Reconstruction fails after switching from Photoscan to Metashape

[yowich]

Recently, after switching from Photoscan to Metashape, I have had multiple scans that failed to reconsturct.
So I tested Metashap and Photoscan with one of my datasets.

As you can see from screenshots attached below, Metashape miserably failed to reconstruct while Photoscan did it nicely for the same set of pictures.
I used the same reconstruction options for Photoscan and Metashape.
Why does Metashape fail while Photoscan does it well?
How can I address this issue?

Attached are screenshots for the results of reconstruction. P and M are Photoscan and Metashape, respectively.

[Alexey Pasumansky]

Hello yowich,

Can you please specify the version of Metashape used (including the build number)? If possible, please also send these two projects with the alignment results only (in PSZ format) to support@agisoft.com.

[Mak11]

Yowich,
Do you mean the alignment process when you say "reconstruction"?
If so then try to disable generic pre-selection before alignment. As I posted in an other thread it looks like the speed optimizationsto the generic pre-selection option introduced in one of the Metashape pre-releases is not behaving correctly with some datasets.

Mak

[yowich]

Hello yowich,

Can you please specify the version of Metashape used (including the build number)? If possible, please also send these two projects with the alignment results only (in PSZ format) to support@agisoft.com.

I just sent an email with psz files you asked.

I don't exactly remember what version and build I used for Metashape alignment because I had to uninstall Metashape in order to rollback to Photoscan. But, considering the date I aligned it and the Metashape installer that was saved in my download folder it might be version 1.5.1 Build 7618 (64-bit).

Thanks!

[yowich]

Yowich,
Do you mean the alignment process when you say "reconstruction"?
If so then try to disable generic pre-selection before alignment. As I posted in an other thread it looks like the speed optimizationsto the generic pre-selection option introduced in one of the Metashape pre-releases is not behaving correctly with some datasets.

Mak

Thanks for the comment, Mak.
Yes, I meant alignment by reconstruction.
Normally, I don't use the generic pre-selection as an option.

[Alexey Pasumansky]

Hello yowich,

Thank you for sending the projects, we'll check them.

As a quick comment: I have selected all NA (not aligned) cameras in the first chunk ("Down") and run Align Selected Cameras operation. It result in a point cloud similar to the one you have in PhotoScan. For the second chunk ("Up") I've reset the alignment for all the cameras and realigned them - the result is also similar to older version result.

What I can suggest, it to limit key point and tie points during the alignment and check, if it helps. I recommend to use 40 000 - 60 000 for key point limit and 4 000 - 10 000 for tie point limit.