Figure 6.

Method for data augmentation. (a) Making changes to figures, the method can usually be separated as a rigid or non-rigid transformation. Rigid transformation geometrically changes the original image, while non-rigid transformation changes the size or shape with respect to the original image. (b) (Upper) The generative part of the GAN neural network yields figures that have similar properties, which can be used to increase the data volume. (Lower) Grad-CAM shows the heatmap of figures, which can be used for annotations, thus reducing manual workloads.