Fig. 2
From: Fully convolutional architecture vs sliding-window CNN for corneal endothelium cell segmentation
a Small, blurred area of a specular image (size 68 ×68 pixels) where the identification of small cells is difficult. b The gold standard (in blue) superimposed on the intensity image. c U-net output for a filter size of 4 ×4. d U-net output for a filter size of 3 ×3 with similar receptive field. e U-net output for a filter size of 5 ×5 with similar receptive field. f Default U-net output for a filter size of 4 ×4, but using the original binary labels. g SW-net output for a filter size of 4 ×4. h SW-net output for a filter size of 3 ×3