search

Roman-Supernova-PIT / phrosty

Basic package for photometry on the RomanDESC sims.
MIT License
4 stars 2 forks source link

Generate 4-5x oversampled PSFs #30

Open laldoroty opened 1 month ago

laldoroty commented 1 month ago

image

Left is original PSF, right is PSF after decorrelation. Mapping is log scale, so it looks worse than it is.

(original notes from @rknop)

laldoroty commented 1 month ago

This can be done in galsim easily--https://github.com/matroxel/roman_imsim/blob/v2.0/roman_imsim/utils.py#L88

it's just a kwarg in the function we use.

wmwv commented 1 month ago

When you say "it looks worse", what's the quantitative thing you're trying to minimize?

I think it's unlikely we will actually know the PSF that much better than a few times smaller than a pixel. You shouldn't expect that a decorrelated image will look like an image. When you're operating near the sampling limit, there's just stuff you can't recover.

laldoroty commented 1 month ago

@thomasvrussell @rknop

Even 2x would help, we think!

wmwv commented 1 month ago

What's the metric for improvement or success?

wmwv commented 1 month ago

On the phrosty side, the following branch now accepts additional kwargs for get_imsim_psf that will be passed on to roman_imsim getPSF_Image, which then passes that on to drawImage.

Next is to [ ] 1. Add a test to verify this behavior [ ] 2. Add change, configuration to diff-img preprocess.py to pass something like oversampling_factor=4 to get_imsim_psf.

Do we always want it to be something constant? Should it be a default in diff-img to be 4, should it be specified on the command line? Do we want different oversamplings for different purposes within diff-img?