According to doc, the num_filters should be int and dilation should be a tuple. And in tensorflow and pytorch, the dilation should be greater than 0, but in following code, when dilation is 0, ConvolutionTranspose2D outputs normally.
f = C.layers.ConvolutionTranspose2D((3, 4), (1, 1), activation=C.relu, dilation=0)
x = C.input_variable((3, 480, 640))
h = f(x)
print(h.shape)
print(f.W.shape)
# (1, 1, 482, 643)
# (-1, 1, 1, 3, 4)
According to doc, the
num_filters
should be int anddilation
should be a tuple. And in tensorflow and pytorch, the dilation should be greater than 0, but in following code, whendilation
is 0, ConvolutionTranspose2D outputs normally.