Dual SAM: Efficient Marine Animal Segmentation with Comprehensive Prior Knowledge

The core message of this work is to propose a novel feature learning framework, named Dual-SAM, that enhances the Segment Anything Model (SAM) for high-performance Marine Animal Segmentation (MAS). The framework incorporates dual branches, multi-level coupled prompts, dilated fusion attention, and criss-cross connectivity prediction to effectively leverage prior knowledge from underwater images and improve the localization and structural perception of marine animals.