Why No. of filters are keep increasing as we go deep in CNN?

In many of the CNN architectures, I have observed that the number of filters(or depth of output layer) keeps increasing as we go deeper in CNN. Is there a rationale or intuition that follows for making such a CNN architecture or is it just empirical for first architecture and then just community followed without bothering to change it? To me the second case seems quite less plausible and I think there must be some concrete reason for the CNN creators to keep increasing (or at least keep constant) the filter size as we go Deeper in the network

Hi mayankb2103,

This may answer your question.