EdX Course Updates and Image Classifiers on the Arduino

I am teaching TinyML to high school students as part of the design challenge through my job at Oracle Education Foundation. There are two areas we are encountering challenges:

  1. TensorFlow 1.5 is now deprecated, leaving many of the CoLabs presented in the EdX course non-functional. I am having challenges making things work with TensorFlow 2.x, but have managed to find some workarounds for a few things.

  2. If anyone has a working image classification example that they have deployed on the Arduino Nano 33 BLE, I would be excited to learn about that. Most of the projects contain image classification, but I can’t seem to get a working example on the Arduino for numerous reasons.

Are there plans to update the EdX curriculum to work with TensorFlow 2.x?

I appreciate any insights anyone has.

Thank you,

1 Like

Hi for #2 this tutorial should help: Adding sight to your sensors - Edge Impulse Documentation. For #1 I had tried to make some fixes to downgrade the TF version back to 1.5 which had been stable a few months ago on some of the Colabs but I guess from your experience that isn’t working anymore either. Unfortunately, we don’t have the capacity to make the upgrade to TF 2.x at the moment. That said, we know this is a major issue and are looking into ways we could find the expertise and capacity to initiate that update.

Thank you for the update. It helps to have the information about the process moving forward.

I have been diving into Edge Impulse for solutions to the challenges we are encountering. It has been great for Custom Keyword Spotting. The image classification will be a little bit more work because we will have to move from the shields in the TinyML kits to breadboarding, but that is my next step with Edge Impulse.

I’m pretty sure you should be able to use the shield with edge impulse — I got it working last year and did a short workshop on it — Africa (see my slides / video from day 3) — but I haven’t tried it recently.

1 Like