Skip to main content
. 2023 Mar 30;13:5177. doi: 10.1038/s41598-023-32514-7

Figure 3.

Figure 3

Mean errors and weighted kappa coefficients of landmark detection with crowding categorization. (A) The maxilla models consistently show smaller error values than the mandible ones. The mean errors decrease in the order of ResNet50, ResNet101, VGG16, and VGG19. (B) All kappa values suggest that the strength of agreement for crowding classification is substantial, and the agreement becomes higher as the layer becomes deeper in the same network models.