I am trying to do upsampling in UNet, but while concatenating the encoder matrix with the decoder matrix I am getting a different matrix, and here I don’t want to downsize the decoded matrix.
You cannot reshape a tensor of the shape
[1, 1, 3, 50] into
[1, 1, 3, 75] as the number of elements differ.
Depending on your use case you could either repeat a subset of the original tensor or use an interpolation method to increase the number of elements e.g. via
out = F.interpolate(torch.randn([1, 1, 3, 50]), (3, 75)).
Thanks, it worked.
Ya, I used the wrong word “reshape”.