Takes long time for meshing

Hello RC community,

i´m really happy to make the decision to switch to RC tool for photogrammetry heritage and other stuff.

But now i´m running into a problem with my current project.Meshing takes a long time.

  • Input files: 2700 images taken by DSLR / 12 MB per image
  • Alignment: Successful was able to align 2590 images out of 2700
  • Meshing Preview Mode: Takes a few minutes and look ok
  • Meshing Normal Mode: First time up to 5 hours.
  • RC can´t show the mesh afterwards, due to less VRAM (using a 10GB VRAM)
  • Export of mesh was possible and got a 8M Mesh, which was quite to heavy for my usage in Blender

2nd run Meshing with smaller Region:

  • Meshing 50-60% takes about 30min, afterwards it slows down drastically… and feeds all my 64 GB RAM
  • Finsh time for meshing jumps then hours and hours.
  • What do i do wrong here and how can speed up the process for meshing? I´ve read in this forum that some people rende reasy 7000k or more.

My Workstation:

  • AMD r9 5900x
  • 64 GB RAM + Additional 64 GB Virtual Memory
  • Nvidia Gefore 3080, 10 GB VRAM
  • M2 SSD with 2 TB -> Project and Cache folder is stored there

Question:

  1. How can i speed up the time for meshing?
  2. Are there any tipps and tricks for best setup for fast workflow?

 

Thanks for any help.

Best

Mcm1003

I have to add that RAM usage is like this:

20 GB in the first 40-50 % of meshing,

45 GB in the 60 % of meshing

62-64 GB at 70%  of meshing - and slown down of the proces, because it´s all the physical RAM.

Why is RAM Usage in the last third so increasing?

 

Reconstruction mainly uses GPU so adding another graphics card to your system would probably be the best. You will not see a great return on investment with a third GPU. Another thing to look at would tweaking your reconstruction settings. 

Obviously a lot of your meshing time is going to be spent creating the depth maps, but you can run some tests to see if you can speed up your meshing times using detail decimation factor in the advanced meshing options.

For example, normal detail model. Depth maps took 1m37s meshing 2s
High detail model. Depth maps took 7mins 11s, meshing 9s (although it is definitely better, a lot of the extra detail is just noise)
The same high detail depth maps meshed in 4 seconds with detail decimation set to 1.3 

Obviously if you wanted to try this it’s best to run some tests on some small reconstruction regions to  hit the sweet spot. Also check your ratio of depth map calculation time to meshing on your current model to see if it would have been worthwhile.​​​​

The other thing you could look into is essentially doing your own custom image downscale to increase the depthmap calculation time. Normal reconstruction uses 2 and high does not downscale, but you could use the folder method of image layers and put your images with a custom downscale into a folder called _geometry and the originals into a folder called _texture. That way you can slightly downscale the images for faster depth maps and keep the same quality images for texturing, and use High detail reconstruction. Bear in mind I have not tested this, but it should work.

Hello Ondrej,

thanks for quick reply. Regarding graphics card… yes, it´s hard to get graphics cards these days. Will it also do an old geforce980?

Also i was reading your comment and honestly speaking i´m not quite sure if i got everything correct. Some functions i just can´t find. Using Version RealityCapture 1.2.0.17385 RC

Have to add, that i was merging two components to one.  Can this result in long calculations times for the mesh?

  • What´s the best setting for detail decimation factor? Right now it´s set to 1.000000
  • Where can i find the option to change the depth mask?
  • Where can i check your ratio of depth map calculation time to meshing?

I also checked a tutorial from CyArk (https://www.youtube.com/watch?v=JeuRd9OquVE&t=63s) to understand Reconstruction workflow better. Even here i can´t find some functions.

(just read some option (Force single part mode) was removed because of better computing process. RC Comment on youtube)

 

 

Hi mcm1003,

as that card has CUDA, then it should work.

There were some changes in UX, so there could be differences, when you are watching older tutorials.

The time of mesh calculation differs and it depends on the used dataset, it could take some hours with your setup.

Best setting for detail decimation factor is pre set 1, but if you want to speed up this process, you can change this number to higher value. But you will lose some details. It is better to test, which value will be best for you on a smaller region.

Depth mask is under Mesh model tab/Export

The whole process of model calculation is written in the console view.

 

 

Hi 374823146340,

I see. I´m still learning more and more about RC. Thanks for your tipps.

So also learned that i´can pause the reconstruction and start from later on. That might also help. Will give it another try for my model.

Even if my current setup is good for me, it´s never enough to have more power in 3D. So will also try to use a 2nd card in another project.