Forum

Author Topic: Balancing dense cloud specs vs mesh specs  (Read 2054 times)

rossnixon

  • Jr. Member
  • **
  • Posts: 92
    • View Profile
Balancing dense cloud specs vs mesh specs
« on: August 17, 2018, 06:32:46 AM »
I was producing an orthophoto from about 260 aerial photos.
I ran out of memory creating the mesh from the dense cloud.
I lowered the mesh face count down to low (2.7 million), but still ran out of memory.
Then I disabled interpolation. This then worked.

1. What is the effective of disabling interpolation?
2(a). Should I have used Medium Quality, instead of High when building the dense cloud?
2(b). Would that affect how many faces the mesh could have?
2(c). If so, could I (somehow) thin down the dense cloud a certain amount?
3. What does Calculate vertex colors do; and does this take much memory?


Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 13211
    • View Profile
Re: Balancing dense cloud specs vs mesh specs
« Reply #1 on: August 19, 2018, 07:49:26 PM »
Hello rossnixon,

Would be helpful, if you can provide all the processing parameters used up to the meshing stage, specify the number of points in the dense cloud and the source image resolution.

If you are building mesh using Arbitrary option, then the face count number has small effect on the memory consumption. In Arbitrary the mesh is generated in the highest possible resolution and then is decimated to the face count value. Vertex colors also has small effect on the memory consumption.

In PhotoScan currently there are no instruments to decimate the dense cloud other then using lower quality settings. However, you can export the dense cloud to the external application, reduce the number of points and import the cloud back to PhotoScan.
Best regards,
Alexey Pasumansky,
Agisoft LLC

rossnixon

  • Jr. Member
  • **
  • Posts: 92
    • View Profile
Re: Balancing dense cloud specs vs mesh specs
« Reply #2 on: August 20, 2018, 02:10:44 AM »
Hi Alexey,
Details are all in my PDF report here www.dropbox.com/s/aum5ghgoccz60ws/Kitchener_Park_Agisoft_report.pdf?dl=0
Some of the values you will see are
254 images: 5472 x 3648 px
Key/Tie point limits: 80,000 / 20,000
Tie points: 471,998 of 699,990
Dense cloud points: 122,457,855

Windows 10 with 16GB of RAM.

rossnixon

  • Jr. Member
  • **
  • Posts: 92
    • View Profile
Re: Balancing dense cloud specs vs mesh specs
« Reply #3 on: August 29, 2018, 02:04:33 AM »
Hi Alexey,
Any comments on the values I used?
Also, what effect would disabling interpolation have on the orthophoto?

jwoods

  • Newbie
  • *
  • Posts: 23
    • View Profile
Re: Balancing dense cloud specs vs mesh specs
« Reply #4 on: August 31, 2018, 03:57:33 PM »
Someone chime in if I'm wrong, but it is my understanding that choosing the smaller face counts in the options dialog when building the mesh doesn't actually reduce the amount of processing the system does.  Selecting a lower face count is just adding a decimation step automatically at the end of processing. 

Say your high face count is 10 million, and your medium is 5 million.  If you select the 5 million, it will generate the entire 10 million face mesh, then decimate at the end down to the 5 million.  That's why it's always best to go ahead and run the high face count for the mesh, that way you can do whatever editing you need to do, first, then smooth and decimate as you need.

stihl

  • Sr. Member
  • ****
  • Posts: 408
    • View Profile
Re: Balancing dense cloud specs vs mesh specs
« Reply #5 on: August 31, 2018, 05:29:06 PM »
Jwoods that's very much correct, unless Agisoft changed this parameter during an update.

If you're running out of RAM during the Build Mesh stage and you're unable to add more RAM or virtual memory then it'd best to duplicate the chunk (for safety) and rerun the dense cloud at a lower quality level. From what I remember, going down one step in the quality level means a factor 4 reduction of generated dense points. With this info you can get a rough sense of the amount points you'll end up with after the dense cloud stage.