I have a custom imu board running arduino+tflite, where I’m able to make classification models using EI using the linear acceleration+gyroscope data, but using the “raw data” processing block, which works well. Our SDK allows users to send tflite models via ble to the board and pipe the raw data samples into the model.
However, I’d like to use processing blocks like the “Spectral Analysis” block, but it seems the inputs to the model are not raw data inputs, but rather some “feature” inputs which calculated outside the tflite model (figured this out when deploying as an arduino sketch to see how the inputs are set). Is there a way to include this feature calculation in the tflite model itself so I can continue to pipe in the raw samples?