Inverse transformations and test time augmentation

[rijobro]

2D segmentation with decathlon brain tumour dataset. Do inverse transformations of single images, and of batches of data.

@wyli @tvercaut Can you check the bottom cell -- does the test time augmentation API look roughly how you expect?

tt_aug = TestTimeAugmentation(
    val_transforms,
    batch_size=5,
    num_workers=0,
    inferrer_fn=partial(infer_seg, model=model),  # fn to infer segmentation
    device=device
)
mode_tta, mean_tta, std_tta, vvc_tta = tt_aug(file, num_examples=10)

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[wyli]

thanks, it looks really nice! I also got a feature request for multi-gpu parallel inference for TTA and ensembling. probably out of this scope but any rough idea here? cc @Nic-Ma

[wyli]

just need to update the first line to use monai-weekly, then it works nicely with colab