The [BioTrove dataset](https://baskargroup.github.io/BioTrove/) is a massive collection of biodiversity images curated from iNaturalist. For this competition, you’ll work with a~50,000-image subset where only family-level labels are provided.

Using image features and optional family-level priors (e.g., in contrastive learning), participants are challenged to learn clusters that approximate the hidden biological taxonomy. Evaluation is based on how well your clusters match hidden genus- and species-level groupings, measured using Normalized Mutual Information (NMI).