Uploading own model to use CMSIS-NN


I’m using a development board with Zephyr OS.
I have implemented my model with TensorFlow Lite Micro and now I try to improve the inference with CMSIS-NN.
Is it possible to upload my own model to Edge Impulse to use the CMSIS-NN library, or is it possible to extract the CMSIS-NN library to my own project?

Hi @Odden,

If you have a .tflite model file that you have ported to TFLM to use directly, I recommend checking out this blog post to see how to enable CMSIS-NN acceleration.

The Edge Impulse SDK inferencing library (i.e. the C++ library that you download from your project on our Studio) has CMSIS-NN enabled by default for any processor that supports it (e.g. many ARM architectures). Feel free to look through the SDK source code here to see how we use CMSIS-NN to accelerate inference in our embedded library.