Auto Focus Beta Consolidation Thread


#1

While I appreciate all the feedback we are getting during the 2.5.1 (Auto Focus Refactor) beta, it is becoming difficult for me to track all of the threads created for it…

If you don’t mind, please continue that feedback here. I just released 2.5.1.5 moments ago… this is an excerpt from the release logs:

2.5.1.5 Changes:

  • More changes to auto focus. Specifically:
  • Replaced the “minimum star size” slider with a numeric entry field (the slider was always meant to be temporary). This will, among other things, allow you to be precise about this number.
  • This new numeric entry box might be more permanent-“ish”… depending on how the rest of the AF beta goes (it is ugly as all get-out, but serves a purpose right now; aesthetics later). This field is also smarter about binning… allowing you to set it for 1x1 frames and then scaling itself as you adjust to different binning modes for AF frames. This will hopefully prep the field for entry into equipment profiles as well (not currently implemented).
  • More star rejection logic to help clean up false positives (specifically for data starved images).
  • Minor changes that do not affect metrics, but should help with performance.

I think this new methodology is coming along nicely… We certainly appreciate all the community feedback here. I am especially pleased that we have come this far into the beta with only one user definable parameter. That is awesome. Making this easy to use reduces the amount of support time we have to commit to it which is ultimately realized as time we can put toward new features.

Please let us know what you think of 2.5.1.5…


#2

I thought I would take a minute to “draft” a quick and dirty version of “best practices” for setting your AF minimum star size value (since beta software has no documentation). This applies only to version 2.5.1.5 (maybe more versions as well, but the nature of beta software makes no guarantees here).

  • Find focus for your rig. Use a mask, use the image history and minimize your HFR… whatever works best for you.
  • Take a frame and focus frame that is equal to the exposure time you use for AF runs (just one). Make sure you are using the proper filter.
  • Click the star icon in the image toolbar. Hopefully a bunch of green circles around stars will show up. Adjust the “Minimum star size” as high as you can get it while maintaining a decent number of green circles (ideally 50 or more because stars disappear as you move out of focus… don’t be too bothered if you can’t hit that number… try not to go below 10). The image stats window will tell you how many stars are detected.

Right now this is a “sequence only” setting. Ideally, towards formal release, we will move this into the equipment profile. You should be able to, in most cases, use the same minimum star size regardless of target (this may not be true all the time).


#3

Last night I moved to SGP 2.5.1.5 and I have to say, I had the best autofocus V-Shape ever.

Great job, thank you Ken.
Horia


#4

It would be useful to know what your imaging equipment is.

Charlie


#5

Hi Charlie,

yes, you are right. This was for an 80mm f/7 Stellarvue refractor with a TS-Flat2 flattener, IDAS LPS-V4 filter and the EOS 1000D camera.

So for focusing an easy job. Nevertheless, I usually get winglets at one end of the V-curve and at least one point off the line.

Regards,
Horia


#6

Working ok here on a less than perfect night.6 INCH RC AND QHY8L WITH FOCAL REDUCER.


#7

Hello,
good result here also, nice V curve the last night, HFR mean value is a little bit higher than with older AF routine but it works.
Looking at the star detection shows than some nearby stars are detected as single star giving a too hight HFR, may this can decrease the AF quality.
Thanks for your great work.
Pascal

M100_120sec_2x2_R_230303.zip (545.5 KB)


#8

Reliable results here too 10"RCT with QSI683 - accurate star detection with min star size at 2-4. V curves look ok at the moment. Not going too far out 9 data points with 22 steps (0.4micron steps). Great stuff. Nice work Ken.


#9

Ken:

A question – when I look at the circled stars on my test frames, I see a whole gaggle of stars in the HFR range of 2.1. But there are also some over exposed stars that are circled that have HFRs of 8+. Additionally, there are some cosmic ray hits with HFRs of 1.0.

Since these are circled, does that mean they are being used in the “average HFR” calculation for the frame or are these “outliers” being discarded before the average HFR is calculated?

Charlie


#10

Sound like maybe your minimum star size is too low?

Depends… I think we will detect up to 300, then those get sorted by mean intensity, then the top 100 are used to calculate HFR.


#11

Ken:

Yes, the minimum star size will eliminate most of the cosmic ray hits. However, if the image only has 50 stars and some of those are big, over exposed stars, it looks like they will end up being used.

One approach is to calculate a mean HFR and then calculate one standard deviation. Eliminate all stars more than one deviation from the mean. That data set should be a better representation of the typical stars in the frame and might yield a more repeatable HFR calculation.

Hopefully all this El Nino rain in central Texas will break Thursday night and I will get a chance to actually try out all the new code.

Charlie


#12

I was think a sigma rejection of outliers too. It will work as long as you have 6 or 7 stars.

Also exclude stars near saturation


#13

Btw, great work on this!


#14

Not working at all in donut range on my RC12", reducer removed so FL=2432, using 2.5.1.5.
Here are 13 AF packs. Target was due North, mount stationary. Very clear. Collimation redone (at least the bench portion) and is fairly good. Donuts are nice and round and uniform all the way around.

When close to focus, about 30 stars counted. If I stay out of donut land, step size=50, starting in focus, I get reasonable V curves, much like in the past.
With step size=100 it gets into the start of donut land, the star count drops to 7 and it often finds a nice V curve way out of focus. Every run with step=100 ended way out of focus.
I tried star size from 2 to 6, did not make any difference. All these AF packs are for 2.

Finding bogus small stars does not seem to be an issue. Losing real stars as it gets into donut land is the problem. Did not seem to recognize my donut stars as stars.


#15

I’m actually getting pretty good results with your 100 step sets (ok with your 70, not great with 50). That said, I am using 2.5.1.6 which unscrews to the min star size bug in 2.5.1.5. These are run with a min star size of 17px (8.5px @ 2x2).

AFPack4 (never crosses focus… smart focus can recover this):

AFPack8:

AFPack9:


#16

I just upgraded from 2.5.1.3 to 2.5.1.5. I have been testing the star selection routine by loading existing images. When I was using the older version I could adjust the slider and the software would calculate a different star count and HFR for each position. After moving to 2.5.1.5 I can set the star size to any size and the same Star Count and HFR is calculated and displayed when I press the Star icon.

Also, I notice that there is field at the bottom of the screen that says:
[ ] Analyzing star data…

Which does not go away after the values have been calculated and displayed. Also the area next to the text appears to be a progress bar, but it remains empty during the calculation process. I assume these are future UI enhancements that will be cleaned up during future updates?

I can provide the image and log file in dropbox if needed.

Fred K


#17

Ken, Jared
Here are a couple of plots from last night’s session.

  1. Target M51
  2. v2.5.1.5
  3. All focusing done with LUM filter at 2x2 and 5 seconds
  4. Autofocus done with 9 steps at 75 counts per step
  5. Star pixel size set at 2
  6. There were a total of 10 Autofocus events through the session
  7. 12"LX200 at f7.16, with Atik 383L+m, OAG and Lodestar guider

The 8:01pm graph is the first AF. The 2:17am graph is the last. There were a few (out of the 10) that were as elegant looking as the 2:17am one. The rest were more balanced than the 8:01pm graph, but with similar raggedness.
All in all, I got a good focus each of the 10 times. Thanks a bunch!
Mark


#18

Mark,
Might be worth upping your step size a bit to get your max HFR around 7ish which will make it more resilient to seeing or lower altitudes.

However…may not be entirely needed if you’re getting good focus. But if you’re imaging on “not so great” nights it may be worth upping the step size.

Jared


#19

Jared:

2.5.1.5 seems to have a bug when first trying to do HFR. I was testing with images previously taken (and processed) and noticed that when I was using an unbinned image, the min star size control had no effect – click the star icon and it always found 250+ stars. Adjusting the spin control for different star sizes had no effect.

After loading a sub that was shot with 2x2 binning, everything worked as expected.

Charlie


#20

That could explain my issue. All of my images are binned 1x1 with my OSC camera. I’ll to wait for clear skies to see if my Focus frames binned at 2x2 allow adjustment of the minimum star size. I assume this is probably a bug.

Fred K.


www.mainsequencesoftware.com