Upsample for 5D tensor is not implemented?

1

Does PyTorch 0.3.0 support Upsample for volumetric input (BatchSize, C, D, H, W)?
Though the docs say it support 3-D, 4-D or 5-D tensor, I got the following error:

A = torch.ones(4,5,10,10,10)
up = torch.nn.Upsample(scale_factor=2, mode=‘bilinear’)
B = up(A)
NotImplementedError: Got 5D input, but bilinear mode needs 4D input

2

Using the PyTorch 0.3.0 as well. Got the same error.
@Shunsuke_Saito Have you figured out why the error comes out here?

Thanks!

3

Just figured it out.
For 5D, we need the upsampling strategy to be ‘trilinear’ rather than the ‘bilinear’.

Good luck!

4 Likes
4

Thanks, this is the correct answer.

1 Like