Global climate change and other anthropogenic stressors have heightened the need to rapidly characterize ecological changes in coral reefs and marine benthic communities across large scales. We are motivated by improving automatic annotation for large underwater image sets, in part with the use of fluorescence imaging, by collecting in-situ data in hard-to-reach environments underwater and by the challenge of observing micro-scale processes in benthic ecosystems. Being that coral reefs are inherently inaccessible and challenging environments to work in, the challenge of collecting photographic data is not small. Nonetheless, underwater imaging can be a powerful tool especially for monitoring long-term changes as well as in-situ processes.
We work to improve semantic segmentation, which attempts to provide per‐pixel semantic labels, which is an essential task when processing survey data in remote and challenging scenarios. In this effort, we propose and validate an effective approach for learning semantic segmentation models from sparsely labeled data [1, 9]. One major contribution to improving automatic image segmentation was the development of the FluorIs system which takes advantage of the phenomenon of coral fluorescence. The FluorIs system is based on a consumer camera modified for greatly increased sensitivity to chlorophyll-a fluorescence. Its success has been demonstrated in many underwater scenarios and used to investigate several processes including coral recruits , in benthic surveys , for underwater wide field-of-view fluorophore surveying during both night and day  and in wide field-of-view images of coral reefs .
Also, with the intention of facilitating rapid analysis of large sparsely labeled underwater datasets we use transfer learning techniques to improve automatic coral segmentation . Our work extends to monitoring over long periods of time to understand why massive Caribbean corals aren’t recovering from repeated thermal stress events during 2005-2013 .
Also, with our underwater microscope we enable in situ observations at previously unattainable scales. This instrument can provide important new insights into micro-scale processes in benthic ecosystems that shape observed patterns at much larger scales . We created a lightweight monopod image-framing system and a custom semi-automated image segmentation analysis program for measuring size and growth rates for individual colonies in coral communities .
Our advancements in underwater in-situ surveying and monitoring have been mostly demonstrated in coral reefs and benthic environments, but we believe that these contributions are significant in underwater research technologies  and can be applied to many research questions that remain about the underwater ecosystem in a time where drastic ecological changes due to local and global stressors are an inevitable part of our future.
This data package contains all images and point annotations used in the present publication.
This dataset contains fully labeled photomosaics from 3 oceanic environments with over 4500 segmented objects. This folder also contains a Matlab script for extracting community statistics from labeled photomosaics.