Five models were developed using RootPainter; four to detect and predict the surface area of the deep-sea sponge Mycale lingua and one to identify laser scales. Three of the sponge models were trained and applied to time-lapse images collected by the Lofoten Vesterålen Ocean Observatory. The fourth sponge model and laser model were developed and used on extracted video frames from an ROV survey of the Tisler reef. The total observatory dataset contained 18,346 images, consisting of 9,173 images each of the Mycale lingua sponges 'Magnus' and 'Mini' from 2017-2019. The total ROV video frame dataset contained 1,420 images from the East of the reef, captured in 2021.

Metrics.csv files - contain corrective annotation metrics from training Models 1-4Model results.zip - contain the output segmentations of Models 1-4, as well as the raw and processed (i.e. with artefacts removed and scaling performed) .csv files of extracted areasModel_1-4.pkl files - final RootPainter models used in the studyScript to crop Magnus and Mini - R script used to batch crop raw images from the Lofoten Vesterålen Ocean Observatory (https://loveocean.no/) to produce the datasets to train Models 1 and 1.1 (targeting sponge referred to as Magnus) and 2 (targeting sponge referred to as Mini).Tisler East frames 2021 - contains the cropped ROV video frames used to train Models 3 and 4