Is ToDeviced transform needed for Cache I/O and GPU transform?

[hw-ju]

Hi!
In the section "4. Cache IO and transforms data to GPU" in fast training guide, it seems ToDeviced transform is needed (after EnsureTyped) to move data to GPU. However, in the fast_training_tutorial.ipynb, ToDeviced is not used, and only EnsureTyped is used to "convert the data to Tensor without meta, move to GPU and cache to avoid CPU -> GPU sync in every epoch" according to the comment.

So is ToDeviced still needed after EnsureTyped? Or just EnsureTyped is enough?

[wyli]

sorry for the confusion, ToDeviced is not needed, EnsureTyped should be enough for this use case. I'll update the the descriptions.

[hw-ju]

@wyli Thanks a lot for the clarification!

[hw-ju]

@wyli Could you also update the description below in the Define the training progress section in fast_training_tutorial.ipynb?

ToDeviced transform: to move data to GPU and cache with CacheDataset, then execute random transforms on GPU directly, avoid CPU -> GPU sync in every epoch. Please note that not all the MONAI transforms support GPU operation so far, still working in progress.

[KumoLiu]

Hi @hw-ju, have updated it in this PR. Thanks!