FADFISH: An annotated underwater video dataset of pelagic fish species associated with Fish Aggregating Devices (FADs) for deep learning applications

Résumé fourni par la source

Dataset Overview The FADFISH dataset is an annotated underwater video dataset specifically designed for pelagic fish species detection and classification. It comprises 3589 annotated images extracted from underwater videos recorded around Fish Aggregating Devices (FADs) in Bali Sea (Indonesia), the Western Indian Ocean (Seychelles EEZ), and the Western Central Pacific Ocean (International waters offshore the Kiribati Islands). Images contain 20,723 bounding box annotations across 25 classes (21 species-level and 4 general categories), including commercially important species such as yellowfin tuna (Thunnus albacares), mahi-mahi (Coryphaena hippurus), and vulnerable species such as silky sharks (Carcharhinus falciformis). Annotations are provided in both YOLO and COCO (JSON) formats for compatibility with popular deep learning frameworks. The dataset captures the unique challenges of FAD environments: high fish density, multi-species aggregations, presence of the FAD structure, predominance of small objects and variable underwater visibility. Funding This research was funded by Biodiversa+, the European Biodiversity Partnership, in the context of the MOOBYF project under the 2022-2023 BiodivMon joint call. It was co- funded by the European Commission (GA No. 101052342) and the following funding organisations: ANR - Agence Nationale de la Recherche and Office Français de la Biodiversité – France, FONDS DE LA RECHERCHE SCIENTIFIQUE – FNRS - Belgium, Ministry of Universities and Research, Italy, and Deutsche Forschungsgemeinschaft eV; BMBF-VDI/VDE INNOVATION + TECHNIK GMBH, Germany. Images were also collected during the INNOV-FAD project (European Maritime and Fisheries Fund, measure n°39, OSIRIS #PFEA390017FA1000004, and France Filière Pêche) and the International Seafood Sustainability Foundation (www.iss-foundation.org) under its Bycatch Mitigation Project. If using the dataset, please cite the data paper:[Title of Paper], [Journal Name], [Year].DOI: [DOI of paper] The dataset is part of an ongoing research project; it will be made publicly available following completion of the project and publication of the associated manuscript. Access may be granted upon request during the restricted period. Data managers: Hartaty, Hety (UMR MARBEC, Univ. Montpellier, CNRS, Ifremer, IRD, Sète, France, National Research and Innovation Agency (BRIN), Bogor, Indonesia) - hety.hartaty@ird.fr; hetyhartaty@brin.go.id Restrepo-Ortiz, Claudia Ximena (UMR MARBEC, Univ. Montpellier, CNRS, Ifremer, IRD, Sète, France) - claudia.restrepo-ortiz@ird.fr

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.