Changes between Version 6 and Version 7 of IPP_Progress_Report_20100719

Timestamp:

Jul 27, 2010, 6:32:21 PM (16 years ago)

Author:

eugene

Comment:

--

IPP_Progress_Report_20100719

@@ -5 +5 @@
 
 === Eugene Magnier ===
+
+I was out on Jury duty for most of 2 days this week.  I made some progress on two fronts: relastro and extended source analysis.  Bill discovered a problem with relastro in how it monitors the fit quality and how it handles poor fits.  It attempts to watch for any fitted source position which deviates by more than a small allowed amount -- evidence that the fit for the chip was allowed to run-away in some region -- and then remove the offending image.  It turned out that this test and fix resulted in the database being re-loaded many times for each repaired image.  Not very efficient.  I re-worked the test to use a grid of test points mapping the area of the image where real sources exist.  This allows the quality check to be done before the data is written in the first pass, so that failed images can be immediately reverted to their original fit.  In testing this analysis, I discovered a bug with the conversion between the internal representation of the astrometric solution (as a standard polynomial) and the WCS representation: the reference pixel is normally set to the telescope boresite, but the non-linear terms may sometimes be large enough that the solution of this position is does not easily converge.  I modified the assumptions in the code to allow this conversion to WCS terms to reset the reference pixel to a pixel on the image.
+
+For the extended source analysis work, I have finished modifications to the psf-convolved fitting process to allow a grid search of the sersic index parameter.  This required some modest re-working of the APIs to allow some additional flexibility.  I have also measured the relationship between the measured moments and the guess scale length as a function of the sersic index, making the initial guess more accurate.  At this point, I'm happy with the results of this analysis, but not happy with the speed.  The PSF-convolved fitting is too slow, so I intend to make some fairly easy optimizations (convert the convolutions to FFTs and carry the FFT of the psf image during the fit).
 
 === Serge Chastel ===