regarding the model parameters specified by Nin, Nout and fsize

[wenouyang]

Hi Parminder,

I have a question regarding one statement made in the paper of HyperNetworks. In section 3.1, authors state

In this section we will describe how we construct a hypernetwork for the purpose of generating the
weights of a feedforward convolutional network (similar to Figure 2). In a typical deep convolutional
network, the majority of model parameters resides in the kernels within the convolutional layers.
Each kernel contain Nin × Nout filters and each filter has dimensions fsize × fsize.

I am not very clear about the definition of Nin, Nout, and fsize. Normally, for each layer of CNN, we can have M filters, each filter has a receptive field X. We can have M*X parameters for a given layer. How should I map this with the Nin, Nout and fsize mentioned in the paper.

Thanks,
wenouyang