Failed parts on large reconstruction

Development Asset Creation
,
1

Hello,

I milled a large reconstruction (2K inputs) on a High settings.
After 2 days the whole thing succeeded with over 1BLN polys, but I got information that some parts were unsuccessful and indeed there’s a hole in the model. Moreover, RC fails to decimate/simplify such mesh every time regardless of settings now.
I wonder what’s possible to save this effort?

2

What I also did was I filtered out the region which had failed parts and tried filtering afterwards.
Filtering went through smoothly, but simplify fails.

3

Hello Patryk,
this seems to be the same problem as here: Bug 892: New Simplify Algorithm Fails With Large Projects

Do you use new simplify algorithm? To check this, go to: RECONSTRUCTION/ Settings / Advanced / Use legacy simplify alghorithm and it should be set to False.

You can also use the old simplification alg. but I don’t think you would have enough memory for it.
Btw. what are your PC specs?

Unfortunately we don’t have a good solution for this right now.
As I can see it now, your options are:

  • Use external tool to simplify your model
  • Split your model into several small parts. Simplify them separately (probably with the old simplif. alg.) and merge them afterwards in some external tool.
4

Kruzma wrote:

Hello Patryk,
this seems to be the same problem as here: Bug 892: New Simplify Algorithm Fails With Large Projects

Do you use new simplify algorithm? To check this, go to: RECONSTRUCTION/ Settings / Advanced / Use legacy simplify alghorithm and it should be set to False.

You can also use the old simplification alg. but I don’t think you would have enough memory for it.
Btw. what are your PC specs?

Unfortunately we don’t have a good solution for this right now.
As I can see it now, your options are:

  • Use external tool to simplify your model
  • Split your model into several small parts. Simplify them separately (probably with the old simplif. alg.) and merge them afterwards in some external tool.

Machine: 72 cores, 128GB, 4x GTX1080

  • I understand that the error I am getting is NOT related to failed parts? It’s just about the size?
  • The Model is 1BLN. So I guess old algorithm won’t cut it. Any idea to estimate how much RAM it costs or what’s the largest mesh size I can mill on 128GB RAM?
  • It’s unlikely any other software would handle 1 BLN. Pointclouds - yes, but probably not meshes.
  • From the other post I understand it’s a bug that’s pending a fix?
  • Simplification is only one problem, the other was the fact that some parts failed (even though the entire reconstruction succeeded and I have holes now). Is there any way to make RC only retry failed parts? Or I have to manually set a reconstruction region, reconstruct the smaller part and merge it?
  • It’s great we have filtering in RC, but here are two other feature requests - easy to implement and would help tremendously.
    A) Filter could work with an option of splitting a model into 2 parts (instead of always deleting the other part)
    B) A very simple merge option to merge split models or partial reconstructions would be a great help.
    Many thanks for your further help!
5

Hi Patryk Kizny

change the RECONSTRUCTION REGION aka BBox size by 10-15 % and change its rotation by 10-15 deg + as mentioned in FB chat, use TASK MANAGER and set AFFINITY to max 16 cores for RC and recon it again, depth maps are stored so it will go pretty fast with recon. then you get properly reconstructed model.

6

Thanks,

So using the legacy simplification worked. Even w/out any serious strain on RAM.
I think I’ll try to reconstruct the missing region and merge outside of RC.
Spending 1.5 days on the reconstruction again does not seem to be worth it.