Overload functions

Opening this to discussion about having a kornia.overload to support api duality

Based on @shijianjian proposal:

I thinking of something like:

from typing import overload

class A():

    @overload
    def generate_parameters(self, batchsize: Tuple[...]) -> Dict:
        ...

    @overload
    def generate_parameters(self, input: Tensor, mask: Tensor) -> Dict:
        ...

Not sure how would PyTorch like this grammar.

_Originally posted by @shijianjian in https://github.com/kornia/kornia/issues/2050#issuecomment-1336374972_

[edgarriba]

I would love to see something like below (or a variant) to support research/production use cases:

• Tensor in Tensor out (TITO) for research flexibility
• Safe typed high level apis semantics for production

@kornia.overload
def grayscale_from_rgb(image: Tensor, weights: Tensor) -> Tensor:
   return K.color.rgb_to_grayscal(image, weights)

@kornia.overload
def grayscale_from_rgb(image: Image, weights: Tensor) -> Image:
   return Image(K.color.rgb_to_grayscal(image.data, weights))

/cc discussed at some point with @ducha-aiki

[edgarriba]

@johnnv1 @shijianjian i found this interesting article https://martinheinz.dev/blog/50