I'll mention that num_filters gets referred to as the number of "channels" later in the lecture. Similar to how R, G, and B describe independent features of the same image (they are color channels), each filter defines a channel containing data relevant to the image phenomenon it is designed to detect. Much of the ensuing material deals with representing these channels in the data layout effectively so that we can improve the performance of our computations.
I'll mention that num_filters gets referred to as the number of "channels" later in the lecture. Similar to how R, G, and B describe independent features of the same image (they are color channels), each filter defines a channel containing data relevant to the image phenomenon it is designed to detect. Much of the ensuing material deals with representing these channels in the data layout effectively so that we can improve the performance of our computations.