Figure 4.
Failed examples caused by decreased vessel density in DeepDRiD dataset. Top row shows 4 pathological images with sufficient image quality and the second row displays corresponding vessel segmentation images predicted by our retinal vessel segmentation model. Misclassification is owing to the presence of only a few visible vessels within the FOV. The file names of color images are (a) “51_r1”, (b) “309_l2”, (c) “67_r2” and (d) “265_r1”.