Does nn.PixelShuffle support 3D data? In other words, the input is 5D.
Based on the docs a 4-dimensional input tensor is expected and you should also get a
RuntimeError, if your input tensor has a different number of dimensions.
is there any solution to 3d data in nn.PixelShuffle or will be in the future?
This PR might have already added the support.
You could use the latest nightly binary or build from source and check, if it’s working.
The latest nightly binary solved the problem of nn.PixelShuffle for input with more than 4 dimensions.
Is that supported for 3D data now?
The MR was merged in Dec 2020 so I think it should be enabled. Did you try it and if so did you seeany issues?