Is there anything special about the --downsample in solve-field?

[Umair Khan]

Hello,

I'm solving a series of 7 images (on an embedded Linux board) and am coming across some behavior that I'm not sure how to explain.

If I do:

1. source images are 1280x960
2. solve-field with --downsample 2 (to 640x480)
   

then it takes about 4 minutes for solve-field to solve all the images.

On the other hand, if I do:

1. source images are 1280x960
2. downsample the image prior to solve-field using magick mogrify (to 640x480)
3. solve-field with no downsample
   

then it takes about 8 minutes for solve-field to solve all the images.

I can't think of any reason why downsampling within solve-field would reduce the time so dramatically, unless there was some other optimization performed with the downsample operation. Am I right about this, or is there something else going on here that someone can shed some light on? For the application I'm working on, the second scenario is ideal since I could use the source images in other contexts without having to downsample again, but I would expect it to also take 4 minutes.

Thanks in advance for any info.

- Umair.

[Eric Sibert]

Are the number of detected stars the same in both cases?

May be the way downsampling is done (for instance average of the 4
pixels instead of an interpolation) can change the background noise
and, hence, the number of detected stars.

Eric

[Umair Khan]

There do seem to be a lot more distractors when the downsampling is done prior to solve-field.

Downsampling within solve-field (4 minutes):

1. 972 sources, 30 match, 0 conflict, 103 distractors, 38 index
2. 1164 sources, 20 match, 0 conflict, 82 distractors, 25 index
3. 1084 sources, 12 match, 0 conflict, 30 distractors, 15 index
4. 1093 sources, 20 match, 0 conflict, 41 distractors, 27 index
5. 1146 sources, 16 match, 0 conflict, 31 distractors, 26 index
6. 1082 sources, 27 match, 0 conflict, 33 distractors, 33 index
7. 1099 sources, 24 match, 0 conflict, 18 distractors, 30 index

Downsampling before solve-field (8 minutes):

1. 813 sources, 31 match, 1 conflict, 121 distractors, 38 index
2. 861 sources, 21 match, 0 conflict, 146 distractors, 25 index
3. 821 sources, 12 match, 1 conflict, 55 distractors, 15 index
4. 842 sources, 20 match, 0 conflict, 33 distractors, 27 index
5. 900 sources, 16 match, 0 conflict, 31 distractors, 24 index
6. 874 sources, 28 match, 0 conflict, 115 distractors, 33 index
7. 992 sources, 65 match, 0 conflict, 146 distractors, 121 index

There are 338 distractors total in the former and 647 distractors in the latter.

I will experiment with different methods of downsampling and see how it goes. Thanks!

- Umair

[Eric SIBERT]

> 1. 972 sources, 30 match, 0 conflict, 103 distractors, 38 index

Overall, you have too much sources. You may increase --sigma parameter
or downsample more strongly to target 100 sources.

Or decrease exposure duration if this is for you nanosat tracking.

Do you apply pretreatment? This may help accelerating astrometric reduction.

Eric

[Dustin Lang]

The --downsample option first scales down the image, detects stars, and then scales the detected star positions back up so that the resulting WCS applies to the original image.

One difference from doing your own downsampling: we assume a default uncertainty in the star positions of 1 pixel -- when you scale down the coordinates, that changes the effective error.  You could try solve-field --pixel-error 0.5 if you're using downsampled images, to compensate.

cheers,

--dustin