3 as a frozen feature extractor. DINOv3 model weights are not redistributed in this archive. The DINOv3-derived embeddings are included for reproducibility and should be used with appropriate attribution to DINOv3 and subject to the applicable DINOv3 license terms. Code and notebooks are available at:https://github.com/aruntipingiri/butterfly-phenotype-taxonomy-clustering