Y-Net: Convolutional Networks for Multi-Domain Image Segmentation

Abstract

In order to solve the drawback that most existing SOD networks cannot extract local details and global contrast information well, and often have insufficient detail on the edges, we design an accurate and compact saliency multi-domain image segmentation algorithm, Y-Net for short. This network combines the new U-shaped network U2NetP and RAS Net segmentation network in the field of deep learning, and well combines the characteristics of each of the two segmentation networks through the self-made module and residual mechanism. It is outstanding in segmentation of different types of images. Y-Net has been tested to show stronger performance than the original two base networks in five major public datasets.