The Convolutional Layer

Think of a convolution as nothing more than an image transformer with three key elements. First, there is an input image, then a filter, and finally, a feature map.

This section will cover each of these in turn to give you a solid idea of how images are filtered in a convolutional layer. The convolution is the process of passing a filter window over the input data, which will result in a map of activations known as a feature map. The input data may be the input image to the model or the output of a prior, intermediary layer of the model. The filter is generally a much smaller array, such as 3x3 for two-dimensional data, in which the specific values of the filter are learned during the training process. The filter passes across the input data with a window size equal to the size of the filter, then, the scalar product of the filter and section of the input data is applied, producing what's known as an activation. As this process continues across the entire...