SUIM stands for Segmentation of Underwater IMagery, which can be used for semantic segmentation of natural underwater images. The dataset includes more than 1,500 images. The dataset has pixel annotations for 8 object categories: fish (vertebrates), coral reefs (invertebrates), aquatic plants, wreckage/ruins, human divers, robots, and seabed.

These images were rigorously collected during ocean exploration and human-robot collaboration experiments and annotated by human participants.