In this paper, we propose a self-supervised twin network approach based on
this a priori. The method of generating the approximate10 edge information of
an image and then differentially eliminating the edge errors11 in the
reconstructed image with a dilate algorithm. This is used to im