Jared,
Prior to last night’s run, I took an older AFPack (NGC3166, v2.5.1.3, min star pxl size 4, see SGPro AF graph) and ran it through v2.5.1.5 with min star pixel sizes of 2, 5 and 6. The 2 and 5 px sizes gave identical values of star count and HFR. The 6 pixel gave different star counts, and in my mind, an inferior focus result.
The red curves on the HFR chart and the STAR COUNT are for the “2 and 5” pxl minimum star size. The HFR shows that location 4 is the best focus (which it was) and location 9 returned to essentially the same HFR value. This agrees well with the SGPro AF graph.
The blue curves on both the HFR and STAR COUNT are for the 6 pxl. Notice once we moved to the 6 pxl minimum size, the HFR values produced an inferior curve. The 6 pxl showed location 1 as the best focus (which it was not).
This is why I chose to use a low value for the pixel size on last night’s run with M51.
Mark
Please discard any results from 2.5.1.5. There is a bug in the min star size logic that will lead to odd behavior.
Also, please remember that if you’d like to participate in the AF beta you will need to turn on and share AF packs. Your feedback and analysis is appreciated, but ultimately there is not a lot we can do with it in this case.
It’s probably not a great idea to use it…
2.5.1.6 has been released this evening…
Ken:
Great! I’m sitting here in my observatory waiting for the last bit of twilight to fade. I will be testing 2.5.1.6 shortly.
Charlie
Ken:
Well, the clouds wiped me out but I did get some good runs with AF. Short answer – worked well after I was able to tweak all the parameters sufficiently. Tested with filters as well as with Lum. I had just removed my 12" Meade ACF and replaced it with my 105mm f/6.2 APO, so no testing with donuts until next week. Will post AF packs to dropbox on Saturday.
AF run made in the middle of a light frame sequence:
Position of focuser selected by AF produced small, well focused stars.
Charlie
Following up with my previous post, I have put two AF packs on Drop Box. The first was and early run where the number of focuser steps way too small. The second was the last AF run of the night where everything worked great.
The Optec Gemini focuser I use has an extremely small step size of 0.1 micron. In order to get the focuser to move enough to get decently out of focus, I had to up the number of steps per data point to 1000. This made the total change in HFR large enough to produce a real “V” curve or at least a U curve. I may experiment with even larger number of steps to see if it improves repeatability. The min star size was set to 10 with focus frames binned 2x2 @ 10 seconds. Scope 105mm f/6.2 (fl 652mm). Focus frames were seeing 200+ stars.
Charlie
Thanks for the report…
I’d just like to make a few quick points. AF seems to work well for sct - but I have long standing requests that would apply here. One is that when the focus curve is complete - you show what the true focus has been decided to be. The plots people are showing now show curves and lines and red crosses - but they don’t show what your routine decided was the true focus point.
I’m still not clear exactly how you use the curves to decide the location of true focus - and what the lines and x’s mean.
In addition - there is a terminology issue where you use HFR explicitly as a radius measure - but then you talk about a minimum “size” for a focus star. Is that also radius? Then you should say minimum HFR - non-binned.
Actually I think everything should be specified in arc-seconds - including centering error. At least as an option. It is what I would use.
Frank
Just tried 2.5.1.6 and it didn’t find stars much at all with my Edge11 at 0.4" per pix. I set min star size to the min value of 2 - and it didn’t find any stars even though they were right there.
I switched back to 2.5.1.5 and it seems ok.
I’ll send logs and stuff later.
Frank
I noticed the same thing on the Edge14. However, when I raised the minimum star size threshold (I settled on 8), it started selecting a lot more stars. I’m not sure why, but it worked.
Tim
AFpacks for Canon 6D + SV102mm refractor using 2.5.1.6.
The general results I have been getting with recent betas shows the following general features:
- When step size is carefully tuned, in particular not so large that stars get very big, the AF routines work well for both refractor and RC. Minimum star size setting has minimal effect, so long as it is small.
- If step size is large enough to increase the star size, far fewer stars are detected, even though all the stars that were detected close to focus are still clearly there, it just quits detecting them.
- Seems to me that if it uses a completely different set of stars at different stages of the process, that is like comparing apples to oranges and is bound to give bogus results.
- Minimum star size should never be set to larger than perhaps 4 or 5. What I have seen on some runs on above hardware is that there are only 1or 2 stars larger than 5, where there are thousands of small perfectly focused stars. With min set to 5 or 6, none are counted.
The old routine always ran perfectly with my refractor, even with a large step size.
The last 2 runs in the zip illustrate this new problem perfectly.
The next to last run started at focus with step size=50 and gave a lovely V close to the start with 300 stars.
Increasing it to 80 on the last run had the effect that the first image 320 steps out had large enough stars that it found very few. From that point on it was toast.
Suggestions for how to deal with these problems:
-
Finding the same stars (or close) in all the frames is important. Clearly when way out of focus this might be close to impossible.
-
The routine should early on be sure it is going in the right direction.
One way to do this:
Assume that the current position is reasonably close to focus. Take one image there at the start and use its star count and HFR as the prime reference. All additional frames should progressively improve on this or it is going in the wrong direction. And it should be finding the same stars. -
Using the HFR curves when out of focus is close to useless.
-
If the routine is finding fewer stars, then it is probably going in the wrong direction.
A possible new approach:
a) start with the initial position prime reference.
b) step away from it going out.
c) if HFR initially improves then starts to get worse, you reliably have both sides of the V.
d) if HFR initally gets worse, go the other way, which should immediately produce improving HFR.
With 2.5.1.6 I set the min star size to 8 (I am binning 2x2 for focus) and SGP at times finding stars that are <1 HFR and hot pixels.
2.5.1.6 worked perfectly on my Astrograph, 50% CO.
1.66 arc seconds/pixel
Then you have a settings issue. 2.5.1.6 is the same as 2.5.1.5, but with a single bug fix for minimum star size.
This is actually due to an implementation of one of your suggestions. The intensity of visible stars does not register high enough against the mean background level.
I don’t believe this to be true…
Please be clear that this is your opinion for your gear and is most definitely not true as a general rule.
Sorry, we cannot provide assistance here without AFPacks.
Also… I ran a few of your AFPacks with min star size in the area you are contending does not work…
Here is size 5 (10 at 1x1):
Here is min size 6 (12 at 1x1):
Not so sure about that… Here is the result of the last AFPack run at min size 6 (12 at 1x1). To me it’s amazing that this worked at all… you target HFR at this binning is ~1.8 and this AFPack starts with HFRs near 17. That said, it did work… smart focus, combined with proper min pixel size would have saved this run:
Well - if I set the minimum star size to the minimum possible value - and it found no stars - i.e. it rejected all stars - how can I change the setting to make it work? Putting it higher would just make it reject more - wouldn’t it? The screen if full of stars - yet none are found - with minimum star size set to the minimum possible value.
I will send an afpack in a bit - but if you somehow fixed the behavior of minimum star size - and now it doesn’t work for me - then that fix appears to be the issue here - since no other parameters have changed.
Frank
Hmm - When I went from 2.5.1.5 to 2.5.1.6 it appears to have disabled the “save autofocus pack” setting - so I didn’t record any autofocus runs. My prior imaging session a few nights ago with 2.5.1.5 saved all focus runs. Then I upgraded to 2.5.1.6 and it didn’t save any. Do I need to turn it on each time I upgrade?
Also - I’m puzzled that the plots people show above show the autofocus run as complete - whereas my plots in the AFPack say “100% complete” with a green bar at the bottom - but also says “Taking autofocus exposure (0s remaining)…” And there is no info on the the final autofocus value. I believe the plot is being created prior to the complete unwinding of backlash and return to focus. I have reported this before with a request to display those values as soon as they are determined - and not to say it is 100% complete - because it isn’t yet until it fully arrives at the intended focus position. Attached is an example from a focus pack - from 2.5.1.5.
Frank
