Figure 4. The U-Net model canonically segments bending sprouts, although it was trained on linear segmentation. The examples demonstrate U-Net’s ability to detect bending sprouts and correct for systematic bias by using only straight lines in the manual analysis.