Forum

Author Topic: Accessing RAM  (Read 4337 times)

Terrence Elliott

  • Newbie
  • *
  • Posts: 4
    • View Profile
Accessing RAM
« on: October 09, 2013, 04:54:39 PM »
Hi Guys

Im new to AGISoft but have we recently purchased a UAV and a decent PC (Intel 3.5Ghz 12 Core, 64Gb RAM)

A typical project consists of about 3000 images and Im just wondering two things:

1. Is it more efficient to break the project up into chunks for initial alignment and then merge?
2. AGISoft only seems to access about 6-8Gb of RAM regardless of what else is running - is this normal and is there a way to enable it to access more?

Thanks for any help you can offer.

David Cockey

  • Full Member
  • ***
  • Posts: 209
    • View Profile
Re: Accessing RAM
« Reply #1 on: October 09, 2013, 09:18:00 PM »
My experience is PhotoScan uses as much memory as it needs, up to using all the available memory.

The amount of memory needed depends on settings and can vary greatly. What settings are you using?

Generally alignment requires less memory than creating the dense point cloud and meshing. My strategy with large numbers of photos (up to 600 photos) has been to align all the photos together, and then create duplicate chunks. The bounding box in each chunk is adjusted to cover a portion of the object. The resulting meshes can be merged before exporting if desired.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14813
    • View Profile
Re: Accessing RAM
« Reply #2 on: October 14, 2013, 02:43:00 PM »
Hello helloterrence,

You can check the approximate memory consumption peaks on our Wiki pages:
http://www.agisoft.ru/wiki/PhotoScan/Tips_and_Tricks#Memory_Requirements
It depend on the number and resolution of images and also from the reconstruction settings selected.

As for the breaking into chunks, we recommend to use the single chunk for processing (I assume that with 64 GB you'll be able to reconstruct Height Field model even in High quality for few thousands of images), but if you face memory limits it's better to duplicate chunks after the alignment and optimization steps and reconstruct several parts of geometry separately, but based on the same camera alignment, thus the georeferencing accuracy will not be affected.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Terrence Elliott

  • Newbie
  • *
  • Posts: 4
    • View Profile
Re: Accessing RAM
« Reply #3 on: October 23, 2013, 02:12:00 PM »
Hi Alexey

Thanks for the information.

I managed to Align the 3000 images without too much of a problem. I have added in about 30 GCPs across the site and the residuals are fairly good (less than 40mm).

However upon export of a 5m Grid and subseqeunt comparison of test points (not part of the Ground Control) - Im not getting the kind of accuracy I was hoping for. (ranges from 50 - 500mm). I suspect that I may need a better spread of GCPs next time.

I am wondering though if the Geometry settings I used (Smooth, Medium) have had an impact on the accuracy of the model.

What you be able to outline the expected accuracy of each setting and make a recommendation as to which I should use?

Many thanks.

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14813
    • View Profile
Re: Accessing RAM
« Reply #4 on: October 23, 2013, 02:16:49 PM »
Hello Terrence,

To improve the accuracy we recommend to optimize the camera alignment based only on GCPs (with camera coordinates unchecked in Ground Control pane, and with the marker accuracy set to 1-5mm in the Ground Control pane settings).

Also it is recommended to use 5-20 millions as face count parameter in the Build Geometry dialog, since low values (like 200 000 faces) may result in very rough geometry with huge polygons.

Best regards,
Alexey Pasumansky,
Agisoft LLC