View my account

Realsense SDK2.0 image alignment

Comments

7 comments

  • MartyG

    Hi 819765928 There does not need to be a value inside the brackets of wait_for_frames(), as this instruction just means "wait until a complete frame is received instead of instantly processing frames".  This helps to prevent the computer's processor being used heavily. ("maxing out " at 100% usage).

    Aligning is also a processing-intensive activity.  So if the Windows computer has a means of aiding the processing, such as an Nvidia graphics chip that can provide "CUDA" support for doing computation on the graphics chip instead of the processor chip, then this may make a noticeable speed difference.

    0
    Comment actions Permalink
  • 819765928

    Hi MartyG,I find that on the ubuntu18.04 of i7 8700k cpu, it takes about 7ms to use the python SDK for alignment, and 20ms for the C + + SDK. I'm puzzled about this.

    0
    Comment actions Permalink
  • MartyG

    If you are building librealsense with CMake, you may get a performance boost if you include the term -DCMAKE_BUILD_TYPE=release if you are not using this term already.

    0
    Comment actions Permalink
  • 819765928

    Hi Marty, thanks for the reply!I tried this method, but it didn't help much.

    0
    Comment actions Permalink
  • MartyG

    I did further research on align processing speed but did not find much other than the advice already given, to use CUDA if your computer supports it and to build in Release mode.

    Another way to gain performance during alignment is to add to the CMake build statement this further term:

    -DBUILD_WITH_OPENMP=true

    This allows multiple cores on the computer's processor to be used when performing depth to color alignment, reducing latency (though the processor will be busier).

    0
    Comment actions Permalink
  • 819765928

    Hi MartyG,

    thanks for the super fast respnse.One of my algorithmic engineers wrote a new image alignment, which seems to solve this problem.But I'll try your advice.Thanks

    0
    Comment actions Permalink
  • MartyG

    Great news - thanks for the update!

    0
    Comment actions Permalink

Please sign in to leave a comment.