Target is imx8qxp
Checked after Enabling opencl.
Command: /usr/share/OpenCV/samples/bin/example_dnn_object_detection -scale=0.00392 -rgb -device=7 -target=1 -zoo=models.yml yolo [provided all dependency files, running fine with -target=0 (CPU)]. but not running if -target is set to 1/2.
Please let us know the behaviour or limitation of software/hardware?
As Vanessa mentioned, OpenCL is currently not supported in the L4.14.98_2.0.0 and L4.14.78_1.0.0 Yocto configurations. This is a software limitation documented in eIQ app note UM11226 that we expect will be resolved in a future release.
Could you please let us know in which existing version for yocto, opencl is working?
We have some demo for machine learning which is working fine with cpu (but slow response and lower frame rate), but not running with opencl/opencl_fp16.
I have enabled the opencl packages and watched the behaviour says, process is taking much gpu memory continously and sometimes process gets terminated as memory gets full.
Is it a way to reduce the time for machine learning while object detection with opencl (gpu utilisation), but there is only one gpu core and 4 shaders, does it perform well?
Let me know about whether opencl_fp16 support? Because it may use sometimes lower memory but faster response for frame rate?
Sandeep K Rai
Get Outlook for Android<https://aka.ms/ghei36>
It is not yet specified when OpenCL support for ML will be available. We continue to work on this enablement but at this time I cannot provide any further guidance on the timeline or anticipated kernel version.