StarAlignment hangs indefinitely if I change Lights processing method

LarryC

Well-known member
Sep 29, 2015
62
0
I've been playing around and comparing processing approaches with darkflats vs. biases in various and sometimes non-traditional combinations for both flat and light frame processing. If, after calibrating lights with a darkflat processed Master Flat, and a "mormal" Master Dark, and then cosmetic correcting with the same Master Dark, StarAlignment works just fine - it takes about 1.5 minutes to register the 24 frames I'm using as test data (M13 Oiii). However, if I try to add a bias frame (either a "0" second Master Bias or a 0.1 sec master dark flat) to the Light frame calibration step, and then cosmetically correct, which all goes well, StarAlignment will load all images and align two of them in about 3-5 minutes, but then nothing happens for at least another 15 minutes. I have to pause/abort the process, which only results in PI becoming "unresponsive", and then I have to force a PI termination to get out. This only happens when I use a dark flat-subtracted Master Flat, a non-calibrated Master Dark and either a "normal" Master Bias or a 0.1sec dark-flats (as a bia frame) during light frame calibration.

The PC is a 12 core AMD running at 4.2mHz, 32Gb ram and a 1Tb NVME drive. I get a benchmark of about 19,500. When the above idleness is occurring the CPU is at 8-15% and the machine is using about 7 GB of RAM. I've tried using a different SSD for all data and processed files in case there is a problem with my NVME, but that resulted in no difference. I also uninstalled and reinstalled PI +/- all updates since Feb. with no effect.

When testing a slightly different processing approach, but still involving a bias or darkflat at two steps, Integration (after successfully registering the frames) would fail on the first frame and report the reference image (regardless of what frame I used as reference) as something like "very low signal or empty file?". All image files showed an image with a STF, but I assumed they way I tried adding the bias frame to to the processing path removed too much signal? In any case, I suspect the problem here is similar -somehow there is too little signal remaining in the frame after my experimental processing approaches for SA to find stars. That's all fine, but the "bug" to me is that after 15 minutes hung up on registering a single frame, I would think StarAlignment would time out. The last entry in the processing console is reporting that it is/was saving the drizzle file from the second registered frame. It does not indicate that a third frame was loaded. I've had plenty of data sets where SA could not find a star match, but that never slowed down the overall process - it just reports the failure of that particular frame and moves on. With the processor at 15% on this current set it seems like SA is just giving up or bewildered and not doing anything.
 
Last edited:

LarryC

Well-known member
Sep 29, 2015
62
0
I think I resolved the previous issue as a problem with Star Alignment with frames having low signal. I've now seen this problem where Star Alignment goes from fairly fast to agonizingly slow, or simply stops making progress, anytime the data set has a mean ADU < ~1e-1. What made the problem hard to nail down the first time was forgetting that STF, when used its own, is applied uniquely to each frame so that two frames processed very differently look, after STF, overall similar in signal. Only when the STF from one image is applied the second is it visually clear that one or the other frames might have a much lower signal- in the case above by over processing. Of course, Image Statistics also makes the signal differences clear.

My current case is from my first set of RGB data with new filters, after years of imaging only with Nb. 60 second exposures of M22 in RGB resulted in very similar and nice looking images after an SFT was applied separately to representative frames of each filter. However, whereas registration of "R" frames went quickly, "B" registration went slowly, but did complete, and "G'" registration ground to a halt with an unresponsive PI. The mean ADU for three representative R:G:B frames were 2e-1, 8e-2 and 6e-2.

An easy fix for low signal is longer exposures, but the problem I reported originally as a "bug" is that Star Alignment/PI, IMO, should either time out or notify the user that there is a problem with the frames instead of just becoming essentially unresponsive ("Pause/Abort" does nothing), while using only 6% of the CPU and 5Gb ram - it won't give up but it's not even trying!
 
Last edited:

Juan Conejero

PTeam Member
Sep 2, 2004
7,555
96
56
Valencia, Spain
pixinsight.com
As noted above, this problem cannot be reproduced. For example, I have just tried StarAlignment with two images where the median pixel value is about 7e-4 and it has worked perfectly. We have much weaker images in our test data sets and there are no problems either.

Please upload a reduced data set where this issue can be reproduced, or your report will be useless to solve the problem.
 

LarryC

Well-known member
Sep 29, 2015
62
0
Juan,

Would 15 images through a dropbox link work for you?
 
Last edited:

LarryC

Well-known member
Sep 29, 2015
62
0
Juan, Here's a link to a dropbox folder with a set of 15 frames each of what were, in this case, Red and Green filter sets. In each set of frames is indicated the frame I used as a reference (selected through SFS).

Slow StarAlignment Test Images

I've also included a StarAlignment process icon with the settings I used.

I tested these sets this morning and while both sets did go to completion, whereas it took 2'30" for the Red frames, it took 6'10" for the green frames.

The full sets contain 70 frames each and when I load all 70 of the Green set into SA, I get 2-4 registered frames in less than 5 minutes but then SA "appears" to do very little - CPU is ~6% and about 5Gb of 32Gb ram is being used. I've let it run for about 10 minutes with no new frames registered (maybe 7-10 minutes). I don't know if it would eventually get to completion, if I let it. I didn't time the complete set of Red frames as it proceeded and completed at what would consider a "normal" rate - maybe 5 minutes overall?

Windows 10, all drivers updated, latest PI release, AMD Ryzen 3900x, OC'd to 4.2. I normally use an NVME M.2 1Tb drive for processing, but the test today was on a SSD. Same results on either drive. Swap files are on the 1 tb NVME drive. Both drives are around or less than 50% full.

Larry