Gravity-aligned Rotation Averaging Example

[htchr]

Hello,

Are you still planning to merge the gravity-aligned rotation averaging into glomap?
If so, would you be able to provide better examples on how to use this?
For instance, where does the relative pose file come from?
The RA doc says that gravity should be [0, 1, 0] when the image is parallel to the ground plane - so this would be when the image is facing the ground right?
What are the axis of the gravity vector?
Pull request 113 also mentions colmap pose priors, would this merge also add support for those?

Thank you for the fantastic work,

[lpanaf]

Hi @htchr, thanks for your interest in the project. For your questions

"so this would be when the image is facing the ground right?", it should be like the image is only rotating around the gravity direction (pitch/roll = 0) and is perpendicular to the ground.

"where does the relative pose file come from?" You can calculate it with poselib, export the results from GLOMAP, or use ones provided by ground truth (for example, 1dsfm dataset provides them)

"What are the axis of the gravity vector?" For this question, can you please elaborate? In general, the reconstruction would result in an arbitrary coordinate with y-axis pointing down.

We are still planning to merge the branch to main; we just need to compose some unit tests before doing so.

[htchr]

Hi @lpanaf, Thanks for getting back quickly, it really helps my work.

To elaborate on the axis, the doc says "The gravity direction g should [ 0 , 1 , 0 ] if the image is parallel to the ground plane" (explained above), does this mean that the vector should be defined as [x, y, z] with +x being left-to-right in the image, +y being downwards, and +z being outwards as relative coordinates for all images?

I look forward to the PR being merged,

[lpanaf]

Hi @htchr, your understanding of the coordinate system is correct. To be more explicit, suppose we can transpose a 3D point from the world coordinate to the image coordinate by RX+ t = x, then the gravity direction should be the second column of the R.

[htchr]

Great thanks @lpanaf