Mesh from Depth Maps - False Surface

[SP74]

Hi,

Using the latest build of Metashape, I am getting a false dark surface as per the attached image, when processing a mesh using depth maps as the datasource.  As my project is urgent I reverted to Photoscan 1.4.4, no issues were found. 

The mesh is a small region from a much larger project (~14300 photos).  I extracted a low resolution dense cloud over the entire project area and no issues we detected in the area of this mesh.  The project aligned well, with ppk camera locations and plenty of GCPs/Check points.

Cheers,

Stu.

[Alexey Pasumansky]

Hello Stu,

Thank you for reporting.

Are all the cameras facing vertically down in this project or there are oblique or even ground-based cameras?

I think as a workaround you can reduce the bounding box, by shrinking it in the vertical dimension and rebuild the mesh in 1.5 just re-using the depth maps.

[SP74]

Hi Alexey,

The imagery as all Nadir, shot with a wide angle (20mm FL) lens and Sony a6000 camera.  However we will be acquiring higher resolution nadir & oblique imagery in the near future.  I downgraded to Metashape 1.5.0 (build 7492) and can confirm the issue is still present.  I have adjusted the region bounding box height as suggested and will advise the outcome.

Cheers,

Stu.

[SP74]

Hi Alexey,

Reducing the vertical extents of the bounding box resolved the issue with the mesh generation.  However generation of a Tiled Model is also effected.  The Tiled Model was also generated from Depth Maps.  I have attached the profile views of the single file mesh and the tiled model for comparison.

Cheers,

Stu.

[Alexey Pasumansky]

Hello Stu,

I can send you a pre-release of the version 1.5.2 that should trim such kind of extra surfaces that you were observing in the original post.

As for the latest message, is it possible that the screenshot represent the lowest (less detailed) zoom level? Maybe you can share the exported TLS file with us?

[SP74]

Hi Alexey,

Please find attached the link to the tiled model as requested,

http://astron-geodata-delivery.s3.amazonaws.com/Test_Model.zip

Cheers,

Stu

[Alexey Pasumansky]

Hello Stu,

I could assume that the problem is related to the "zero" face count used for the tile model generation. We'll try to check it on similar size projects, but meanwhile can suggest to re-build the tiled model using valid fact count number (or one of the pre-sets).