Deep learning accurately identifies fjord benthic foraminifera

This dataset includes images (n=3,003) and associated annotations for benthic foraminifera (22,138 individuals) from four Skagerrak fjords (Swedish west coast). Annotations were collected using the Roboflow platform (https://roboflow.com/) and are provided in both YOLOv7-PyTorch and COCO format. These data were used to train a YOLOv7 deep learning segmentation model which is able to identify 29 species of fjord benthic foraminifera with 90.3% mAP (mean average precision).