Live stacking performance

[lowenthalm]

I have noticed a slow down in live stacking performance and am trying to track down the source. I have used the same camera (a ZWO ASI294 Pro Cooled Color camera) throughout, but other software changes along the way muddy what might be the cause. I am using Parallels on a Mac the entire time, but have observed decent in the past. Parallels has 8gb of memory, 4 CPU cores and up to 4gb of graphics memory allocated to windows.

Initial configuration was SharpCap 3.2 and Sharpcap 4.0 under Windows 7. Live stacking performance was typically less than 2 seconds per captured frame with a render time of about 0.3 to 0.5 second to 1/2, if I recall correctly.

I upgraded to Windows 10, and recently have switched to SharpCap 4.1. I also am running PHD2 guiding now, which grabs some CPU, but not too much. I always use sigma clipping. I observe in the Status tab that stacking time is now on the order of 7 seconds prior to passing the sigma clipping initial frame count of 5 and doubling to a stacking time of 14 to 15 seconds per frame thereafter. Sometimes the stacking time goes up to as high as 21 seconds. It was even worse before I turned "optimize for faint stars turned" off (I saw this recommendation in one of your older forum posts). Render time still seems to be quite fast, so I have focused on what is causing stacking time to be excessive.

I may want to try going back to SharpCap 3.2 to try to sort out what might be causing my issues, but have been trying other things first as I don't want to break my setup.

Is it typical that once the initial sigma clipping frame count has been passed during live stacking that stacking time should double? I have checked the task manager and see that SharpCap does appear to be using the GPU. I have more testing planned the next few nights to see if I can see what might be slowing things down. I do see the CPU either close to or at 100% during live stacking.

Any thoughts you might have on what I should check would be helpful.

[admin]

Hi,

yep, can give you some more pointers of things to check - no problem...

The first thing to do is to turn on the additional logging options for 'Live Stacking' and 'Image Processing' in the settings - that will add some more info to the log giving timings on various parts of the process that you can't separate out in the UI. Also, use the folder monitor camera with some previously saved subs to enable quick, reproducible testing.

So, it sounds like star detection is at least part of the slowness, since turning off the 'optimize for faint stars' helped. Star detection can be quite a variable thing, depending on how many candidates are found in the image. Turning down the sensitivity or tunring up the 'Noise Reduction' may both help by reducing the number of candidates that are considered during star detection. If you have 20 to 30 stars being found consistently then that should be fine for live stack alignment. It's worth noting that every so often, SharpCap will re-detect stars in the stack as well as the stars in the new frame - that could add even more time if the star detection is slow.

Switching to default stacking as a test is a good way to work out how much time the sigma clipping is adding to the processing time (default stacking is quicker). Sigma clipping will definitely be slower after the initial frames, since it then needs to start testing the new values against the sigma interval for each pixel to see if the new value should be used or not, so the observation that it slows after the initial frames could well point the finger at sigma clipping.

Do let me know what you find out - there is often scope to optimize parts of the code to make them go faster (but if I optimize everything then I would never get anything else done, so it has to be targeted).

thanks,

Robin

[lowenthalm]

I've been working on this on and off. Its highly variable from star field to star field. Also depends on if I am using just an UV/IR Cut filter with my OSC or my dual-band Ha/OIII-Hb filter (usually worse as I have to turn on optimize for faint stars). If I notice a clearer pattern I'll post it here.

Saw that you did some sigma clipping optimization in a recent new version feature. Thanks for this. Some preliminary testing last night seemed so show the difference before and after sigma-clipping kicks in after a few frames is now only about a 50% increase in stacking time rather than a 2x to 3x increase in stacking time when I last tested it on a SharpCap a couple versions back.

[admin]

Hi,

thanks for the update and good to hear that the sigma clipping improvement helped.

If it is varying from one field to another then I would still suggest looking at the star detection, as it does more work if more potential stars are found in the image. Turning up the 'Noise Reduction' setting for star detection can be quite good at speeding this up if you can still get enough stars for alignment after adjusting the setting. Basically the noise reduction blurs a copy of the image slightly before the star detection happens - the blur helps stop faint stars and pixel noise being considered as possible candidate stars during the detection process.

cheers,

Robin