The conservation of marine resources and the mitigation of marine pollution require strengthened knowledge of marine biodiversity, particularly in the deep sea. Videos and images are valuable for documenting the distribution of deep-sea organisms, but manual processing is labor-intensive and variable, emphasizing the need for automated methods. To address this, the J-EDI Organism Detection Dataset (JODD) is introduced. This dataset comprises 8151 images and 15,621 bounding boxes annotated in the Common Objects in Context (COCO) format. The images were captured during deep-sea surveys conducted by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) between 1984 and 2021, using remotely operated vehicles (ROVs) and human-occupied vehicles (HOVs). All images were derived from publicly available videos in JAMSTEC's E-library of Deep-sea Images (J-EDI). The dataset includes 20 object categories-19 biological groups and one machine category-providing a reusable resource for developing and benchmarking machine learning models for the automatic detection of deep-sea organisms.
Keywords: Artificial intelligence; Benthos; Deep learning; HOV; ROV; Seafloor.
© 2026 The Author(s).