can anyone tell the matrix representation of transposed convolution filter when filter size is two and stride number is two with more than one input channels on both one and two dimensions?
I googled my problem and all the articles talking about transposed convolutions are intuitive. I want to know the exact matrix representation of transposed convolution filters especially when input channel is greater than one. Thanks.
The Theano conv arithmetic might help with the basic understanding here.
I’m not sure, what “matrix representation” means. Do you mean the convolution operation written as a matrix operation?