StarTools 1.9 preview

[Mike in Rancho]

Neat! New toy to play with.

Will kick the tires more later, but a few thoughts -

SVD seems to be "thinking" more about my star selection. Like, did you really want to pick that star? Why yes, I did. Maybe it's that there are more candidates with the one feature on the temporary injured list? Or I could be wrong too, I didn't get out a stopwatch or anything.

Deringing starting off at something (50 now) is good. It depends on the data and the samples I imagine, but generally I've been needing to move that into the 30-40 range with 1.9 so far. Anybody else? Here, however, in a test of 539, even default 50 wasn't enough and I bumped it to 60. I then ran the same data in 536 and got away with 35. Again I'll have to re-test this, as I didn't make sure my processing leading to SVD was identical, and the samples could have been different.

And not terribly important, but the new image types in Compose don't contemplate bicolor with discrete filters, only OSC/DSLR duo/tri. I'm not sure if that leads to anything other than the initial defaults in Color (and maybe whether one gets green or purple cap). Just by chance I had decided to do an HSS (but could easily have been HOO) with mono NB filters, and selected the appropriate L, RGB non-OSC bicolor and loaded my S once, into G. For the image type, I picked SHO, since that seemed the closest even though I was missing O. My bad. I tried again with bicolor OSC/DSLR duo/tri and as far as I could tell that got me into Color with the right initial presets.

[Mike in Rancho]

As a follow up, I am in fact continuing to have trouble with the new SVD deringing in 539a, and so have reverted back to using 536a.

Across multiple datasets, processed identically leading up to SVD and then within SVD as well, I would typically be needing 30-40% deringing, and the results would be great. Nice stars, no rings, and pinpointed sufficiently even though the deringing backs that off some. In 539a, the starting point of what it considers 50% deringing usually has a lot of ringing showing. But, upping the deringing to 60, 70, 80, either doesn't fully get rid of all the ringing, or, at the level where it finally does, the star shapes, sizes, and core pixels have become mangled. Yuck!

I do have third option intra-iteration centroid tracking on as well.

Anyone else, or is it just me and my data?

Can provide screenshots if useful.

[decay]

In 539a, the starting point of what it considers 50% deringing usually has a lot of ringing showing. But, upping the deringing to 60, 70, 80, either doesn't fully get rid of all the ringing

Same here, Mike. I was not sure, because I have a new dataset, which I had not processed with prior versions of ST. But I do see this unusual heavy ringing as well.

For me there seems to be a second issue: The sampling of the starfishies is incredibly slow After every click to select a fishy I have to wait for at least one minute or so, until it goes on. This was not the case before, and I'm quite sure, my system didn't change in between.

Best regards, Dietmar.

[admin]

Hey @Mike in Rancho and @decay, I released 1.9.541 a few hours ago, which has some changes (again) to the SV Decon module. Are you still seeing regressions in this version?

[alacant]

Hi everyone

SVDecon is fine on both 539 and 541 under Ubuntu 22.04 even on an old i5, 8Gb.
Could the issues be os specific?

Observations
Decon of the stars seems harsher than 1.8. 5 iterations in 1.9 seem to give similar stars to 10 in 1.8.

The starfish method of sampling is an improvement.

The transformation Decon makes is stunning.

Cheers and HTH
Steve

[Mike in Rancho]

Yeah still having trouble here.

The apod mask is improved, however! Better sampling in 541, on the one dataset so far.

Speaking of, the apod mask is no longer used for deringing support in 1.9, correct? Since it still leaves some things on the table or even cuts stars in half, and there's no touch-up.

Anywho I took some screenshots, but might take two posts (3 attachments tops, if I remember right?).

I ran this data through normal full processing in both versions exactly the same by referencing my log. Compose, Crop, Bin 35, Wipe, Contrast defaults, HDR (close to reveal core), Sharp (saved and reused the same modified mask), and then to SVD. Defaults except intra-interation centroid and then deringing. The only difference would be I could not match samples as the apod masks were different. So I loaded the 15 sample PSF's used in 541, and ended up with only 6 that carried over to 536. So I sampled a few extra stars here and there.

Here's the 541 image with the before button pressed.

541 SVD Before.jpg (212.58 KiB) Viewed 13835 times

And then the 541 after with default 50 DR.

541 SVD Default DR 50.jpg (213.92 KiB) Viewed 13835 times

Finally, 541 with the DR upped to where I think I removed all obvious ringing artifacts. But, the interior of the stars has become kind of scrambled with light and some more shaded pixels, and the star cores themselves have just turned into a flat disk that really isn't smaller than the original (if not actually bigger). Most resolving inward towards a point source seems to be lost, and other than the non-star detail, I think the star field is degraded from the pre-SVD state.

541 SVD DR 84.jpg (211.62 KiB) Viewed 13835 times

[Mike in Rancho]

Meanwhile, back in 536, this would be the default state of things but for intra-iteration. A little different to start out with.

536 SVD default DR 0.jpg (211.5 KiB) Viewed 13834 times

But, just 35% DR leads to this. A much more normal and expected result from SVD+DR, IMHO.

536 SVD DR 35.jpg (210.52 KiB) Viewed 13834 times

Wow! Look at that. Beauteous! This guy has mad SVD skillz.

[Mike in Rancho]

Ok, an addendum...

After a lot of extra testing this morning, I think my binning may be at the root of things here, at least to a large extent.

At full scale, SVD in 541 is a lot less objectionable on the artifacting and ringing, and a reasonable bump of DR to 65% or so from the default 50 was producing an acceptable result. And a subsequent bin, while not otherwise optimal as it sort of mushes things up, does not appear to resurrect the ringing and other issues that are showing up if binned first.

Now, I generally don't process at full scale, both because I want (need?) that extra SNR, it takes a lot of computer crunching time, my current monitor is only FHD, and other than perhaps a-bin, which I don't have, you can only post so much resolution anyway.

In 1.9.536 and before (incl 1.8), at my imaging scale and sensor size, a 35% bin would provide processing that I liked including through SVD and deringing. Maybe I need to amend my workflows going forward?

Obviously there will be a huge range of equipment and imaging scales utilized by ST users - have I been doing something unusual? I don't know what the common scales and binning workflows are out there, though I know from the past that pretty substantial bins were oft-suggested, especially with APS-C/DSLR sized sensors.

I suppose I could get more integration in lieu of heavier binning. Oh the horror!

[decay]

For me there seems to be a second issue: The sampling of the starfishies is incredibly slow After every click to select a fishy I have to wait for at least one minute or so, until it goes on. This was not the case before, and I'm quite sure, my system didn't change in between.

Unfortunately, this problem is still there. It takes about 30s after picking a sample until ST gets responsive again. This is/was not the case with ST 1.8 and with prior alpha versions. It was not simple to take a screenshot while system is not responding ... So I took my smartphone

signal-2023-01-24-195810_002.jpeg (383.82 KiB) Viewed 13800 times

[decay]

OK, it's addendum time ... sorry.

I just retried ST 1.8 with the same dataset and likewise workflow, to exclude any mistakes. It's absolutely no problem to select the samples - from time to time (perhaps every 5th sample) the system needs a few seconds, but no problem at all.

And I have the impression, that ST 1.8 in general is much faster in processing

I noticed, that ST 1.8 allocated about 3.5 GB memory, but ST 9 alpha only about 1.5 GB. That's odd, isn't it?

2023-01-24 20_30_07-Clipboard01 - IrfanView.jpg (20.18 KiB) Viewed 13798 times