When looking into /opt/nvidia/vpi/vpi-0.1/samples we couldn’t see an example for writing our custom code for the PVA compiling and running it.
We could see in CMakeList.txt find_package(vpi 0.1.0) which we guess uses the pre-compiled library libnvvpi.so.
Can anyone please forward us to an example for writing custom code for this accelerator?
Hi Gabi, the Xavier’s Vision Accelerator doesn’t run custom user-defined code (i.e. that implements custom routines/functions). It is used through the vision primitives exposed by VPI.
Or is your question how do you link your application against libnvvpi for using these computer vision primitives?
My question was about writing custom code to PVA/VPI.
Will it be possible in next JetPack releases?
Please setup your device with JetPack4.3.
There is a new API called VPI, which can control PVA directly.
You can find some VPI sample in /opt/nvidia/vpi/.
Gabi, I think there is some confusion regarding which level of custom code you are asking about - e.g. programming the PVA directly (to implement custom functions that are not included in VPI), or to invoke/control the PVA to run vision functions from VPI. The later scenario is what is supported.