Forum

Author Topic: Memory unload  (Read 4517 times)

RHenriques

  • Full Member
  • ***
  • Posts: 228
    • View Profile
Memory unload
« on: April 09, 2019, 12:04:13 AM »
There seems to be a few problems in memory unload, in Metashape 1.5.2 (7838), after finishing a few steps. Lot's of RAM stays locked within Metashape even if it is not needed anymore. This seems to be causing a few "out of memory errors" in further steps. In a complex project, with about 1117 photos, even using RAM cleaner apps, from 64GB only 7 GB are free. If we quit and open Metashape again, 57GB will be free with the same features opened. Even after visualize a dense 3D model, 48GB will be free. However, if a workflow is followed, in the end, after all processes finish, the RAM is still blocked. If we open a project and keep Metashape opened, memory keeps leaking and adding to the app, even without doing nothing.
Cheers
« Last Edit: April 09, 2019, 01:11:49 AM by RHenriques »

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 15117
    • View Profile
Re: Memory unload
« Reply #1 on: April 09, 2019, 12:50:37 PM »
Hello RHenriques,

Which OS version you are using and which steps are followed, so that we could try to reproduce the problem on our side?
Best regards,
Alexey Pasumansky,
Agisoft LLC

RHenriques

  • Full Member
  • ***
  • Posts: 228
    • View Profile
Re: Memory unload
« Reply #2 on: April 10, 2019, 02:45:50 AM »
Hi Alexey

The OS is MacOS High Sierra. The steps are Align; Build Dense Cloud; Build Mesh; Build DEM and Build Ortho. After the end, from the 64GB available, 57GB will be locked by the applications and are not released until we quit Metashape.
However we do not even need to follow these steps. If we open a completed project (with all the elements already calculated), with the sparse point cloud as default view and let Metashape opened without doing nothing, the memory keeps on growing (or leaking). You will easily see this effect by checking the Activity monitor build in MacOS. Yesterday I've opened a project that took initially 1.35GB of RAM and, in 18 minutes without doing nothing, went to about 18GB and kept pilling until I've quitted Metashape. I have a recorded screen video that I can send if you want.
Best Regards.

PS. It gets worst. If we go to 3D model view in any project, above about 30 000 000 faces, Metashape gets very slow and it impossible to change to sparse point view or anything else without force quit the app. If we do so, Metashape gets locked even with force quit and the only solution is to force start the computer. Clearly this version has a memory leak problem into MacOS.

PS 2: The problem of huge latency seems to be linked to 3D Model Solid View. Using this view Metashape gets virtually frozen. If we change the 3Dmodel view to Shaded or Wireframe things behave as expected. In solid view it's impossible to do anything like change views, select and cut, etc. Metashape becomes completely unresponsive. This seems to be a different issue from the memory leak I talked about above.
« Last Edit: April 11, 2019, 04:47:02 AM by RHenriques »

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 15117
    • View Profile
Re: Memory unload
« Reply #3 on: April 11, 2019, 03:28:04 PM »
Hello RHenriques,

Model view display is using graphic card virtual memory to render the mesh, for example. Which GPU is used for OpenGL rendering on your machine and which amount of VRAM does it have?
The requirements for rendering in different modes are different, for example, the model rendering in Solid view mode would require about 3.5 times mode VRAM than in Shaded view mode.
Best regards,
Alexey Pasumansky,
Agisoft LLC

RHenriques

  • Full Member
  • ***
  • Posts: 228
    • View Profile
Re: Memory unload
« Reply #4 on: April 12, 2019, 05:28:37 AM »
Hi Alexey

The GPU it's an Nvidia GTX 1080TI with more than 11GB of VRAM. It should not be a problem.
However, the slow 3D model display is not my main concern. The memory pilling problem (even without doing nothing) seems to be the source of some problems, mainly Heighfield Surfaces that do not finish (they finish without problems in PS 1.4) and memory errors while using the new reconstruction method based on depth maps.
Cheers