Output PSF model at input star positions
Opened this issue · 3 comments
Feature request: For further processing, we need postage stamps of the PSF model at input star positions filtered with the input star SEDs.
These outputs can then serve to perform further diagnostics, e.g. rho statistics.
The code is already capable of doing this, as it is done when the metrics are computed. To implement this issue we just need to save the produced PSFs by the model in a given format.
The question is in which format would you like the output to be saved? At which resolution, at super-resolution? (If you are going to measure shapes, as I expect if you will compute rho statistics, it should be at super-resolution)
@martinkilbinger for the metrics we are already computing the shapes using GalSim's HSM module. It could be something interesting to add in the output file in addition to the pixel PSFs. What do you think?
@jeipollack for this we could reuse something as it is done in the metric computation, but instead of computing the metrics with respect to a ground truth, just save the PSFs in a given format for postprocessing.
Hi @tobias-liaudat, yes, with Alex we figured that during or after the metric computation, we can simplly output the computed PSFs. Ideally, it would be the same format as the input star catalogue.
If the shapes are already computed during metrics, I agree it could be interesting to save ellipticity and size. Same for the stars.
Is the simulated input catalogue already super-resolved?
@martinkilbinger Ok, great. It should not be too hard and we'll need to do that at some point.
In the original version of wavediff, the simulated input catalogue can or cannot be super-resolved. If it does not contains the super resolved stars, they will be regenerated to compute the metrics. If super resolved stars are in the dataset, they will be used directly. We should confirm that this is still the case in the refactored version of the code.