torch default for Flip

[rijobro]

Description

Starts to address: #2231.

To base class convert to/from numpy as necessary, and then convert back to the original at the end. Use torch for the main functionality.

This is an example case for the rest of the transforms.

I also created a class to contain types frequently used in our transforms:

class TransformTypes:
    """Common transform types."""
    Images = Union[torch.Tensor, np.ndarray]
    ImageDict = Dict[Hashable, Images]

By using these, we should be better able to unify our transforms (some only have np.ndarray, some use torch.Tensor, some have both).

Status

Ready

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[ericspod]

We can put this conversion block into a separate utility function somewhere:

def convert_data_type(data, current_type, required_type):
    if current_type is np.ndarray and required_type is torch.Tensor:
        return torch.Tensor(data)
    elif current_type is torch.Tensor and required_type is np.ndarray:
        return data.detach().cpu().numpy()

    return data

[rijobro]

good point, done.

[wyli]

Addressed by #2822