ValueError: cannot reshape array of size 18 into shape (1,128,1,1)

[Esaada]

You did: N = offset.size(1) // 2, In that level the offset.size(1) is equal to 128 , and N suppuse to be the kernel size *2, means N=|R|

And I haven't change anything in the code...

Thanks.

[Simpatech-app]

I have the same issue, please help. Beside it is not clear that why out_channel is 18. Can you
explain briefly?
self.conv_offset = nn.Conv2d(in_channels=inC, out_channels=18, kernel_size=3, padding=padding, bias=None)

[bkvie]

Size mismatch because of the linear layer... do a reshape or adaptive pooling or leave the linear layer out.
have a look at this implementation: https://github.com/bkvie/Locally-Consistent-Deformable-Convolution

[H-Wenfeng]

I have the same problem……