Regression in SDnetDecoderBlock

[glgerard]

In the forward method of SDnetDecoderBlock, the call to self.unpool can fail if out_block is None.
Code should be changed as such

    if out_block is not None:
        unpool = self.unpool(input, indices, out_block.shape)
        concat = torch.cat((out_block, unpool), dim=1)
    else:
        unpool = self.unpool(input, indices)
        concat = unpool

[viethoang303]

I also got this problem and I can' t run the code