feat: validate the minimum number of expected images in `validate_list_of_dataset2d` (Useful for `DriftCorrection`)

[bobleesj]

Continuing our discussion from #75 with @gvarnavi

Problem

validate_list_of_dataset2d is used for setters. At the moment, it doesn't support the following behavior: "I want to ensure there are at least two images in the list of dataset2d (in the case of DriftCorrection).

Proposed solution

Add an optional img_count_min parameter:

def validate_list_of_dataset2d(
    images: List[Dataset2d] | List[NDArray] | Dataset3d | NDArray,
    img_count_min: int = None,
) -> List[Dataset2d]:
    """
    Validate that the passes images is a list of Dataset2d objects.

    Parameters
    ----------
    images : List[Dataset2d] | List[NDArray] | Dataset3d | NDArray
        The images list to validate
    img_count_min : int, optional
        The minimum number of images expected to be present in `images`.

In practice, it can be used as the following?

def test_validate_list_of_dataset2d_with_wrong_min_image_count_specified():
    img = np.random.random((10, 10))
    with pytest.raises(ValueError, match="The 2D list requires a minimum of 2 images but only 1 has been provided."):
        validate_list_of_dataset2d([img], image_count_min=2)

@gvarnavi

[cophus]

@bobleesj I think we want to allow single images for DriftCorrection. The reason is the following workflow (not yet implemented):

1 - import 1 atomic resolution image
2 - find the unit cell lattice coordinates / indexing (several possible methods)
3 - create a "mean unit cell"
4 - tile out the mean unit cell to fill a a "reference image."
5 - drift correct the single input image against the reference image.

This way we can apply prior knowledge in some cases - i.e. that atomic res images of crystals should (on average) be distortion free. This is also potentially useful for "lattice unbending," e.g. https://doi.org/10.1016/j.jsb.2006.07.012

Feel free to take a crack at implementing this - at some point I will make a project entry for it.

[bobleesj]

@cophus Sounds fun. I will ask a few follow-ups via slack.

[gvarnavi]

@cophus Still, we shouldn't be shipping broken code.

What you're describing is a separate classmethod, e.g. from_single_image(), which does all the preprocessing you described to make a reference_image and then calls cls.from_images().