Hi! Question about the calculation of φp .

[gzhhhere]

How to match the output of convolutional network G and MLP parameters φp? Since every Yp uses a different MLP, the
amount of parameters of MLPs(64-64-64-64-3) is far more than the amount of elements of the feature map（e.g. 16161024）.I believe somewhere my understanding was wrong. Could you please give a simple example to correct me?

[vinthony]

Hi, I have asked the same question by email to the author, here is the response:

After the last layer of G we add a 1x1 conv layer with a number of channels that equals the number of parameters needed for the MLP (instead of 1024). So for example for 5 layers MLP this will be 5*(64*64+64) (up to the number of input and output channels). The kernel size in this layer is 1x1 (instead of 3x3). I'll correct the SM. Thank you for noticing this.

Hope it helps!