Hi Arm Community
I was wondering is there an efficient/any way to transfer the data(weights & biases & architectural info) or read the hdf5 file in CMSIS api from a tensorflow model (saved in hdf5) into a nucleo-f446re microcontroller unit programed in keil.
Because the hdf5 file gives a data output like this, which was saved using the "model.save('model_v1.h5')" command:
as I need to pass in these convolution or dense layer "weights", "biases" and "activation function information" into the arm CMSIS-NN api.
The online documentation I have seen so far only seem to explain the weight matrix transformation and quantization only and seem to have assumed that the data is already in the microcontroller and extracted, and not how to extract the weight, bias, and activation function information from the saved model.
For example this ARM documentation : https://developer.arm.com/solutions/machine-learning-on-arm/developer-material/how-to-guides/converting-a-neural-network-for-arm-cortex-m-with-cmsis-nn/single-page that only explains the tensor reshaping's and quantization, and not the initial steps of importing the data into the microcontroller and extracting the information from the hdf5 file.
I hope that some one familiar with this can provide some guidance
View all questions in Cortex-M / M-Profile forum