Transforms tutorial fails if GPU available

[adamjstewart]

Description

The transforms tutorial crashes if you run it on a system like Google Colab where a GPU is present.

Steps to reproduce

Run the tutorial on Google Colab. The following cell:

%%timeit -n 1 -r 1
_ = transforms_gpu(batch_gpu)

fails with the following error message:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
[<ipython-input-15-75fa7a521613>](https://localhost:8080/#) in <module>
----> 1 get_ipython().run_cell_magic('timeit', '-n 1 -r 1', '_ = transforms_gpu(batch_gpu)\n')

8 frames
<decorator-gen-53> in timeit(self, line, cell, local_ns)

<magic-timeit> in inner(_it, _timer)

[<ipython-input-3-61c1063eeb18>](https://localhost:8080/#) in forward(self, inputs)
     13         # Batch
     14         if x.ndim == 4:
---> 15             x = (x - self.min[None, ...]) / self.denominator[None, ...]
     16         # Sample
     17         else:

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu!

Version

releases/v0.3

[calebrob6]

FWIW this was caused by the fix from a few days ago. https://github.com/microsoft/torchgeo/pull/756/files#diff-30f150fdd744014d8396edf56eec96f7285e9f26375f39cc182b1dc4d07bc178

Previously, because everything was incorrectly on the CPU anyway, we didn't see this conflict.