Fig. 6
From: Fully convolutional architecture vs sliding-window CNN for corneal endothelium cell segmentation
Three representative examples (high ECD in a-f, low ECD in g-l, high CV in m-r) for both networks. (a,g,m) Intensity images. (b,h,n) Outcome of the SW-net. (c,i,o) Segmentation after postprocessing of the SW-net outcome. (d,j,p) Outcome of the U-net. (e,k,q) Segmentation after postprocessing of the U-net outcome. Green arrows indicate true edges that were weak in the CNN output but detected by the postprocessing. Blue arrows denote true edges that were missed by the postprocessing, either because they were weak edges or because a small cell surrounded by large cells was smoothed away. Red arrows indicate false edges and mistakes in general. (f,l,r) Segmentation provided by the Topcon microscope’s built-in software