Bundle adjust parameter ranges and/or cost function?

27 views
Skip to first unread message

Rainey Aberle

unread,
Jan 28, 2025, 4:35:25 PM1/28/25
to Ames Stereo Pipeline Support
Hello!
Is there a list of the potential parameter values for bundle adjust somewhere in the documentation? I haven't been able to find it. I'm wondering specifically what the ranges of --camera-position-weight and --tri-weight values are and how these are weighted with --heights-from-dem and --heights-from-dem-uncertainty. 

Also, is there a way to see the specific cost function being used? I see a link to the generic function, but not with all of the specific parameters. 

Thank you! 

Oleg Alexandrov

unread,
Jan 28, 2025, 4:44:24 PM1/28/25
to Rainey Aberle, Ames Stereo Pipeline Support
The full list of options of bundle_adjust is in https://stereopipeline.readthedocs.io/en/latest/tools/bundle_adjust.html#command-line-options.

The  --camera-position-weight option has a complicated explanation and is not easy to set right. 


The implementation of --heights-from-dem is described here: https://stereopipeline.readthedocs.io/en/latest/bundle_adjustment.html#heights-from-dem

The --heights-from-dem-uncertainty is in units of meter. Any measured discrepancy between a point being optimized and its initial value is normalized by this to convert to units compatible to pixel units. 



--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/07dc630d-5ca7-482e-91c0-c9de12f7933en%40googlegroups.com.

Rainey Aberle

unread,
Jan 28, 2025, 4:57:24 PM1/28/25
to Ames Stereo Pipeline Support
Hi Oleg,
Great, thank you! I'll play around with the --camera-position-weight. 

Rainey Aberle

unread,
Feb 11, 2025, 4:40:18 PM2/11/25
to Ames Stereo Pipeline Support
Hello! 
I have another bundle adjustment question when you have time. Is there a way to provide a list of images associated with each camera to optimize their orientations? I see that you can do what I'm looking for with the intrinsics, where the same intrinsics are applied to all images for each camera (e.g., 12.2.2), but I don't see an analogous solution for extrinsics. When I provide the --camera-list without the --solve-intrinsics parameter, it returns an error because I don't have the initial intrinsics parameters (which I don't want adjusted). For reference, I have ~70 SkySat triplet stereo images and the orientation varies quite a bit for a given camera, resulting in some artifacts between frames in the final DEM. Keeping orientations consistent might help. 

Thanks again!

Alexandrov, Oleg (ARC-TI)[KBR Wyle Services, LLC]

unread,
Feb 11, 2025, 5:50:09 PM2/11/25
to Rainey Aberle, Ames Stereo Pipeline Support
>Is there a way to provide a list of images associated with each camera to optimize their orientations?

It depends on what you mean by "camera". Normally, there are N images and N cameras, so one camera per image. You should be able to use the --fixed-image-list option to fix poses of desired images.

If you are talking about sensor properties, then a single sensor can acquire multiple images, each with its own camera orientation, but all these will share the same intrinsics.

>When I provide the --camera-list without the --solve-intrinsics parameter

Interesting. You should be able to provide --image-list and --camera-list without any mention of intrinsics. If no luck, you can share the exact command you use.



From: ames-stereo-pi...@googlegroups.com <ames-stereo-pi...@googlegroups.com> on behalf of Rainey Aberle <rainey...@u.boisestate.edu>
Sent: Tuesday, February 11, 2025 1:40 PM
To: Ames Stereo Pipeline Support <ames-stereo-pi...@googlegroups.com>
Subject: [EXTERNAL] [BULK] Re: Bundle adjust parameter ranges and/or cost function?
 
CAUTION: This email originated from outside of NASA.  Please take care when clicking links or opening attachments.  Use the "Report Message" button to report suspicious messages to the NASA SOC.



Reply all
Reply to author
Forward
0 new messages