I am also wondering about this.
It must depend both on convolutional sub-network output size (N) and number of classes (M).
Maybe there are some rules of thumbs depending on (N, M).
Why 2 dense layers and not, say, 3 or 4 ?
Is it better to have all dense layers (except last) the same size ? or decreasing ? or increasing ? or pyramidal ?
Is it better to have ...