Benthos Dataset; segmented photomosaics from three oceanic environments
Yuval, Matan; Alonso, Iñigo; C. Murillo, Ana; Treibitz, Tali (2021), Benthos Dataset; segmented photomosaics from three oceanic environments, Dryad, Dataset, https://doi.org/10.5061/dryad.8cz8w9gm3
Reefs constitute some of the richest and most diverse environments on the planet, manifesting in complex biogenic 3D structures. As such, they are intensively studied. In benthic images, full segmentation (where every pixel is classified) is difficult and labour-intensive to achieve using human-based manual labeling. To solve that, we utilize label-augmentation, i.e., propagation of sparse manual labels to accelerate this task and achieve full map segmentation.
The Benthos dataset contains fully labeled photomosaics from three oceanic environments with over 4500 segmented objects useful for research in computer-vision and benthic ecology. This folder also contains a Matlab script for extracting community statistics from labeled photomosaics.
Microsoft, Award: AI for Earth: AI for Coral Reef Mapping