Coral reefs, vital for marine biodiversity, are difficult to monitor due to their complex structures and the manual effort required for analysis. This project addresses the challenge by offering an advanced foundation model for the automatic dense segmentation of coral reefs. It uses a novel parallel semantic branch to precisely label different coral entities and delineate their boundaries. Trained on the CoralMask dataset, which includes over 41,000 labelled images, the technology enhances coral analysis by providing high-quality, detailed coral masks, improving scalability and accuracy in reef monitoring