Repository navigation
Add a new unit equivalency to handle pixel scale specified with length unit. #10110
Description
I came across this situation that I need to build a transformation to describe the projection of a detector array specified in u.um to some fiducial plane which is used in lab testing the array.
We construct maps from such tests and would like to build a GWCS object as the WCS of these maps.
In this case, the WCS is an AffineTransform2D that takes inputs unit 'u.um' to 'u.pix', which requires pixel_scales specified as u.m / u.pix as the input_units_equivalencies.
I was wondering if this makes sense and I am glad to contribute a initial pull request. Mostly it gonna be a duplication of u.pixel_scale (say, u.pixel_scale_phys), but just replace u.deg there with u.m. Or it makes more sense to extend the current pixel_scale equivalency to cover this situation.
Activity
pllim commentedon Apr 3, 2020
Personally, I think it makes more sense to extend either
pixel_scaleorplate_scale. @Cadair or @nden , what do you think? You must have come across this conversion already in your pipelines? cc @mhvkhttps://astropy.readthedocs.io/en/latest/units/equivalencies.html#pixel-and-plate-scale-equivalencies
mhvk commentedon Apr 3, 2020
I'd be tempted to just generalize
pixel_scaleto be anything whatsoever per pixel.Though I also wonder why it isn't possible to give one's own equivalency in the first place - but I'm not sure what exactly is being initialized here!
pllim commentedon Apr 3, 2020
pixel_scaleandplate_scalewere added by @eteq in #4987 .Jerry-Ma commentedon Apr 3, 2020
Using one's own equivalency does not permit serialize the model to ASDF file, because it is not installed into
.../astropy/io/misc/asdf/tags/unit/equivalency.py.It seems that to make the serializer work for arbitrary user defined
Equivalenceinstances, we probably could leverage the models?Say, we can allow defining equivalencies via a tuple of
(u_from, u_to, model_unitless), e.g, a pixel scale of 2. * u.m / u.pix can be defined asCustomEquivalency([(u.pix, u.m, models.Scale(2.))], 'my_pixel_scale').Cadair commentedon Apr 4, 2020
I don't see why we need to have
pixel_scalebe limited to angular units personally.Another work around @Jerry-Ma (which I have confused @nden with in the past) is you can use the
Multiplymodel to do the scaling before the affine transform. See https://github.com/DKISTDC/dkist/blob/master/dkist/tests/generate_eit_test_dataset.py#L51 for what I mean 😄Jerry-Ma commentedon Apr 6, 2020
I just tried models.Multiply, and it works well. Thank you!
On the pixel_scale, I think generalize the current pixel_scale is a good idea.
pllim commentedon Apr 6, 2020
@Jerry-Ma , do you have interest to work on it? 😉
Jerry-Ma commentedon Apr 7, 2020
@pllim I've put together a pull request to try addressing the issue.