Visual Consistent Mesh Generation

[Michael_]

I've mentioned this before, but I try againn and again... ;-)
We're scanning insects with delicate structures (wings, setae) and the visual consistent mesh generation (experimantal in late Photoscan versions) was a great tool to avoid holes in wings and thin structure to disappear in the final model. We also üpromoted this option in our publication on the insect scanner (https://zookeys.pensoft.net/article/24584/).
Metashape has "removed" or further developed this approach (reconstruction from depth maps), but the results in our cases are far worse than the late Photoscan-results. So, after every Metashape update I try and then I return back to Photoscan with frustration.
It would be great to have the option in its last Photoscan-version back in Metashape, even as experimental hidden in the advanced settings...
See me crawing on my knees and begging for the return of vcm...

Best regards,
Micha

[Mak11]

What kind of results are you getting now? Holes in the meshes? Unwanted geometry inside/outside the meshes?

Mak

[3DandVR]

I have also difficulties with holes and unwanted vertices in the meshes (from depth maps) with Metashape versions after 1.5.1.
I use the standard version of Metashape 1.5.3.

With MS 1.5.1 there were less holes at the same positions and fine struktures were also better, unwanted vertices inside meshes were also less or not a problem.
Meshing with Photoscan 1.4.5 with Mesh from Dense Cloud was bether, but only at larger and closed struktures like wooden or good struktured walls or similar. At fine struktures like an open roof 1.4.5 was not so good.

With MS 1.5.3 in dense clouds is much more noise from background around edges.

Agisoft please provide a list of tweak funktions, because each try takes days to process with my Models (800 to 1200 Pictures at 24MP).
I have not the time to try and try and try… it is also very frustrating when you don't know how to change these special parameters. Thank you.

[Alexey Pasumansky]

Hello,

If you would like to switch back to experimental meshing method (visibility consistent) from the version 1.4, then you should create the following Tweak via Advanced preferences tab and set its value to False:

Code: [Select]

main/mesh_tvl1Then just use the Depth Maps as a source for the Build Mesh operation.

Note that the old experimental method has not been developed and optimized since version 1.5.0 has been released, so the processing time and resource consumption could be suboptimal.


As for the current Depth Maps based approach, in the 1.6.0 release (and hopefully before that in some pre-release update) the method should become out-of-core, i.e. would not require excessive amounts of RAM, like it could happen now for large projects and Ultra/High quality used.

[Mak11]

Alexey,

Good to know that we can still use the older visibility consistent method via this tweak and that out-of-core processing should be comming soon. Hopefully also improvement to the quality/integrity of the generated mesh..

PS: please create a page listing all the useful tweaks.

Cheers

Mak

[JRM]

Great to hear about the out-of-core improvement Alexey ! Could Agisoft updates its memory requirement documentation for 1.6 ?

[Michael_]

Alexey,

thanks for the Tweak. I will upgrade to Metashape and test it immediately ;-)
Results are more important than time consumption to us.

Best,
Micha

[Michael_]

Talking about Tweaks - is there also a Tweak to disable filtering of depth maps?
We had best results without filtering using dense cloud-approaches, but "mild" filtering seems the only option with the new meshing approaches

Thanks,
Micha

[Alexey Pasumansky]

Great to hear about the out-of-core improvement Alexey ! Could Agisoft updates its memory requirement documentation for 1.6 ?

The out-of-core mesh is still in development, but we are planing to make it fit 16 GB RAM limits.

[wojtek]

Great to hear about the out-of-core improvement Alexey ! Could Agisoft updates its memory requirement documentation for 1.6 ?

The out-of-core mesh is still in development, but we are planing to make it fit 16 GB RAM limits.

Will this apply to point cloud filtering as well?

(And meshing from a point cloud source)