Fig. 8: Schematics of the architecture of our surface regression deep learning prediction model.

a Overall architecture of the model. b Basic architecture of the Adapter Net. c Basic architecture of the modified DenseNet121. d Structure of the Dense Block. e Structure of the Dense Layer. f Structure of the Transition block. g Schematic illustration of data flow in the neural network. h The principle of guided back-propagation. i The principle of Class Activation Mapping.