ReefNet: A Large-scale, Taxonomically Enriched Dataset and Benchmark for Coral Reef Classification

Yahia Battach, Abdulwahab Felemban, Faizan Farooq Khan, Yousef A. Radwan, Xiang Li, Luis Silva, Rhonda Suka, Karla Gonzalez, Fabio Marchese, Ivor D. Williams, Burton H. Jones, Sara Beery, Francesca Benzoni, Mohamed Elhoseiny KAUST · MIT · Red Sea Global NeurIPS 2025 (Submitted)
Paper Supplementary Code Dataset

Abstract

ReefNet introduces the first large-scale, taxonomically enriched benchmark for coral reef classification. It features over 32k images annotated at the genus level across 400+ sites globally. We define three core tasks — domain-adaptive classification, coral localization, and temporal monitoring — enabling coral reef research across geographies and seasons. We present baseline results and evaluate state-of-the-art methods across task-specific splits. ReefNet enables scalable ecological analysis and paves the way for domain-adaptive AI models for marine biodiversity monitoring.

BibTeX

@article{battach2025reefnet,
  title={ReefNet: A Large-scale, Taxonomically Enriched Dataset and Benchmark for Coral Reef Classification},
  author={Battach, Yahia and Felemban, Abdulwahab and Khan, Faizan Farooq and Radwan, Yousef A. and Li, Xiang and Silva, Luis and Suka, Rhonda and Gonzalez, Karla and Marchese, Fabio and Williams, Ivor D. and Jones, Burton H. and Beery, Sara and Benzoni, Francesca and Elhoseiny, Mohamed},
  journal={arXiv preprint arXiv:XXXX.XXXXX},
  year={2025}
}