Question regarding quads fitting that doesn't converge

[Andreas Faisst]

Hello!

 

I am running some images using the astrometry.net API and although the extraction looks fine, the fit never converges.

Specifically, it keeps “trying quads” and it never ends (until it time outs after 10 minutes):

 

….

Field (processed): 1000 stars

  Field (orig): 1000 stars

Limiting search to first 1000 objects

Quad scale range: [204.8, 163863] pixels

object 1 of 1000: 0 quads tried, 0 matched.

object 2 of 1000: 0 quads tried, 0 matched.

object 3 of 1000: 2 quads tried, 190 matched.

object 4 of 1000: 33 quads tried, 828 matched.

object 5 of 1000: 143 quads tried, 2066 matched.

object 6 of 1000: 264 quads tried, 3547 matched.

object 7 of 1000: 557 quads tried, 5386 matched.

object 8 of 1000: 1006 quads tried, 8765 matched.

object 9 of 1000: 1717 quads tried, 12446 matched.

object 10 of 1000: 2975 quads tried, 17685 matched.

object 11 of 1000: 4907 quads tried, 24381 matched.

object 12 of 1000: 7485 quads tried, 32872 matched.

Got a new best match: logodds 1.27822.

object 13 of 1000: 10950 quads tried, 44184 matched.

….

 

See this job log: https://nova.astrometry.net/joblog/12853054

 

I know the parity of the image is negative, so I tried to restrict it to that, but it doesn’t help.

My input parameters are as follows:

 

subf_data = {"session": init["session"],

                 "allow_commercial_use":"n",

                 "allow_modifications":"n",

                 "publicly_visible":"n",

                 "scale_units": "degwidth",

                 "scale_type":"ul",

                 "scale_lower": 0.1,

                 "scale_upper": 5.0,

                 "center_ra":78.0058235

                 "center_dec":-60.0609995

                 "radius": 10.0, # [deg]

                 "downsample_factor":2,

                 "tweak_order":3,

                 "crpix_center":True,

                 "parity":1, # 0 = positive, 1 = negative, 2 = both -- flipping

                 "use_sextractor":False

    } 

 

 

If I try a similar image (actually the same but parity reversed, so positive), it converges very quickly.

See for example this log: https://nova.astrometry.net/joblog/12853042 

I know the center_ra and center_dec pretty well (within 0.5 degs), but even restricting this doesn’t help.

 

Is there a way to restrict the search even further to make it converge?

 

It’s just very strange that it can fit very similar image (at the same position, just flipped) but not this one. I’m probably doing something wrong.

 

 

Thanks for your help!

 

Best,

Andreas

[Dustin Lang]

Hi,

If I look at the job logs, I see that they both have "Parity: 2", the only thing that looks different between the two is the scale ("Arcsec per pix range:").  This makes one run much faster than the other.  The one that solves look at objects up to #84.  Oh, but actually the other one also looks beyond object #84, so I'm not sure why it does not also get a solution.

Do you have the submission numbers / links to the submitted images?

thanks,

dustin

--
You received this message because you are subscribed to the Google Groups "astrometry" group.
To unsubscribe from this group and stop receiving emails from it, send an email to astrometry+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/astrometry/6e768d24-26ed-4204-a612-a6f3e3ea4b8fn%40googlegroups.com.

[Dustin Lang]

Also -- it is *very* rare to require looking at 84 objects to get a solution.  Usually if we don't find a solution in the first 10-20, we'll never find a solution.  Usually this means there is something wrong with the source detection... handling of saturation, or something other issue like that.  The source detector we use is *not* smart.  Are you sure the source detection is working well?

eg, if you go to the status page for your submissions, eg https://nova.astrometry.net/status/2129809 ,

does the "source extraction image" look good -- are the brightest stars circled perfectly?

cheers,

dustin

[Andreas Faisst]

I think that might be the culprit!

Some of the red circles are in the middle of nowhere. The image that worked has a “higher contrast” (or lower background) and the circles are better aligned with actual detections.

I tried to use the SExtractor = True argument, but that also doensn’t make it better. Is there a way to add additional argument to the star detection method?

Or is there a way to upload a catalog with (x,y) of stars if I extract them myself?

 

Cheers,

Andreas

[Dustin Lang]

Hi Andreas,

Currently, in the API, the source detection methods (both the default and source extractor) do not allow setting any arguments, but you can definitely upload your own x,y list.  (https://github.com/dstndstn/astrometry.net/blob/main/net/client/client.py#L316)

cheers,

dustin

[Andreas Faisst]

I was able to solve the issue:

I'm extracting the sources myself using SEP and played with the threshold. This seems to work as I have more control over the extraction!

Andreas

[Dustin Lang]

Great!

To view this discussion visit https://groups.google.com/d/msgid/astrometry/87ebbc5d-9f73-48c4-86ec-83b9665132bfn%40googlegroups.com.