UnderwaterDG: Domain Generalized Semantic Segmentation in Underwater Scenes via Decoupling Visibility and Appearance (opens in new tab)

Domain generalized semantic segmentation (DGSS) has made significant progress in urban scenes but remains unexplored in marine environments. Underwater images present diverse appearance and visibility variations caused by water quality, lighting, and habitat conditions, which significantly differ from those in urban scenes, making urban DGSS methods less effective in such environments. To fill this gap, we formally investigate DGSS for underwater scenes, termed UnderwaterDG, and propose a nov...