AA-Caps Architecture Clause Samples

AA-Caps Architecture. Figure 2 shows the architecture proposed for AA-Caps. The architecture consists of a 2D Convolutional layer (Conv2d), a Primary Capsule Layer (Primary-Caps), a Self-Attention layer (Self-Attention), a 1D Con- volutional layer (Conv1d) and it ends with a classifica- tion layer based on the attention outputs. The input im- age of the model is defined over m channels, the number of channels depends on the dataset to be considered. The image is fed into the Conv2d that applies a 2D convo- lution on m channels and extracts 256 feature maps. It applies a kernel = 9 9 and stride = 1 1. This layer is the original feature map extraction described for CapsNet([45]). The Primary-Caps layer takes the 256