none
i7-860 CPU RRS feed

  • Question

  • Want to use Kinect v2 on my system using an old i7-860 CPU, a 2010 produced 4-core 2.8 Ghz CPU.

    The Kinect Configuration Verifier Tool state "the system has enough CPU core" so the test is passed, but the system requirements say "Physical dual-core 3.1 GHz (2 logical cores per physical) or faster processor".

    This CPU will be enough or will I experience inconsistent FPS during capture? What's more important, the CPU or the GPU?

    Sunday, February 26, 2017 10:25 AM

Answers

  • The specs are more like Recommended, not Minimum. People have have reported good perf with i5 cpus etc.

    You certainly need a good gpu if there's too much work to be done. But that doesn't mean cpu isn't important.

    I guess you're gonna have to test it yourself.

    You can take a look at the FPS given by the last test in the Configuration Verifier. If you more or less see a steady 27-30FPS it should be ok. Bare in mind though that when faced with lots of people in the fov , the sensor has to do more work on the gpu to account for all of the recognized humanoids and do full skeleton tracking for 6 of them.

    Depends on your needs.

    • Marked as answer by davide445 Monday, February 27, 2017 5:23 PM
    Sunday, February 26, 2017 12:52 PM

All replies

  • The specs are more like Recommended, not Minimum. People have have reported good perf with i5 cpus etc.

    You certainly need a good gpu if there's too much work to be done. But that doesn't mean cpu isn't important.

    I guess you're gonna have to test it yourself.

    You can take a look at the FPS given by the last test in the Configuration Verifier. If you more or less see a steady 27-30FPS it should be ok. Bare in mind though that when faced with lots of people in the fov , the sensor has to do more work on the gpu to account for all of the recognized humanoids and do full skeleton tracking for 6 of them.

    Depends on your needs.

    • Marked as answer by davide445 Monday, February 27, 2017 5:23 PM
    Sunday, February 26, 2017 12:52 PM
  • i7-860 2.8 Ghz Passmark it's in the range of i3-4330 or Pentium G4560, about 35% less than an i5-6600.

    GPU it's a Radeon HD 7950 Boost = R9 280, about a GTX 660 Ti. System have 12 GB RAM, 7200rpm HDD + small SSD for just the OS.

    Will use the Xbox for tracking just one person in a room.

    Since I need to purchase all the setup (Xbox + adapter + USB PCIe card) I wanted to avoid purchasing all the things just to discover the only element I can't update in my rig (the CPU) it's what make the whole ting unusable, so will be glad for any idea.


    • Edited by davide445 Sunday, February 26, 2017 7:17 PM
    Sunday, February 26, 2017 7:15 PM
  • Is this a personal project or some kind of research/work project?

    If it's the latter, ask them for a better development rig.

    Sunday, February 26, 2017 9:52 PM
  • It's a project for a NPO, in a case creating an animated stage set for a ballet school (with mocap needed to cut animation clip creation time, to be used with Unity character and next rendered in a video) and for a science museum (a presentation about dance mechanics).

    So no way currently to have a better rig, just need to understand if will be possible to use what I have.

    Sunday, February 26, 2017 10:17 PM
  • You can record full streams in files using Kinect Studio. You can then playback the file and KS will hijack the service and feed any open application with data from the recorded clip.

    To be honest though, Kinect has a drawback. It was mainly designed to be used in a home setting where users are facing the sensor at all times and they don't move around too much when multiple users exist. There are two things that will pose problems. Kinect can only generate 3D data from 2D input.

    That means that a ballet dancer will never have full 3D data. Whatever parts of the body are occluded, the SDK will try to compensate by inferring,but most of the time it doesn't know what to do and the results are bad. Recently, I read about a studio that made a dance game where they completely discarded Body Tracking and developed their own silhouette tracking which was faster it seems and was more reliable in their use case than the SDK would have been using the Body frames.

    So if a dancer is doing pirouettes or anything like that, the skeleton starts acting weirdly.

    Secondly, it means that when a user is occluded by another, she's effectively lost to the sensor. Imagine you have locked on a specific user, ignoring the others, and either another user passes in front of the locked user or something else happens to occlude the parts of the user that Kinect SDK needs to recognize a candidate for body tracking. The previously active user's tracking id would be lost, and the SDK would replace her with another candidate being loosely tracked, or not at all if there's no one else in the fov.

    I can see this happening if you're recording more than one person.

    Usually mocap uses multiple sensors. But in Kinect v2's case, this means one pc per sensor. And a lot of work in order to stitch the multiple angle recordings in one coherent thing.

    That said, depending on how you record stuff you might be able to it even with one sensor. But that would mean no poses where the angle of the recorded party is greater than what the sensor likes, if you need skeleton data.

    Hopefully someone else can tell you more in terms of hardware compatibility.

    Sunday, February 26, 2017 11:56 PM
  • I'll track just one person in a room, with slow motions will be possible to track 360 movements in your opinion.
    • Edited by davide445 Monday, February 27, 2017 11:33 AM
    Monday, February 27, 2017 8:17 AM
  • Probably not. You might get reliable data while the user is facing the sensor and up to a certain angle. Anything else towards having the back facing the sensor will probably have none or unreliable data.

    Monday, February 27, 2017 5:15 PM