Convert all index transforms to Kornia

[adamjstewart]

This PR converts all of our index transforms to be valid Kornia augmentations. These transforms can now be used with AugmentationSequential just like all of the other transforms we use.

[adamjstewart]

Can't get the dimensions to work in the transforms tutorial, please help.

[nilsleh]

Can't get the dimensions to work in the transforms tutorial, please help.

The only thing I can get to work is to match the size dimension of the input.

class MinMaxNormalize(K.IntensityAugmentationBase2D):
    """Normalize channels to the range [0, 1] using min/max values."""

    def __init__(self, mins: Tensor, maxs: Tensor) -> None:
        super().__init__(p=1)
        self.flags = {
            "mins": mins.view(1, mins.shape[0], 1, 1), 
            "maxs": maxs.view(1, maxs.shape[0], 1, 1)
        }

    def apply_transform(
        self,
        input: Tensor,
        params: Dict[str, Tensor],
        flags: Dict[str, int],
        transform: Optional[Tensor] = None,
) -> Tensor:
        return (input - flags["mins"]) / (flags["maxs"] - flags["mins"] + 1e-10)

[adamjstewart]

Well, all the dims work now, but the plotted image is all black. Not sure what changed to cause that.

[calebrob6]

Heads up that running transforms.ipynb breaks because the SSL cert for the EuroSat dataset expired. I'm wondering whether we want to add a UserWarning or something about this?