What is causing the spectrogram block to display the following error message:
Error: Product of "output_config.shape" width, height should be equal to the number of features returned for "spectrogram" block
I’m trying to use Edge Impulse with a Arduino Nano RP2040 Connect after downloading a newly built firmware from:
https://github.com/edgeimpulse/firmware-pi-rp2040
after starting the edge-impulse-daemon EI recognizes the Arduino board as a:
RASPBERRY_PI_RP2040
The Data Acquisition screen will let me collect various data samples using the Inertial sensors and the data graph looks somewhat reasonable for the type of gesture motion recorded.
I created an Impulse with a Spectogram block and 2 learning blocks: NN Keras) block and with the two learning blocks: 1. NN (KERAS) and 2. Anomaly Detection.
Then when I click on the Spectogram block thats when I get the following error message:
Error: Product of "output_config.shape" width, height should be equal to the number of features returned for "spectrogram" block
Any explanations or suggested ways to resolve this issue will be greatly appreciated.
I saw that the project id is part of the URL I pasted above. Regardless, the Project ID: 130909
I wonder if its something about my local build of the firmware for the rp2040 based on the instructions in that github repos’ README.md:
https://github.com/edgeimpulse/firmware-pi-rp2040
Anyway, please let me know if you can tell from the project what might be wrong. If its potentially my build of the firmware … Can you point me to an official build of that firmware that is proven to work with Arduino Nano rp2040 Connect?
Sorry for the late reply, I was traveling for TheThingsConference last week in Amsterdam.
I can see several issues for your project.
Your dataset contains both audio samples (that you disabled) and accelerometer + gyroscope samples. The accelero + gyroscope have been sampled at different frequencies (some at 62,5Hz, some at 100Hz). In your impulse, the frequency has been set to 100Hz which will add a couple of zeros for your 62,5Hz samples.
The spectrogram only accept one axis, for gesture recognitions, I would suggest to use the spectral analysis. However, if you really need to use the spectrogram for any particular reason, you can stack 3 individual spectrogram blocks and associate them with one axis each.
I do not have an Arduino Nano RP 2040 so I won’t be able to help you on the deployment part but let me know if you have further questions.