Clarity-Former: Tackling Image Degradation and Geometric Diversity in Underwater Instance Segmentation (opens in new tab)

Effective visual analysis of underwater environments for applications in geoscience and remote sensing is fundamentally constrained by two principal factors: severe image degradation inherent to the aquatic medium and the extreme geometric diversity of instances. Existing methods often struggle as they rely on separate enhancement modules or complex architectures that are not optimized for these conditions. To overcome these limitations, we propose Clarity-Former, a novel one-stage, end-to-en...