Help! Image type changes mid-DP!

[Jeffrey Larsen]

Hi!

We were just getting started with AstroImageJ with some data in 16 bit unsigned int and tried to use the DP to do our bias and flat corrections.  After sorting the data the right way, we get a strange behavior.

Our bias images have means in the ~200 ADU range.
The master bias output by DP has a mean in the 33000 range (actually 200'ish + half of 65538).
The reduction is ruined because that large pedestal takes otherwise proper flats and distorts them badly.

Here is the strange part though....
1.  Start AstroImageJ fresh.  Load one of our images and examine it.  Values in the ~200 ADU range.
2.  Run DP.  Things get ruined due to the previous problem.  Processed images popped up with values in the -23000's.
3. Take that same image window that popped up, open the same image I did in step 1.  Values now appear in the 33000 range for this image.
4.  Quit and restart aij and then load the same image again.  Values back in the ~200 ADU range again.

This happens for both a Windows 64 bit and Linux 64 bit install.  Somehow there seems to be a signed/unsigned thang happening!

Help?

[Jeffrey Larsen]

Additional information.

In desperately playing around with the DP options,  I seem to have cured my problem.  Why?  Dunno because it makes no sense.  I was running with all defaults here!

In ONE run of AIJ with my pipeline I can do the following for a set of bias images, a set of flat images and a set of science images set up in a DP which asks to build the bias and flat and with no other options enabled other than save images.

1) DP->view "Show raw calibration images...." and DP->View->"Show master calibration images" ON
RESULT: get correct behavior!  Images output from pipeline are properly calibrated and disk files look well calibrated when viewed with ds9.

2) Remove masters and calibrated science images.

3) Rerun pipeline with DP->view "Show raw calibration images...." and DP->View->"Show master calibration images" OFF
RESULT:  images that pops up in AIJ window show a badly applied flat and dark and values of 33000 counts.  However if I look at the output images with ds9, they are actually full of zeros.

4) Remove masters and calibrated science image files AGAIN.

5) DP->view "Show raw calibration images...." and DP->View->"Show master calibration images" ON
RESULT: get well calibrated images again both on-screen and in disk files!

The Log file shows steps and not a lot of debug information, but one section is telling -- that the problem is in making the combined bias image:

For a BAD run, the listed flat images have some lines about normalization:
[2015-11-04T18:45:42.189]  Normalizing g07d321.112.fits with mean = -23759.2564
[2015-11-04T18:45:42.227]  Normalizing g07d321.113.fits with mean = -23429.0443
[2015-11-04T18:45:42.265]  Normalizing g07d321.114.fits with mean = -23393.1936

For a GOOD run, they appear different:
[2015-11-04T18:49:28.565]  Normalizing g07d321.112.fits with mean = 9008.7436
[2015-11-04T18:49:28.603]  Normalizing g07d321.113.fits with mean = 9338.9557
[2015-11-04T18:49:28.641]  Normalizing g07d321.114.fits with mean = 9374.8064