Error on extreme crowded fields

[Hevinxue]

Hi! I was wondering if AstroPhot could be used for fitting images that contain random and crowded fluorescence splats (for the purpose of photometry of course) . I tried the segmap_models_fit.py with only source extraction parameter tuned, and it came out that more than 900 sources extracted yet the initialization of model group crashed every time. The reason of crash seemed was contributed to too small area of some sub-models, do anyone have any ideas on that? Thanks!

[Hevinxue]

Also, is it possible for us to use Astrophot(or any other tools) to fit a densely-packed brightness field withoud explicit models designated prior to fitting? :)

[ConnorStoneAstro]

Hi @Hevinxue , Ah yes for very small objects of just a few pixels it can be hard for AstroPhot's automatic parameter initialization to get started. You can of course manually initialize those parameters, but it is more work. I am constantly tinkering with the automatic initialization so I'll try in the next version to make it more robust to very small sources (you're not the first person to request this).

The segmap_models_fit.py script is really meant as a starting point. For simple images it can likely handle everything, but for a crowded field or other challenging situations you would need to make some manual adjustments. You can for example filter out the smallest objects and not fit them, there are helpful hints in the group models tutorial.

AstroPhot only works with model objects when representing an image, it doesn't do other techniques like aperture photometry. Also you need to know where all your objects are since AstroPhot can't find sources either, that is a whole separate and very challenging problem. I would suggest looking into SourceExctractor which is able to identify individual objects in an image on its own and measure basic properties such as aperture photometry.

Let me know if you need any more help!

[Hevinxue]

Thanks for your kind and timely reply :). Actually the purpose of my photometry under current situation is to measure the ratio quantity distribution between two registrated channels so I just switched from aperture photometry to avoid error caused by minor registration misalignment.
Auto initiation seems to be a potential solution for further automation under complex situations, but I'm thinking if there's any theory for the formal representation for image data.(like compressive sensing,etc.)
Also you may will be interested to check out gaussian splatting, computer vision fellas now use density-adaptive gaussian functions (and gaussian function only) to represent a complex 3D environments. Thats somewhat inspiring for me.

[ConnorStoneAstro]

Hi @Hevinxue ah indeed gaussian splatting is an interesting concept for representation of complex structures. Along similar lines, I have recently implemented a "Multi-Gaussian Expansion" model for AstroPhot: https://astrophot.readthedocs.io/en/latest/tutorials/ModelZoo.html#multi-gaussian-expansion

It is based on this paper: https://arxiv.org/abs/astro-ph/0201430
and is apparently very efficient at modelling galaxies with detailed radial structure (although all the models have an aligned center). I haven't done much testing of it, but it sounds like something you would be interested in. As a bonus, under certain conditions it can easily be deprojected into a 3D model! See the paper for more details :)