82 replies to this topic

[ShortLobster]

PixInsight has released a new gradient removal process, GradientCorrection. In the latest release, though I haven't upgraded. They also posted a video:

 

https://youtu.be/3fs...GLywWJSA_E7ausv

 

Has anyone tried it yet?

[auroraTDunn]

Playing with it right now. Here's what I posted in another thread. Give me a couple more minutes and I'll comment on a particularly tricky project I've struggled with and am testing right now.

 

As for tool itself, I'm playing with it right now on images that I have struggled with when using DBE and/or GraXpert. I have to say that so far I indeed have only needed to make minor changes and so far only for objects with lots of dark dust (I lowered the Structure Protection and life is grand)

I do wish we could get it to create a new image upon request instead of updating your current image (but then mainly because I like looking at multiple attempts to get things just right and making clones of the original fixes this issue). Plus I find the mask to be harder to interpret exactly the strength per region that was manipulated, but this is minor.

Al in all, so far after just a few tricky images, I LIKE IT!

 

[EDIT] I also find 'Automatic Convergence' really helpful on my widefield nebulous regions dealing with fainter regions as I like to take long total exposure times to get the best SNR possible (per sanity level for the given project)

 

[auroraTDunn]

OK I'm sold on this new tool!!! (Sorry DBE and Graxpert!)

 

I've been trying to reprocess my last Falls image of IC396 (the Elephants Trunk) where I have a lot, but not enough, signal on dimmer parts of the outer region using my RC71 and ASI6200MM.

 - GraXpert fails miserably on the background extraction unless I run it on the individual channels (S,H, O) first and then Channel Combine.

 - DBE has been hard since I have close to nothing but gas in my image, then after I do a BXT(correction only) SXT, then a normal BXT and NXT, the NXT no matter how I deal with BXT and NXT makes the image unusable (banding like you don;t want to see unless you are looking at an 8bit image) So I've had to do a second DBE right after SXt to get things to work.

 

With GradientCorrection I lowered the Protection threshold to 0.05 and checked Automatic convergence and boom everything works all the way through, no need for a second background correction and it looks magnificent.

If anyone's interested I can make a comparison showing exactly what I mean by these comments (noting it would be much later this evening as I'm about to start WBPP up on another project to test.

[nwcs]

I haven’t tried it yet but I’m looking forward to it. I like Graxpert but I would prefer to stay in PI for processing. Although I wish they weren’t so protective with third parties.

[Dan_I]

I have tried on luminance data for a galactifc field with pretty severe LP gradients (a very different use case than auroraTDunn). I found that GradXpert did a better job than the new tool. I have tried to tweak a bit the parameters of the latter but did not succeed in matching GradX's results. It is quite close though.

Edited by Dan_I, 01 March 2024 - 06:15 PM.

[pfile]

juan cracks me up. local normalization never worked until someone else wrote a similar tool for PI and that lit a fire under him. result? awesome local normalization tool. now graxpert comes along and lights the same fire and suddenly we have this tool. he's been threatening a new gradient removal tool for about 10 years now, and all it took was the threat of competition.

 

i did try it on one particularly problematic image that i got from the astrobiscuit discord. it's not clear what knobs to tweak, but i did try the automatic convergence and lowering the smoothness and this improved the result in a particularly dark corner of the image. interestingly running it a 2nd time did not seem to clean up the residual dark corner.

 

i have a lot of images with really bad LP gradients so i'll be going back to those and seeing what i can do with this process.

 

rob

[calypsob]

I have tried on luminance data for a galactifc field with pretty severe LP gradients (a very different use case than auroraTDunn). I found that GradXpert did a better job than the new tool. I have tried to tweak a bit the parameters of the latter but did not succeed in matching GradX's results. It is quite close though.

Can you by chance show an example of the differences you found and tell us more abt the sky conditions? 

[calypsob]

PixInsight has released a new gradient removal process, GradientCorrection. In the latest release, though I haven't upgraded. They also posted a video:

 

https://youtu.be/3fs...GLywWJSA_E7ausv

 

Has anyone tried it yet?

I just watched the video.  Im pretty pumped about this one. I have some galaxies that ride along the light down of the city near me and Im hoping this tool is the ticket to gradients. I thought DBE worked fine already but it took some time and patience to get setup right. 

Edited by calypsob, 01 March 2024 - 06:56 PM.

[DeanS]

I am far from an expert but I tried an M42 OSC  with this, and graxpert and I think the graxpert did a better job.  This is with only the default settings on both.

[Elusivephotons]

I suggest watching the Pixinsight Youtube videos.  This gradient removal tool is not designed to work with 'default settings'.  You have to adjust in order to get best results.  (at least that is my limited experience so far)

[Dean J.]

I'm sure it won't be long before Adam Block puts out a video explaining the new tool.

[Marcelofig]

I'm sure it won't be long before Adam Block puts out a video explaining the new tool.

Yes, here waiting for Adam's video.

[Rustyd100]

Just updated Pixinsight. New Gradient Tool doesn't seem to be listed anywhere.

[Rustyd100]

Duh...figured it out. Separate complete application installation. The usual update procedure will not work with this version. Downloaded installer from my user page on pixinsight.com

Edited by Rustyd100, 01 March 2024 - 11:34 PM.

[shenmesaodongxia]

In cases where flat field calibration is problematic, graxpert is used in separate channels, which is much better than gradient correction

[Arie]

Duh...figured it out. Separate complete application installation. The usual update procedure will not work with this version. Downloaded installer from my user page on pixinsight.com

Don’t forget to copy your tensorflow.dll along with the new version.

[*Axel*]

Made a quick test yesterday on data taken from my SQM18 place.

Fiddled a lit bit with the tool but it looks very promising.

 

 

[TodRiley]

The big test for me will be how the new tools deals with gradients in an image with bright and faint background nebulosity.     I have yet to view the video so perhaps that issue is addressed. 

[calypsob]

Does it harm ifn or faint background color?

[Dean J.]

Made a quick test yesterday on data taken from my SQM18 place.

Fiddled a lit bit with the tool but it looks very promising.

 

429931427_1493733777849683_3500471886342982496_n.png

That looks great.

[Dan_I]

Can you by chance show an example of the differences you found and tell us more abt the sky conditions? 

From left to right : original, GradXpert and GradientCorrection. I used a pretty aggressive stretch in order to highlight the differences. I find GradX  a hair better, but not buy much admittedly.

 

 

 

Data was acquired from my Bortle 7 backyard with a rather low transparency. The final image is there: https://www.cloudyni...-1-and-arp-285/

Edited by Dan_I, 02 March 2024 - 09:55 AM.

[auroraTDunn]

Wow I hop on to transfer (spelled delete - stupid clouds) last nights images and found a ton of videos on this new tool.

Going to play more with it today on images with IFN.

So far my thoughts are the same as always, it all depends on the data as to which is best for the needs, Easy wise its smack dab between GraX and DBE.

What it NEEDS is a preview window so you don't have to keep undoing, running, undoing, or making a ton of clones in order to find the sweet spot for the data at hand.

[jlpollard10]

Anyone have issues with Starnet 2 and the new update?

All of a sudden Pixinsight doesn't want to keep Starnet 2 loaded so I have to install it every time I start Pixinsight now.

If I try to image solve with Starnet 2 loaded the program crashes to the desktop.

I've tried on a couple different images and the same thing happens on all of them.

If I image solve first and then load Starnet 2 it works fine.

 

Jim

[Higgsfield]

I thought it would be interesting to compare PI's GradientCorrection against GraXpert and DBE, using an OSC image with a pretty severe gradient.  What a shocking difference!  While the GC did leave some gradient present on the RHS, there was no perceivable change in color of the nebula or stars. This was not true for the other two methods. DBE was the worst, leaving significant low frequency color noise, but this same noise is also present in the GraXpert result too, but to a lesser degree. By selecting the different options in GC, slightly different results can be gotten. In terms of the other slider options, I really would not know what to adjust. I'm sure Adam Block will have a 10hr video on it soon though, it it not already out   In terms of not removing signal, I would say GradientCorrection is the hands down winner. For mono images, this may not the true, but I suspect it will be the new goto method. 

 

[nwcs]

I finally had a chance to do some testing and Graxpert and GC are very similar with my data but I’m just blind doing it since I’m not sure what options work best. Yeah, Adam Block video to the rescue. And I agree, a preview would be nice!