Abstract: Highlights•We propose a novel non-pretrained deep supervision strategy for polyp segmentation.•We design a parallel dual attention module to enhance the global modeling ability.•The method has been evaluated widely, outperforming state-of-the-art methods.
Loading