Automating the assessment of biofouling in images and video footage

Images and annotations used for training the computer vision models in the <i>Automating the assessment of biofouling in images using expert agreement as a gold standard</i> (2021) paper. Please cite this paper if you use this dataset.We include biofouling (SLoF), paint damage (Paint quality), and niche area annotations in the metadata. For biofouling, we use the Simplified Level of Fouling (SLoF) scale0: No fouling organisms, but biofilm or slime may be present.<br>1: Fouling organisms (e.g. barnacles, mussels, seaweed or tubeworms are visible but patchy (1-15% of surface covered).<br>2: A large number of fouling organisms are present (16-100% of surface covered).<br>For paint quality, we use the following scale1: Paint not present or in poor condition (16-100% of surface scratched/corroded/fouled).<br>2: Paint visible and in fair condition or slightly obscured (1-15% of surface scratched/corroded/fouled)<br>3: Paint visible and in good condition.We are also releasing models trained on this dataset. Please see this github page for further information on using them.We also include footage from an underwater ROV inspecting two small vessels, to illustrate the utility of the model.We further include the expert versus expert annotations, and an experiment with participants in grading imagery using Amazon Mechanical Turk. For completeness, we also provide the original LOF annotations, of which the quality will vary significantly depending on the dataset.<br>