StarTools 1.8.518 Beta 1 now available

[admin]

Hi all,

In the download section of the website, you will be able to find the first public alpha version of the upcoming 1.8 version of StarTools.

Documentation for the new modules (like the SVDecon and NBAccent modules) is under development here.

Thank you to all the alpha testers; your test reports, contributions and suggestions have been extremely helpful!

Clear skies all!

Ivo

[jackbak]

Hi Ivo,

Thanks for all your work. I just tested 1.8.506 on my Intel i7 with Ivy Bridge internal GPU (16 cores) running on openSuse Tumbleweed 64.

The non-GPU version comes up fine (haven't yet played with it), but the GPU version crashes with the following shell output

ASSERTION FAILED: sel.hasDoubleType()
at file /home/abuild/rpmbuild/BUILD/beignet-Release_v1.3.2/backend/src/backend/gen_insn_selection.cpp, function void gbe::ConvertInstructionPattern::convertBetweenFloatDouble(gbe::Selection::Opaque&, const gbe::ir::ConvertInstruction&, bool&) const, line 6227
Trace/breakpoint trap (core dumped)

Best,
Jack

[admin]

Hi Ivo,

Thanks for all your work. I just tested 1.8.506 on my Intel i7 with Ivy Bridge internal GPU (16 cores) running on openSuse Tumbleweed 64.

The non-GPU version comes up fine (haven't yet played with it), but the GPU version crashes with the following shell output

ASSERTION FAILED: sel.hasDoubleType()
at file /home/abuild/rpmbuild/BUILD/beignet-Release_v1.3.2/backend/src/backend/gen_insn_selection.cpp, function void gbe::ConvertInstructionPattern::convertBetweenFloatDouble(gbe::Selection::Opaque&, const gbe::ir::ConvertInstruction&, bool&) const, line 6227
Trace/breakpoint trap (core dumped)

Best,
Jack

Hi Jack,

Could you try this version and let me know if that works?
https://download.startools.org/StarTools_1_8_506_1_alpha.zip

Thank you!

[jackbak]

Sorry Ivo,

Same result. But I'm wondering if you grabbed the right zip file. The 506_1_alpha StarTools-Linux64-GPU has the same CRC checksum as 506_alpha.

Jack

[Mike in Rancho]

Boy I don't log in for a week and all sorts of stuff happens here! Including a new alpha.

Should this double as the feedback thread on 506, or...?

I ran through some data last night, all sorts of new things to try.

- Is the initial stacking artifact warning gone?

- Things seemed faster, though maybe because I had binned down to about 2000 x 1500.

- All sorts of new Mask tricks will have to be learned, the feature changes seem substantial. I like the new circle creator, though grow still turns circles into squares. And elongated nebulosity (such as in the Veil) still seems to get picked up as stars.

- SVD sampling went great - quick, no hangups, and I selected and deselected blue boxes fairly fast.

- Only real problem I had was SVD creating little planets around certain stars out of whole cloth - typically single or sometimes double full-white pixels. I could not find settings to ameliorate this other than backing off iterations to 2x, which at that point isn't doing a whole lot of good on the resolving. Whether the particular star was in the apod mask or not didn't seem to matter. One little bright planet/moon I was able to make go away by changing around a couple blue box stars. But most others persisted and I couldn't figure out why. I thought maybe it was because I used somewhat different settings in AutoDev, but this morning I tried a more normal stretch and SVD did the same thing.

In the end I went with enough iterations to produce the resolution enhancement, and after denoise I just went over the entire image and selected all those hot pixels for the heal module.

In general, of course, 1.8 is really getting cooler by each release.

[admin]

Same result.

Hmmm...
Can you try this one?
https://download.startools.org/StarTools_1_8_506_2_alpha.zip

But I'm wondering if you grabbed the right zip file. The 506_1_alpha StarTools-Linux64-GPU has the same CRC checksum as 506_alpha.

The difference is actually in the 'resource' file.

Should this double as the feedback thread on 506, or...?

That would be great!

]- Is the initial stacking artefact warning gone?

Shouldn't be... (though detection is still hit and miss)

- All sorts of new Mask tricks will have to be learned, the feature changes seem substantial. I like the new circle creator, though grow still turns circles into squares. And elongated nebulosity (such as in the Veil) still seems to get picked up as stars.

Have you tried the new mask generator (try the AltMask preset)? It should specifically be much better at not picking up elongated nebulosity.

- Only real problem I had was SVD creating little planets around certain stars out of whole cloth - typically single or sometimes double full-white pixels. I could not find settings to ameliorate this other than backing off iterations to 2x, which at that point isn't doing a whole lot of good on the resolving. Whether the particular star was in the apod mask or not didn't seem to matter. One little bright planet/moon I was able to make go away by changing around a couple blue box stars.

Hmmm... This could be because of an outlier pixel around such a star (e.g. noisy data), or poor samples.
But most others persisted and I couldn't figure out why. I thought maybe it was because I used somewhat different settings in AutoDev, but this morning I tried a more normal stretch and SVD did the same thing. Did enabling (or disabling) PSF resampling do anything to improve the situation?

[Mike in Rancho]

Thanks Ivo,

You're right, on some other data the stacking warning still worked. Of course Wipe will smack you in the face with it anyway, so not super necessary.

There's a lot of new stuff to play with in Mask, so have yet to try many things. Am looking forward to it. I will note with happiness that the field of view stayed put in SVD when switching to mask.

As for noisy data, well I'm not sure I have anything but. It's just how I roll. However, zoom and pixel peep on the pixels that are going "wild" here, I don't see anything untoward there. Just relatively normal star areas, no hint of little stars hidden in the halo that deconv will reveal or anything like that.

Changing PSF resampling, or any other controls really, can have some effect but not really a resolution. In fact, sometimes a few wild pixels will disappear but new ones will take their place elsewhere. Could I really be that bad at picking sample stars? Maybe, but I don't recall this on 505 (unless I wasn't looking as closely).

So after trying my noisy bicolor West Veil and my noisy bicolor Crescent, with the same results, I figured I try some full RGB data for which I have a lot of integration. But why even do that, I have plenty of tutorial data on my drive.

So I tried the Elf M33, and had the same problems with...whatever we are going to call them. Wild pixels, planets, moons? Then I tried the Misti Mountain Cone Nebula. On that one, I left the image at full res after crop, no binning. I took default Wipe (not that it even needed it) and went straight to SVD. Sampling was a bear, because the starfield is crazy busy. Almost to the point it could use a "conservative" apod generation to go along with the normal and sensitive. But I found a bunch of green stars eventually scattered about the image.

It was much less, and they were harder to find, but I still ended up with some wild pixels around stars. Changing to intra+centroid did help make some go away. But not all of them, even after I dropped iterations to 5x. But at that point of course a good chunk of the resolving benefit of SVD has been left on the table.

That said, nobody else has said anything, so maybe it's me, or my machine?

[admin]

That said, nobody else has said anything, so maybe it's me, or my machine?

Hmmm... is this the GPU or non-GPU version?

[Mike in Rancho]

Hmmm... is this the GPU or non-GPU version?

GPU. I haven't used anything else in a long time.

i7-8550U, 32GB, 940MX. I could go try it on my old desktop, Q9550, 4GB, N460GTX-OC. It always crashes entering the color module, but I can test this issue out before getting to that point.

I tried another set - my infamous once-purple M106 which is about 9.5 hours integration and a pretty diffuse starfield surrounding the galaxy, and still had wild pixels.

Also separately gave the AltMask a try. A very interesting jigsaw puzzle effect. Not sure yet how to work that in. I haven't read through any updates to the development instructions though, so will go do some reading if it's there.

[jackbak]

Same fail with same Assertion error message.

Intel i7-3770K with in-built GPU, 16G. Tumbleweed current as of 07/27/2021.

Thanks Ivo,

Jack