Using EdgeImpulse for model that are not machine vision

Arduino Boards
1

Hello,

I have a model which is not machine vision in EdgeImpulse. Would it be possible to deploy it as a OpenMV library ?
As Nano BLE sense 33 (and also the RP2040) are compatible with the OpenMV IDE it would be interesting to be able to do audio model with the OpenMV firmware.
Thanks

2

The plan is to upgrade the TensorFlow library later this year to support this through the tf module.

I have more IDE updates to do in the mean-time.

3

ok thanks. Is the tf module supported by the RP2040 at the moment?
I m trying to classify a raw image from a model that I know can run the openMV cam but I get the following error :

OSError: This function is unavailable on your OpenMV Cam.
MicroPython: v1.15-r57-dirty OpenMV: v4.1.1-dirty HAL: v1.2.0 BOARD: Arduino Nano RP2040 Connect-RP2040

this is reffering to :

objs = tf.classify(model_file, img)
4

Is the model store on the internal flash?

5

yes, indeed, is this a problem?

6

No, but, I’m not sure what the error is referencing. Maybe you are using a TF lite macro we do not support. Please check what ops you are using.

7

OK. The strange thing is that the exact same code works perfectly well on the portenta so surely the TF lite macro is supported.
I was wondering what the “-dirty” in the Micropython and OpenMV version referred to?

8

Dirty means the binary was generated from code that wasn’t committed.

9

OK thanks.
What are the TF lite macros supported on the RP2040 ?

10

Actually, nothing right now as our library is not compiled for the M0. This is the issue, it only runs on M4/M7 arm cores.

Please wait until next month. We will have a new firmware release which enables this.

11

Hello,

Do you still plan to release the new firmware for the M0 that supports TF lite macros?

Thank you

12

Yes, and the m55. Just have to find a working version of the library.