A module is a reusable network substructure made from a sequence of layers. It packages a repeated pattern so a larger model can organize its backbone into clear, replaceable blocks. section 13.4.1
A convolutional module can combine convolutions, activations, and pooling into one forward-processing unit. The example block uses two 3 × 3 convolutions, each followed by an activation, then max-pooling. section 13.4.1
A convolutional module
LunaBlock accepts an input channel count and a convolution channel count, applies two Conv3d layers with ReLU activations, and returns the result of MaxPool3d. Similar blocks can form the backbone, even when channel sizes change between blocks. section 13.4.1
