[AstroPy] Gaussian models fitter non-functional?

Fri Oct 23 12:00:14 EDT 2015

Hi John,

You might also be interest in IQMon, a package Josh Walawender put together for assessing image quality after-the-fact. It is at https://github.com/joshwalawender/IQMon

Matt Craig

schedule:  http://physics.mnstate.edu/craig
——

Professor
Department of Physics and Astronomy
Minnesota State University Moorhead
1104 7th Ave S, Moorhead MN 56563

phone: (218) 477-2439
fax: (218) 477-2290

On Oct 23, 2015, at 10:40 AM, John Wainwright <johnw at lyric.com<mailto:johnw at lyric.com>> wrote:

Thanks Larry.  Sorry I’m coming up to speed on this somewhat haltingly.  I’m an amateur astrophotographer (some of my images here<https://www.flickr.com/photos/jbw/albums/72157631316034648>), as well as a software engineer, and I’m working on an image-cataloging program that, among other things monitors exposures as they are captured and reports things like FHWM, estimated SNR, any correlated PSF elongation, etc., to help tune a capture session.  Astropy has been an absolute boon for this effort.

From what I can see the 2D-Gaussian fit in find_peaks() is in the dev release, which I’ll upgrade to shortly, but it doesn’t seem to report mean or std-devs or orientation of the fitted model, which I think I need for some of the above.

So, the trick would be to use find_peaks() to get the centroid and then create cut-outs around them that I give to the fitting & modeling functions in astropy, is that right?   Or simply use detect_sources() + segment_properties() to get the cut-outs and covariance matrix from their image-moment analysis to seed the 2D Moffat fit, or whichever model I might use.  How different would you expect say a Gaussian 2D model’s centroids & orientation & standard deviations be from the segment_properties() figures on the same source?

I do hope I’m not overstepping things here as an amateur bugging you guys for help.

Best,
John.

On Oct 22, 2015, at 9:19 AM, Larry Bradley <lbradley at stsci.edu<mailto:lbradley at stsci.edu>> wrote:

Hi John,

Just to clarify, the morphological properties calculated by photutils.segment_properties() are derived from the 2D image moments of the source (within the defined segment) -- 2D Gaussian fits are not used there.  Photutils also has a local peak finder (photutils.find_peaks()), which does have an option to use a 2D Gaussian fit for better precision.

Cheers,
Larry

On Thu, Oct 22, 2015 at 12:08 PM, John Wainwright <johnw at lyric.com<mailto:johnw at lyric.com>> wrote:
Thanks Erik, and sorry, I should have found those.  I realize now I was asking relative to the detect_sources() and segment_properties() functions in the photutils<http://photutils.readthedocs.org/en/latest/index.html> affiliate package, which I think use a 2D Gaussian fit to generate the basic morphological properties I've been getting.   Presumably, I could use the centroids from that data as peak seeds for a Moffat2D model fit.

Thanks again,
John.

On Oct 20, 2015, at 7:01 AM, Erik Bray <embray at stsci.edu<mailto:embray at stsci.edu>> wrote:

On 10/19/2015 05:56 PM, John Wainwright wrote:
Just on the topic of PSF fitting, is there any plan to implement Moffat PSF
models?  I'm trying to replicate some of the PSF fitting
<http://pixinsight.com/doc/tools/DynamicPSF/DynamicPSF.html> capabilities found
in PixInsight <http://pixinsight.com/> which provides excellent Moffat modeling
support.

Well, there are are already these implementations:

http://docs.astropy.org/en/latest/api/astropy.modeling.functional_models.Moffat1D.html#astropy.modeling.functional_models.Moffat1D

http://docs.astropy.org/en/latest/api/astropy.modeling.functional_models.Moffat2D.html#astropy.modeling.functional_models.Moffat2D

So you can check how they compare.  You can also always add your own version:

http://docs.astropy.org/en/latest/modeling/new.html

On Oct 19, 2015, at 11:45 AM, Jeff Mangum <jmangum at nrao.edu<mailto:jmangum at nrao.edu>
<mailto:jmangum at nrao.edu<mailto:jmangum at nrao.edu>>> wrote:

Thanks for the responses.  Adam Ginsburg and I iterated a bit on this problem
and found that the issue was the fact that the image I was fitting contained
nans.  Once I removed the nans the gaussian fit routine works.  Note that if
the fitter in astropy.models is not able to handle images with nans, then
perhaps it should issue a warning.

-- Jeff

On 10/19/15 11:00 AM, Erik Bray wrote:
On 10/16/2015 09:47 AM, Giuliano Iorio wrote:

Hi Jeff,
this is the kind of staff that usually happens if the position of the
initial guess is too far away the right one. Did you plot the contour of the
fitted gaussian on the image?
It can be also that the image is too noisy around the target to fit. I
usually use only a little subarray around the target, then I search the
position of the maximum and I set the initial guess of xo and yo to its
position. After I normalize all the subarray to the value of the maximum and
cut all the signal below a certain threshold (usually 0.1).
In this way I am able to obtain a robust fit also in very noisy map.

I agree pretty much with all of Giuliano's advice.  And again it might help
to exclude data outside a smaller subarray.  Without actually seeing the data
it's hard to provide much other advice.

Have you tried a different optimizer?  There are a couple other non-linear
fitters in the package and some have different sensitivities to initial
conditions.  But it's certainly not "non-functional".

Erik

Hi Jeff--

The fitted model will not be in g_init (this remains unchanged), but in g.
 If you take a look in g, you should be able to see the fitted results.

Cheers
Steve

----- Original Message -----
From: "Jeff Mangum" <jmangum at nrao.edu<mailto:jmangum at nrao.edu> <mailto:jmangum at nrao.edu<mailto:jmangum at nrao.edu>>>
To: astropy at scipy.org<mailto:astropy at scipy.org> <mailto:astropy at scipy.org<mailto:astropy at scipy.org>>
Sent: Friday, October 16, 2015 3:11:56 PM
Subject: [AstroPy] Gaussian models fitter non-functional?

Hello,

Long-time listener but first-time caller.  Have been trying to use the
2D Gaussian fitting capabilities within astropy.models to fit a 2D
Gaussian to an image (in FITS format).  Here is my script:

import numpy as np
import radio_beam
from astropy.modeling import models, fitting
from astropy import units as u
from astropy.io<http://astropy.io/> import fits
import astropy.coordinates as coord
import astropy.wcs as wcs
import matplotlib
import pylab as pl

# The following is the position of the peak in both J1 and J2, so use
# it as the starting guess
ra0 = '13h15m03.50s'
dec0 = '24d37m08.2s'

radeg = coord.Angle(ra0, unit=u.hour).degree
decdeg = coord.Angle(dec0, unit=u.degree)

# Now open the image
hdulist = fits.open('IC860CbandCarrayH2COJ1_moment0.fits')
#hdulist = fits.open('IC860KubandDarrayH2COJ2_moment0.fits')
w = wcs.WCS(hdulist[0].header, hdulist)

yy,xx = np.indices(hdulist[0].data.shape)

x0,y0 = w.wcs_world2pix(radeg, decdeg, 1)

# Now fit 2D gaussian
g_init = models.Gaussian2D(amplitude=1., x_mean=x0, y_mean=y0,
x_stddev=1, y_stddev=1)
fit_g = fitting.LevMarLSQFitter()
g = fit_g(g_init, xx, yy, hdulist[0].data)

When I look at g_init (which is where I believe the fit results should
be), I see that the fit has not moved from its initial settings.  It
appears that the fitter just did not run.  With Adam Ginsburg's help I
have a workaround using gaussfitter, but would certainly like to make
this work with astropy.models.  Thanks!

-- Jeff

_______________________________________________
AstroPy mailing list
AstroPy at scipy.org<mailto:AstroPy at scipy.org>
https://mail.scipy.org/mailman/listinfo/astropy

_______________________________________________
AstroPy mailing list
AstroPy at scipy.org<mailto:AstroPy at scipy.org>
https://mail.scipy.org/mailman/listinfo/astropy

_______________________________________________
AstroPy mailing list
AstroPy at scipy.org<mailto:AstroPy at scipy.org>
https://mail.scipy.org/mailman/listinfo/astropy

_______________________________________________
AstroPy mailing list
AstroPy at scipy.org<mailto:AstroPy at scipy.org>
https://mail.scipy.org/mailman/listinfo/astropy

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/astropy/attachments/20151023/ca4c9a73/attachment.html>