Variations of the U-Net model with enhancements for retinal vessels segmentation using the DRIVE dataset. It is used weight masks for the BCE loss specifically designed for this task, alongside novel metrics for evaluating models performance regarding thin vessels only. This code was originally used for conducting the experiments described in the paper {LINK TO THE PAPER}.