Changes between Version 10 and Version 11 of IPP_Progress_Report_20120709

Timestamp:

Jul 16, 2012, 10:55:38 AM (14 years ago)

Author:

bills

Comment:

--

IPP_Progress_Report_20120709

@@ -41 +41 @@
  * Continued to work on psphot & psphotStack for stack images.
    * fixed bug where RA and DEC values were not being set in the outputs
-   * Changed the code to not fail if one or more of the inputs fails the deconvolution limit test. Since this only affects the convolved stack inputs, changed the code to skip the psf matched radial aperture measurements which is the only place where the convolved images are used in psphotStack.
-   * Ran a test with some modified recipe values for a MD09 stack that generates large number of false sources around very bright stars. In the example the false detections are not seen if the recipe value +LINEAR_FIT_VARIANCE_MODE is changed from "CONSTANT" to "IMAGE_VAR". Further testing is required to determine if this is a good general solution or not.
+   * Changed the code to not fail if one or more of the inputs fails the deconvolution limit test. Since this only affects the convolved stack inputs, changed the code to skip the psf matched radial aperture measurements. This analysis is currently the only place where the convolved images are used in psphotStack.
+   * Ran a test with some modified recipe values for a MD09 stack that generates large number of false sources around very bright stars. In the example the false detections are not seen if the recipe value LINEAR_FIT_VARIANCE_MODE is changed from "CONSTANT" to "IMAGE_VAR". Further testing is required to determine if this is a good general solution or not.
    * Verified that the latest version can successfully process skycells in the galactic plane without consuming unwieldy amounts of memory. We still have a performance problem with some inputs. See below
    * The current versions psphot has some sections that take unreasonably long times to process some images in very dense fields. The problem is triggered when very large footprint structures are detected (close to the size of the input image).  These structures have very large numbers of spans and some of the algorithms cpu usage blows up. In one example this seems to be triggered when a stack has a small number of inputs so cell boundaries appear in the stacks as NAN pixels. Setting the NAN pixels in the footprint detection image to zero seems to avoid the problem by breaking up the footprints into pieces that can be processed more efficiently. More investigation is required.