Hello,
I’m trying to merge 2 components of 8994 and 6016 cameras. In total 15010 cameras.
Images are acquired with Mavic 3E RTK and are aligned with normal settings and 20/40K max features.
Unfortunately it takes too long and the Matching is not moving over 10% for the last 24 hours.
The question should i leave it another 24 hours or is it a waste of time?
I never had such issues with older versions, I have 128GB RAM and it’s not even using it all.
After 40 hours of waiting to finish merge I gave up.
I just don’t understand why there is no obvious feedback when there are issues / or problems with merging.
Also I found out that aligning on the new Feature detection quality: high it takes longer to align and when you export components it takes like 100 times more to export the component.
I used to work with components in the older versions without any issues.
Right now I’m working on this project with this workflow:
And more specifically using overlap cameras. I will post more information when I’m done with all components and final alignment.
Hello Dimitros,
unfortunately there is a bug when the component has more than 65M reprojections (which is your case). To get fewer reprojections try to set the Max features per image value to a lower number.
(post deleted by author)
Thank you for the feedback.
Some sections include dense vegetation, which reduces repeatable features and tie points. To achieve a stable alignment, I had to increase the max detected features/tie points to 40K or more.
Will this bug get patched next version?
All my cameras are about 37K. I’m trying the “Merge using overlaps” workflow with 5 components.
Do you think this will work? or the bug will still prevent the merging? Maybe try an older version of RS to merge components?
Update: I tried “Merge using overlaps” it didn’t work. Probably I’m reaching the threshold of 65M reprojections.
The bug should be solved in the next release.
I’m posting this in case anyone else runs into a similar problem.
I downloaded RealityCapture (Scan) version 1.4.2 and processed my dataset there instead of the latest release.
I had around 40,000 photos, which I split into five separate parts, each representing a distinct region of the scanned area. The total area was quite large, so this segmentation was necessary even from the scanning stage.
For each region:
I identified the overlapping cameras and included them in both adjacent parts, ensuring proper overlap between datasets.
Each part contained 6,000–11,000 photos and took roughly 3–5 hours to align.
After aligning each part individually, I exported the alignment registration for each.
Then I merged the components one by one:
Combined the first and second components
Exported the new merged component
Repeated the process, merging that with the third, and so on
In the end, I had a single, correctly merged component without alignment errors or misalignments — as long as each part was well aligned beforehand.
After that, I continued with the normal workflow.
My Takeaways
Stability: Version 1.4.2 was noticeably more stable than the latest release.
Accuracy: The estimated values (e.g. alignment precision, reprojection errors) were much closer to reality, whereas the latest version produced inconsistent results.
Speed: Alignment was slightly slower, but overall more reliable for users familiar with RC’s workflow.
Export performance: The export registration process was significantly faster — in version 1.4.2 it completed in just a few minutes, while in the latest version (high mode) it often took 30+ minutes per alignment.
Hopefully this helps anyone struggling with large datasets or instability in newer releases of RealityScan.
