Raspberry Pi Pico Compatibility

Hey all you EI folks!

I’m looking into alternative boards that can run my object detection and classification models made via Edge Impulse, and was wondering if the Raspberry Pi Pico has that ability? I’ve heard that other people are using TinyML like projects on them, but I can’t see it listed in the officially supported boards like the Raspberry Pi 4.

Any information/help you could provide would be, as always, very much appreciated :smiley:


Hi Tom,

The pico can definitely run edge impulse models, but the main limitation is the amount of flash and ram available to store more complex models on a particular device. The pico has far too little RAM to run object detection models. It looks to be able to fit image classification models, but note it will run much slower compared to the higher powered RPI4.

Arducam has actually done a lot of community contribution and has some guides available for running EI on a pico, so definitely check that out if you haven’t seen it already:


1 Like

Hey @daschwar, thanks for the response! I guessed the pico might be slightly underpowered, the form factor would be great though for a few projects. Guess I’ll have to wait for higher powered smaller boards to come out :wink:

Well, it seems like it might be possible to run object detection on the Pico in the future:

However, the Pico does have a lower performing Cortex-M CPU (M0).

@janjongboom @daschwar what do you think?

1 Like

Yes good point, a more precise wording on my part would have been to say the pico is too small for our current object detection transfer learning networks. Getting object detection on smaller and smaller devices is definitely something we are working on from multiple angles.

It is hard to say when we will have object detection models that will fit the pico specifically, the only way to tell will be getting accurate flash/ram estimates once we have those optimizations in hand.

And finally just to note, there are often cases where image classification can be used a bit creatively to achieve an originally object-detection based use case!