Yocto imx8m tensorflow-lite

arno_0 — Thu, 06 Aug 2020 14:28:57 GMT

I would like to build C++ code and use tensorflow as a library. What do I have to ad to the Yocto image recipe to have header and required libs in SDK?

For the rootfs it seems

that IMAGE_INSTALL += " \
tensorflow-lite \

does the job to have the examples in it (no lib, no headers).

But how about the SDK? I need headers and libs. The idea was about:

SDKIMAGE_FEATURES_append = " \
staticdev-pkgs \
tensorflow-lite-dev \
tensorflow-lite-staticdev \
"

But I can't compile label_image (just as an example) because of undefined references. What do I miss?

Also a bit strange. Building the libtensorflow-lite.a out of Yocto results in 250MB lib, but building it external as described here (Build TensorFlow Lite for ARM64 boards ) results in a just 8MB lib ?

Any help welcome!

Re: Yocto imx8m tensorflow-lite

jamesbone — Thu, 06 Aug 2020 15:53:33 GMT

Hello,

I recommend you to follow the Getting Started, that we create for the eIQ, which it is our software based on Tensor Flow, to add Machine Learning to our i.MX8 Devices. Here is the link:

https://community.nxp.com/community/eiq/blog/2019/07/01/getting-started-with-eiq-software-for-imx-applications-processors

topic Yocto imx8m tensorflow-lite in i.MX Processors

Yocto imx8m tensorflow-lite

Re: Yocto imx8m tensorflow-lite