Filament identification workflow (left). Inputs for the algorithm are the CD map and the skeleton map, which are passed through the preprocessing module (Section 3.2.1 ), followed by input to the machine learning algorithm. Architecture of the U-Net model with an image subdivision size of 64 × 64 (Section 3.2.1). The size of each layer is indicated in the figure. The dotted orange lines represent the skip connections between the encoder and decoder layers.