I have a 2d Gaussian filter h, and I have initialised a Conv2d filter with
conv_filter = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=(sigma))
I attempted to set the weights with
with torch.no_grad():
conv_filter.weight = torch.nn.Parameter(torch.from_numpy(h).float())
But I get an error saying that 3 dimensions is expected for the weight. I expanded the dimension with a newaxis by
with torch.no_grad():
conv_filter.weight = torch.nn.Parameter(torch.from_numpy(h[:,:, np.newaxis]).float())
but then I receive RuntimeError: expected stride to be a single integer value or a list of 1 values to match the convolution dimensions, but got stride=[1, 1]
I finally tried to set stride to be 1, but the line:
conv_filter = torch.nn.Conv2d(in_channels=1, out_channels=1, strides = 1, kernel_size=(sigma))
says that "strides" is an unexpected arguments, and changing the spelling to "stride" has no effect and still yields the same Runtime error message.
