In this update I want to show you how easy it is to create custom machine learning models for your Piunora using tools like Edge Impulse. Edge Impulse recently added support for the Raspberry Pi 4 and, along with it, the Compute Module 4. In fact, you don’t need anything but your Piunora for the entire workflow covered in this update! I used the Raspberry Pi HQ Camera connected via the camera port to gather data using Edge Impulse Studio. Everything is running in the browser and connected to the
edge-impulse-runner software running locally on Piunora to stream the camera feed to your browser.
Edge Impulse Studio will automatically detect the running instance and allow you to create a dataset by capturing photos and labeling them, all of which is done in the browser without downloading any extra software. Next, all you need to do is tell it what sort of model you want to use with your dataset and train it on the Edge Impulse servers, so there’s no need for expensive GPUs. You can do this from any device, even your Piunora. Once the training is done, you can deploy the model to Piunora and execute it locally, using the same camera you used to capture your data set. It works at an impressive 60 FPS and it does so completely offline.
You can do the same thing with arbitrary sensor data from an I²C sensor attached to Piunora’s Qwiic connector. The Edge Impulse Python SDK is well suited for that, as it integrates seamlessly with a huge number of Adafruit CircuitPython drivers! I will explore this in a future update.
In the meantime, checkout the video below for a run down of the process.
A Raspberry Pi HAT that lets you develop edge AI camera applications in minutes, not months
A single-board computer in the Adafruit Feather form factor
An embedded platform for combining Depth and AI, built around Myriad X