Figure 1.

U-Net Lumen Segmentation Example

Upper images(A to D) show the contractive path from input image, and lower images(E to G) show expansive path to output segmentation map - first into an increasing number of lower-resolution features along the contractive path (A to D), and then into a decreasing number of higher-resolution features along the expansive path (D to G). At each level, one of many features was arbitrarily chosen and transformed to cartesian coordinates for ease in interpretability.