Hello,
I am trying to install tensorflow lite on an EVT RB5 board with the preinstalled Ubuntu Linaro OS.
The software reference guide here:
Says that NNAPI and DSP Delegate supports in Tflite are supported. However, I am unable to compile tensorflow lite with NNAPI enabled because the image is missing the android ndk libraries that are required. What packages/sdks do I neUed to install to build this correctly.
I found that my system is missing preproccesor defiinition: __ANDROID___
which causes tflite to build correctly but disable NNAPI.
After forcing that definition it is missing headers and libraries like android/api-level.h
Is this something that is still in the works? Or am I missing something. I see that the hexagon sdk has the relevant android ndk tool chain, but I do not know of instructions on how to build the android sdk and install on the RB5.
Also after installing the snpe toolkit and running the runtime validator only GPU prerequisites are present.
I would really appreciate an update on this issue or a push in the right direction! Thank you.
Hello mike,
Can you please clarify few things.
You mentioned the board is EVT RB5, May I know where you got it from?
The Linaro OS is either Debian or OpenEmbedded, not Ubuntu. Can you confirm which OS is on the board?
Thank you
Hi,
Yes sorry for misspeaking, the OS is the linaro debian version.
The boards I have is a pre release from ModalAI.
Hi Mike,
The software reference manual that you are refering to on QDN is for the LE and LU linux versions available from Thundercomm using the SDK Manager. I think the upstream versions from Linaro might not be feature compatible with the LE/LU versions.
But I am not sure if that is the source of your error.
Is there any way you can reach out to Linaro or ModalAI for support? |
ModalAI releases their own version of OS for the RB5 flight boards.
https://docs.modalai.com/Qualcomm-Flight-RB5-system-image/
I am not really sure if the Linaro version would work on the drone boards.
I see you have posted on the modalAI forums as well:
https://forum.modalai.com/topic/422/building-tensorflow-lite-from-source-w-nnapi
I think they might be the best resource to support you on this.
Best Regards,
Rajan.