Closed sbr-dev closed 3 years ago
Without knowing anything about your company tools we are just guessing at the issue. One area to look at is memory transfer between the GPU and CPU.
Hi @qm13 Thanks for your reply !
Indeed, it might be difficult for you to guess what happens without knowing the setup. I'm just asking for some advice or if something is wrong in the way I handle the kinect .
Datas transfer could be an issue, but neither the CPU nor the GPU are overloaded (25% and 80% respectively) The GPU bandwith usage is also in the usual limit.
Could it be related to the OpenGL context? What happens if I start the Kinect on the main thread and then I execute only some code on a dedicated thread? Is there any autonomatic context switch?
What's the data worflow of the kinect ? Is it Kinect -> CPU -> GPU ?
@aspioupiou the perf tools especially GPU tools often do not take into account DMA orchestrated data transfers.
The data flow is Kinect -> CPU -> GPU
Hi all,
This is more of a question than a bug report.
We integrated the azure kinect SDK for headtracking purpose. The integration works like a charm with our company tools.
However, we spotted a huge performance drop when connecting our libraries with 3rd party softwares. After some benchmark, we spotted the bottleneck:
ka4bt::tracker::pop_result();
is the culprit. It seems this call freezes for some time the application and causes the FPS drop.timeout
parameter had no effectAm I missing something. What does it still hangs the app when called on a dedicated thread ??
Cheers,