OK… Here is the first beta release of the new Auto Focus subsystem. These are pretty big changes so I feel the need to set expectations for 2.5.1… If you are a normal consumer of our beta software you might have an expectation that they are pretty stable and might have few quirks at worst. This is usually true…
With this beta, you need to have the expectation that AF might not work at all for you (yes, even if you use a refractor). There is no special AF routine based on the type of scope you are using.
Be on the lookout for accuracy, star detection error, bad readings when the data clearly shows what should be “V”, etc…
Auto focus, in this state, should not be run unattended. It is not dangerous… we just need to get data. The easiest way to run this is by clicking “Run” manually on the AF docking module.
Here are some important notes about testing:
Anyone is welcome to to download and muck about if you are just curious.
If you are inclined to test and provide feedback (the only way AF will improve), you need to use and make AF Packs available for download (see the thread at the bottom of this post). If you cannot do this, we probably cannot use your feedback. Dropbox is a great free place to share this data.
In addition to the AFPacks, screenshots of the graph produced by the AF run can be really helpful.
I’d like to stress that, with this beta, positive feedback is just as important as negative. We need to know if changes are actually working. If you had trouble with AF in 2.5.0 and before and AF in 2.5.1 is like magic, then knowing that is just as valuable in terms of moving toward a formal release.
If you have previously turned “smart focus” off because of limitations with donut stars, please turn it back on before testing.
Nebulosity rejection and Sample size sliders are still present but have absolutely no effect on the new AF routine (we will remove them later depending on which direction we go…).
The new AF routine has several knobs and levers we can use if necessary. Because I am not a fan of settings, the initial approach will be to provide absolutely none (other than the existing step size and focus point count). If during the course of testing, we find that there is no way to universalize our approach, a setting will be added to compensate.
AF still does not allow for fast download frames. We will stabilize the new routine and then see if it is more tolerant of noise in “fast downloads”. We need to minimize the number of changes in each release so we can tell where change is originating.
Refractor users: If your current step size and data points settings are working for you, there is no need to adjust them. We still recommend 9 points.
Central obstruction / Long FL users: You will likely want to expand your step size (and use 9 data points). If you are using the current AF routine (in SGPro 2.5.0 and before), step size is likely very carefully adjusted in order to avoid donut stars. Since donut stars are now handled properly, you will probably want to expand the outside points of the run to have some “tight” donuts in the AF frame. Too much is still too much… AF is not magical (meaning vague, low intensity donut-like structures will do you no good). In general, though, you should have to be way less careful about step size.
Knobs and Levers
Right now these values are preset in the SGPro code, but, in general, are things we could allow users to adjust (if necessary):
Initial histogram stretch… we need to stretch the data before we can work with it
The number of wavelet applications (structure isolation)
The image binarization threshold
A few Circular Hough Transform params (mainly minimum allowable radius, aka minimum star size)
Please feel free to ask questions…
This thread shows how to grab AFPacks. Beta available in about an hour from now.
Odd… I’ll have to look into that again. Last time I checked our installer did not have an issue with this. I might put in a temporary AF selector mode (new and legacy) just to make testing easier for folks that want to participate.
thanks, have not had a clear night since i posted this, but might try tonight. so if i’m running a release 2.4 version and the new beta is the only 2.5 beta, it will work OK to simply install the beta into a different folder?
I think so? @joelshort indicated it might behave badly within the same version. Last time we checked this was OK. Maybe from 2.4 to 2.5 will be OK. Now there are questions and between AF and prepping for NEAF we are short on time.
I don’ know if I tried it correctly. I simply created a new folder on my
desktop and installed the 2.5.1 beta to that folder. It installed fine and
ran fine, but the installation also wiped out the previous 2.5.0.23
(beta?). I simply downloaded the official 2.5.0.23 installer and
re-installed using defaults and it all behaved properly, but I couldn’t
install both versions at the same time.
I get it. You wanted to have a stable release, as well as the beta release
so you could try the beta but then switch back to the stable version. That
does not appear to be possible as installing one (even to a different
folder) wipes out the other.
2.5.0.23 is a stable, non-beta release. You don’t need 2.4.x. In fact I’m
not even sure 2.4.x is officially supported anymore.
What I’m saying is you can install 2.5.0.23 over 2.5.1 beta and the
installation will remove 2.5.1 beta, but will still retain your settings
and GUI layout. If you use some other method to uninstall 2.5.1 beta then
when you install some other version you will need to set up the SGP screen
(modules etc) to your liking.
Of course, I may still not be understanding what you are saying…wouldn’t
be the first time.
ok - my mistake, i thought you guys were saying that you successfully installed successive 2.5 betas over one another.
in fact, it’s a double mistake, because it turns out that what i actually installed yesterday was 2.5.0.23 and not the beta. that might explain why 2.4 went away, if in fact a beta installer and release installer behave differently. anyway i still learned something which is that i guess there can be only one stable release on the machine at one time.
this is somewhat concerning though, since i had all kinds of problems with PhD integration last night, including an API guider timeout, which has never happened since upping the timeout to 350s. i don’t see anything in the 2.5.023 changelog specifying a minimum version of phd, but i’m on 2.6.0 which is hopefully new enough.
just installed the beta and yep, it deleted 2.5.0.23. so no differences between installers. hopefully the 2.5.1 beta works better than 2.5.0.23 did for me.