Hi Jarmo,
Just checking - in the images above:
Are Top Left and Bottom Left the same i.e. both BXT non-linear
OR
Top Left BXT non-linear and Bottom Left just BXT?
I have found a dataset of M51 that more closely matches yours and I am testing also.
I have run:
BXT non-linear
BXT
AutoIntegrate Default
AutoIntegrate non-linear.
My observations are:
“3. AutoIntegrate Default” > more blue
Both non-linear images (BXT and AI) are the closest color match and seem to have better color than both linear pics (they look less washed out)
Both BXT images show better detail BUT BXT with default (“2.” above)
My pick of the 4 images is BXT with non-linear. It has best definition and color.
Yes, that is correct. Both leftmost images are the same. BXT non-linear and non-linear noise reduction.
I tried also BXT on non-linear after AI default processing and it was pretty good.
I forgot to mention that I reduced AutoSTF background from 0.250 to 0.100 (or whatever the option name was). I also use auto crop. I am not at home right now but I think I had one One click observation from Telescope Live, four images for each channel.
Hi Jarmo,
That’s fine…it doesn’t bother me at all.
You are providing a wonderful free service and I certainly don’t expect tech support.
I’m simply trying different combinations in the hope that it may uncover something useful and save you a bit of time. I certainly don’t want to take up any of your time.
I would be more than happy if you specified certain scenarios that I could run, or any other basic tasks that might make things a bit easier for you. Just let me know if I can help out and also if I’m not i.e. you need a bit of space instead of replying to posts.
Any scenarios you would like me to run that might save you some time?
Please just specify any settings and I can run them.
Also happy to test out settings for Nebula and Globular Clusters (for example).
Just let me know what data sets you are using and I will download and use the same (I’ve got plenty of credits to use and lots of data already e.g. I had all M51 sets off iTelescope already so I can use the same as you if I know the date).
Happy to post xisf results on a shared drive also if that helps with analysis.
I was thinking about the NXT problem - I reran it with the 0.10 AutoSTF background setting and it’s the same result. I’m wondering if it’s because NXT applies Artificial Intelligence to the image and that somehow messes things up? Not a big deal anyway.
My M51 data is dated as 2021-04-03T20-56-31. It has 16 files so it must be from one observation.
Many youtubers seem to use M16 as an example. If M51 is a test case for LRGB/Galaxy then M16 could be a test case for nebula/HSO data. Then some Globular Clusters would also be an interesting test for BXT too.
I am working on some noise reduction changes. Just trying to clarify noise reduction and try to understand myself what is actually run in AutoIntegrate. I have already pushed some interface changes to GitHub. I have moved some sections to a new tab.
I have installed the latest changes and can see the new Tab in AI (looks good)
I can’t find that M51 data anywhere on Telescope Live.
I have plenty of M16 data from there (91 frames in total of SHO) 2020-05-24 to 2022-06-18 all T-25 600s
Pick anything you have from Telescope Live in that range and I should have it and will use for testing.
What base AI settings do you recommend for M16 and I will get started?
Call out a few settings to try and I will run them and post results if this helps.
BTW: just checking that you are aware of the detailed BXT documentation accessible from the tool itself?
I only raise this because it’s different from STX and NXT
Sorry for not checking the M51 data more carefully. Using the same data set makes sense so we can just pick one. I suggest the SPA-1 one-click observation from 11 Mar 2021, 13:15. From the preview image it looks quite ok. For me searching for M51 finds that observation. Searching with Whirlpool Galaxy should find it too.
For M16 I did some testing with CHI-1 HSO observation from 08 Oct 2022, 5:15. But any observation is fine for me if you have started with some other one. For M16 I had to search for Eagle Nebula to find it.
I got quite different results from M16 depending on where I do noise reduction. If I did noise reduction after stretching the image colors were different. The only difference I was able to find was different AutoSTF. I did not see any real color difference in M16 between BTX vs no-BTX.
Nice improvements from BTX in your M16 images. Similar to what I have seen, although not as dramatic as in some youtube videos. I have also noticed that running BTX on the final image usually works pretty well.
Btw, the latest GutHub version has some changes on noise reduction settings.
Thanks for tipping on the documentation on the tool. I had just found it but might just as well have missed it because I typically do not check those. Are you using MaxOS? Maybe that would explain your bad results in that one image as I am using Windows.
Quite a few things that could be done, not sure what would be useful
Running BTX always on HSO image and then separating channels for actual NB palette.
Options to run BTX as in current version or separate for each channel (although not recommended but could be interesting).
BTX and removing stars in linear image, not done correctly now. But I have not seen removing stars early as a very useful option in AutoIntegrate.
If running noise reduction multiple times at different phases, each could have its own settings.
Hi Jarmo,
No problem at all.
I already have that One-Click…only thing is that it’s RGB vs LRGB…will that be a problem for testing?
M16 CHI-1 HSO 08 Oct 2022 is fine…I have that.
Yes, I am using MacOS…yes, that could explain it.
I can run Windows in a virtual environment but wondering if it might be helpful to stick with MacOS to cover 2 operating systems?
I noticed the changes in noise reduction settings, but thought these were cosmetic only…wasn’t aware that you had made other changes.
As for potential changes, a lot of that is over my head as I’m still a beginner. But I’ll provide my comments, but just take them lightly as you know a lot more than I do.
I can foresee some challenges because BLX seems to want things to happen in a different way than how autoIntegrate works by default, to allow its Artificial Intelligence to work correctly. I imagine this will make coding difficult.
But there appear to be some common themes:
do as little as possible to the linear data before BLX;
give it color data vs separate channels (RGB or SHO) because the AI is trained to recognize certain things in color;
No noise reduction prior to BLX.
It seems to indicate that enabling these things will make the biggest difference (I think?)
So your point 1: BTX on HSO image - seems important
Point 2: BXT on each channel - documentation seems to recommend against this as the AI isn’t optimised for this.
Point 3: BXT removing stars - difficult to implement…requires manual PSF value (more below 2.2.2 Stars) and probably not worth the effort?
Point 4: If running NR multiple times then different settings for each. Good idea, but I think the main one for BXT is no noise reduction prior to BXT.
It really seems to want linear data as close as possible to integration and channel combination (2.2.1 in the doc). It mentions "perhaps color calibration and background flattening, but prior to any other processing’ it goes on to say that the one exception is a “reasonable amount of simple stretching…”
Related points from the documentation:
“Performing noise reduction prior to using BlurXTerminator is not recommended” (2.2.1)
2.2.2 Stars - “If you plan to remove stars…it is best to perform this operation after applying BlurXTerminator” … star removal prior to BXT requires manual PSF settings which is probably going to be difficult to code and of little value.
2.2.3 Color - “Best results…running on RGB as opposed to each monochrome channel independently”
It goes on to say that the neural network is trained in a way that it sometimes recognizes certain things only by seeing all three channels simultaneously. Same theme for NB - do as little as possible before BXT e.g. just combine channels, don’t mix colors. For LRGB - run BXT on L and RGB separately, then combine.
Oops, I missed that it was RGB and not LRGB. How about 06 May 2021, 7:15? Or pick something else if you like.
It is a good idea if you have a Mac and I have Windows. Potentially helps to catch some OS specific issues.
Yes, noise settings are just interface changes. I did change the code but only for settings so it should be the same unless I made a bug.
I pretty much agree with your comments on potential changes. Current behavior is close to what is recommended for BTX if noise reduction is moved after BTX (using Combined image noise reduction or Non-linear noise reduction). I need to take a look at BTX on HSO image and then separate channels, it should be relatively easy to do as similar code is already there.
Sounds good - I’m now testing with M51 06 May 2021, 7:15 and M16 CHI-1 HSO 08 Oct 2022
Not sure how much time I’ll get in the next few days with Christmas upon us.
I have loaded the lastest updates - 1.56 Test 4 so I will try and run a few tests in about 4-5 hours time.
I live in Sydney Australia so it’s 6pm Xmas eve now.
In case we don’t catch up, I want to thank you for all of the hard work you do on this great tool and wish you a Merry Christmas.
Hi Jarmo,
I hope that you had a great Christmas and New Year.
I haven’t spent much time on AutoIntegrate over the break.
I applied your changes today to my Mac.
I am getting excellent results on M51 with Crop, SPCC colour calibration, SXT, NXT and BXT, autoSTF TargetBackground = 0.10
SPCC seems to make a noticeable difference - well done!!
I have noticed a small error on my Mac where the AutoIntegrate window goes almost blank near completion. I can still close it by clicking the red X top left.
