1.4.2 Mesh Generation Problem

[Tunkali]

Hello,

it's now my 4th time to generate a mesh out of my Point Cloud. It takes extreme long and it aborts with a bad allocation error.

Project Data:

1251 / 1426 Cameras aligned
Dens Cloud 17.905.712 Pts low quality (8 hours 38 min)

i7 5820k
64 GB Ram
GTX 970 4GB
GTX 640 4GB

It takes unusual long almost 11 hours on low quality and ends with bad allocation. I remember on earlier projects which were bigger on old Photoscan release... the creation of the 3D model was much faster (around 2 hours).

So can be wrong?

Thank you!

[Alexey Pasumansky]

Hello Tunkali,

Can you please specify the mesh generation parameters used and provide the processing log for the failed operation?

[Tunkali]

Hi Alexey,

Build Mesh
Arbitrary (3D)
Medium
High
Mild
Calculate vertex colors

Processing Log

https://www.dropbox.com/s/bdxd00omw4ud2nn/log_badallocation.txt?dl=0

[Alexey Pasumansky]

Hello Tunkali,

Thank you for additional information.

It looks like the part of the operation required more RAM than possible, even thought the first sub-step has been completed. You can try to enable disk swap and see it it helps (can suggest to extend it for additional 16-32 GB, just to be on the safe side).

The new meshing method is basing on the depth maps, whereas the old one took the dense cloud as a source, so if you compare the time for both approaches - you need to include the dense cloud generation for the older method.

Also is it possible, that you have used Height Field option before and not arbitrary?

[Tunkali]

Hi Alexey,

thats very strange because with earlier version I calculated similar models with more Photos and points with the same system. I will try again and monitor the RAM.

I have system controlled disk swap, but I will try with one manual set swap space (I have two SSDs with > 200GB free space).

Is it right, that I can not anymore create a arbitrary model from sparse cloud and cannot set the filtering method? It's greyed out and set to mild automatically? Maybe it is the reason I used everytime aggressive, so this could be a lot faster.

[Tunkali]

Oh maybe I got the problem. I used the new experimental method. I will try again with the old one!

[Tunkali]

Mesh generation has been succesful in less than 5 minutes :-). So it's something with the new experimental method ;-)

[Alexey Pasumansky]

Hello Tunkali,

Turning on the experimental mesh generation method means that in Arbitrary mode you would be able to generate mesh from the depth maps only (only new method), so if you need to build mesh from the Sparse Cloud in Arbitrary mode - you should disable experimental meshing method.