Model File Types - Conversion

Help
#1

I’m looking for explanation to further my understanding.

I am playing with a Silicon Labs Thunderboard EFR32BG22. It appears similar in many ways to the Sense2, but because of these differences I make no assumption that the Edge Impulse documentation or any binaries (*.bin) I deploy should work for this model.

What I have working on the EFR32 is the example solution, “TensorFlow Lite Micro - Hello World”, provided through Simplicity Studio. Further, I can read serial output over USB with Putty.
.
.
What I would like to understand is how to work with and convert models between formats…

The example solution implements a model exported in C (.cc) and appears to be a single array (pictured). Contrast that with the C++ compiled model using Edge Impulse (.cpp) and what I have is a lengthy file with a multitude of #includes and other tensorflow specific definitions.

example model
example model798×372 13.7 KB

.
.
So, my challenge is, how can I keep the existing file structure provided by Silicon Labs’ example solution and “drop in” an Edge Impulse generated model? What terminology describes what I want to accomplish?

#2

Hi @comlogical,

We also provide an open source solution for the Silicon Labs Thunderboard that you can drag and drop your deployed Edge Impulse Studio C++ library into here: https://github.com/edgeimpulse/firmware-silabs-thunderboard-sense-2

The README of this github repo describes which folders to replace to quickly update the included Tensorflow model with the one who have trained in the Edge Impulse Studio. Then you can compile the firmware similarly to the SiLabs solution in Simplicity Studio.