1.8 closed alpha feedback

[Stefan B]

Update: next SVDecon crash after selecting two stars not far away from each other within a short time period.

EDIT: Sorry, Ivo, I missed your previous post saying a new alpha is on the way. Thanks!

[Mike in Rancho]

I got a chance to try out 1.8.503 the past couple nights, first just a quick kick-the-tires on an M104, and tonight a bit more comprehensive run throughs on my M5 glob. I know a new release is on the way, but my bunch of random, disorganized thoughts...

No crashes or hangups on my semi-souped up four-year-old i7 laptop. Things ran pretty fast, actually. Or maybe it felt that way as I was staring at the new spinning timer thing in the lower corner, trying to figure it out?

The color/lum toggle in Wipe says NB as the alternate state (correctly Lum down below). This is on single file DSLR stack using "open," have not tried compose or NB features yet.

Must learn more about the Correlation Filter, how to use and what to look for. Might be useful as I am likely always oversampled.

Only tried on the M104 so far, but the new default gamma in SS did not work well for me on the Sombrero halo. Took me a minute to track down what changed and then revert it to the 1.7 setting.

On my M5, when using 1.7 I actually ran a bit of sharpening on it after 2 HDR's (one normal, one reveal core), with the dark/light slider at 100%/0%. In 1.8 I was not getting any effect here at all. Mask was fine and as intended (conservative). I opened 1.7 and ran through the same log to confirm.

In SVD the sampling toggle does not have an alternate state name.

SVD is going to take a lot of learning and experimenting.

Searching for green star cores is a bit tedious, though perhaps fun in an Easter Egg hunt sort of way.

I had to reset the blue square size twice (grew it). Does it start out at any kind of auto-sensed state?

Emulating deconv from 1.6/1.7 is not so easy, what with the new masking regime. But I tried anyway, using the same settings for M5 that I used in 1.7 (no secondary, Moffat only, 2.0 rad, conservative mask with the glob core opened up more via lasso). It really didn't work the same. Stars did not seem to resolve and tighten, instead I just got bright star centers (often squares). Playing with the spatial error helped out some, but...

Using the variant sampling deconv, I still noted just brightening star centers/squares rather than resolving. Again I opened in 1.7 also and ran through the log to see the difference. Granted I am zoomed in and pixel peeping the glob stars, but the difference is fairly stark. Obviously I need to play with this more. Could it be the stars I selected in sampling? They were generally a bit larger and scattered about the FOV. Not easy to get sampling inside the busy glob of small stars, especially after I had to increase the size of the blue box. As with the non-variant attempt, I did try playing with the spatial error slider too.

However, despite (so far) not being able to get M5 to resolve as good as I could in 1.7, 1.8 did spectacular on the double star 5 Serpentis, clearly resolving the large A and small B components. In 1.7 I had to run a second deconv with a custom mask to get that to clarify nicely, along with some repair work afterwards, and it still couldn't hold a candle to 1.8's results.

That's all I got for tonight. I'll continue throwing things at it. Glob and a tiny galaxy are probably less common targets anyway.

Mike

[admin]

On my M5, when using 1.7 I actually ran a bit of sharpening on it after 2 HDR's (one normal, one reveal core), with the dark/light slider at 100%/0%. In 1.8 I was not getting any effect here at all. Mask was fine and as intended (conservative). I opened 1.7 and ran through the same log to confirm.

The Sharp module indeed may need tweaking.

SVD is going to take a lot of learning and experimenting.

Searching for green star cores is a bit tedious, though perhaps fun in an Easter Egg hunt sort of way.

I had to reset the blue square size twice (grew it). Does it start out at any kind of auto-sensed state?
Emulating deconv from 1.6/1.7 is not so easy, what with the new masking regime. But I tried anyway, using the same settings for M5 that I used in 1.7 (no secondary, Moffat only, 2.0 rad, conservative mask with the glob core opened up more via lasso). It really didn't work the same. Stars did not seem to resolve and tighten, instead I just got bright star centers (often squares). Playing with the spatial error helped out some, but...

Using the variant sampling deconv, I still noted just brightening star centers/squares rather than resolving. Again I opened in 1.7 also and ran through the log to see the difference. Granted I am zoomed in and pixel peeping the glob stars, but the difference is fairly stark. Obviously I need to play with this more. Could it be the stars I selected in sampling? They were generally a bit larger and scattered about the FOV. Not easy to get sampling inside the busy glob of small stars, especially after I had to increase the size of the blue box. As with the non-variant attempt, I did try playing with the spatial error slider too.

It sounds something went wrong... SVDecon should always yield improved results to 1.7 for both synthetic AND sampled PSF deconvolution. It should still work the same too; set PSF radius and the amount of iterations and it will do its thing. You cannot, however, use the same parameter settings for 1.7 in 1.8 and vice-versa. 1.8 requires very different tuning.

However, despite (so far) not being able to get M5 to resolve as good as I could in 1.7, 1.8 did spectacular on the double star 5 Serpentis, clearly resolving the large A and small B components. In 1.7 I had to run a second deconv with a custom mask to get that to clarify nicely, along with some repair work afterwards, and it still couldn't hold a candle to 1.8's results.

That's all I got for tonight. I'll continue throwing things at it. Glob and a tiny galaxy are probably less common targets anyway.

Mike

If you'd like to share the dataset with me, I'd love to have a look. Thank you for your comments and testing!

Ivo

[almcl]

Don't know if this is of any importance, but when attempting to select stars in the GSVdecon module, I try and click on the central pixel (example below) and while this sometimes produces a blue box centred on the subject star, sometimes it doesn't. In the example below it seems pretty obvious where the centre is (the only green pixel) but the generated blue box is shifted away from this. Occasionally by several pixels.

Not sure if this matters normally but when the box is close to the star size it sometimes misses a bit of star?

The target star:

selecting a star1.jpg (22.27 KiB) Viewed 7391 times

The result of clicking on the green pixel:

selecting a star2.jpg (28.66 KiB) Viewed 7391 times

[Astroscapist]

The new deconvolution is much weaker for moon shots.

Old

old deconvolution

Startool old.jpg (401.43 KiB) Viewed 7387 times

New

new deconvolution

Startool new.jpg (412.79 KiB) Viewed 7387 times

[admin]

Don't know if this is of any importance, but when attempting to select stars in the GSVdecon module, I try and click on the central pixel (example below) and while this sometimes produces a blue box centred on the subject star, sometimes it doesn't.

In the example below it seems pretty obvious where the centre is (the only green pixel) but the generated blue box is shifted away from this. Occasionally by several pixels.

Thank you. This is not something to worry about - the center of a sample is not necessarily the brightest pixel. Instead, the center is the "luminance centroid". It is the weighted (by brightness) mean of all pixels in the sample. This is so that, for example, samples of stars that are deformed or heavily defocused (where their centre is less bright than their surroundings) are still captured correctly. I will update the documentation accordingly. Thank you again!

Not sure if this matters normally but when the box is close to the star size it sometimes misses a bit of star?

The Sampled Area bounding box should ideally be extended so that the samples fit in the box and are not cut-off by the box.
From the documentation;

The Sampled Area

The immediate area of a sampled star is indicated by a blue square. This area is the 'Sampled Area'. A sampled area should contain one star sample only; e.g. you should particularly avoid parts of other stars in the blue square surrounding a sample. This size of the blue square is determined by the 'Sampled Area' parameter. The 'Sampled Area' parameter should be set in such a way that the single samples fall well within the blue area's confines and are not 'cut-off' by the blue square's boundaries.

Hope that helps!

[admin]

Thank you Stefan,

The new deconvolution is much weaker for moon shots.

Old
Startool old.jpg

New
Startool new.jpg

From the screenshot it appears you are using far to few iterations; try the dedicated Planetary/Lunar preset.

I have updated the documentation now to reflect this;

A preset for lunar, planetary, solar use quickly configures the module for lunar, planetary and solar purposes; it clears the apodization mask (no star sampling possible/needed) and dials in a much higher amount of iterations. It also dials in a large synthetic PSF radius more suitable to reverse atmospheric turbulence-induced blur for high magnification datasets. You will likely want to increase the amount of iterations further, as well as adjust the PSF radius to better model the specific seeing conditions.

Settings you are used to in 1.7 are not compatible with 1.8.

Do let me know if you still cannot get better results with 1.8 vs 1.7.

[Mike in Rancho]

It sounds something went wrong... SVDecon should always yield improved results to 1.7 for both synthetic AND sampled PSF deconvolution. It should still work the same too; set PSF radius and the amount of iterations and it will do its thing. You cannot, however, use the same parameter settings for 1.7 in 1.8 and vice-versa. 1.8 requires very different tuning.

If you'd like to share the dataset with me, I'd love to have a look. Thank you for your comments and testing!

Ivo

Yes, I am starting to realize the reality of the parameter differences between 1.7 and 1.8! I know we have extra controls that need to be learned (very much so), but I did sort of figure the iterations/radius pixels would match up when using SDV in its synthetic 1.6/1.7 mode. Nope! I can push the radius to points that would make an abomination of things in 1.7.

I also can't find that the mask (normal, not apod) in synthetic mode does....anything? Pixels seem to be fully affected regardless of full mask, empty mask, or a star mask.

I have mostly stuck to synthetic mode for now, figuring I better learn how to replicate 1.7 before moving on to bigger things, but it has been a bit of a slog. Eventually with parameter changing I have been getting closer to 1.7 deconv results, though I'm still getting a bit too bright and square on the M5 glob stars. Expanding the dynamic range turned out quite necessary to tame it back down too. However, a working mask would be helpful as the same settings do not seem to work well for both M5 and 5 Serpentis.

For 1.8 SVD, this has also somewhat been the case playing with the parameters. I can make M5 or 5 Serpentis better, not both. But I'll get there I'm sure once I get the hang of it. The next 1.8 release may also remedy a few things that are unbalancing me at the moment, so there's that.

I did upload the file to Drive - though with all the things you have to work on it is not of great import (unless you think it can help). It's 2.5 hours of M5 in DSS, all settings as ordered by ST. I did set RGB align (not a no no?) due to Pepsi-symbol stars, though I haven't tested yet to see if it was from atmospheric dispersion or I have a lens issue. Nikon D5300 data.

https://drive.google.com/file/d/1vqWT6I ... sp=sharing

This (if I can get upload to work) jpg was my final result for it in 1.7. Crop to the shown FOV, bin to 71%, Wipe, AutoDev, a normal and reveal core HDR both at 1.4, sharpen dark, deconv (custom mask), second deconv on 5 Serpentis, color, and denoise, followed by various little cleanups using shrink and layer.

Mike

[deanST]

that is one sweet M5 Mike! I have also been plugging at M5 with 1.7 and now 1.8. I'd be interested
to read your workflow if you could post it from the log in code blockers.

[Mike in Rancho]

that is one sweet M5 Mike! I have also been plugging at M5 with 1.7 and now 1.8. I'd be interested
to read your workflow if you could post it from the log in code blockers.

Thanks, Dean. (also presuming DeanST = DeanCN? just a hunch. could be wrong)

If so, yeah this is the same image in the monthly challenge already - so acquisition specs all in there. Except I think I forgot to mention dithering, which I now do (thanks to Ivo and the Elf).

I gave it another run in 1.8 tonight, and had much better luck than last night with SVD. Really takes some changing stuff around and learning new balancing though with the dynamic range extension and spatial error - though my iterations ultimately ended up right back where I started lol. Oh and I also didn't bin, to see if maintaining more resolution could be at issue. Anyway, despite M5 looking better, I still couldn't shake the big "gray hole" in 5 Serpentis A.

The 1.7 is still the best of my M5 though. For now. The workflow is relatively straightforward except for a few custom masks, but then it gets pretty loony with the touch ups - both at the end of my main log and then two extra little logs. Totally glad to post them up, though I don't know how easy that stuff would be to follow. Even beyond the garbled ascii, I sometimes have trouble figuring out what mask means what in the ST logs.

Nonetheless, and I'm sure this will sound lame - but, is code blockers a sub forum here for log posting? Or something that strips the mask codes out? Or...? But I could also put it somewhere where it wouldn't be clogging up the 1.8 feedback thread too. Too new here to know though!