Machine Learning with Swift

Convolutional layer

In CNNs convolutions happen in the special layers, called convolutional layers. Each layer has an array of convolutional filters, which also can be seen as one 3D convolutional filter with width, height, and number of channels (or depth). In the first convolutional layer, we usually want to have 3 channels corresponding to the RGB channels of the input images:

Figure 9.7: The first convolutional layer takes a batch of images as its input and outputs a batch of feature maps. The results of each of the 16 red, green and blue filters are being summed up to obtain the final feature map

The output of a convolution layer is ...