Raspberry Pi Pico (ArduCAM Pico4ML) Realtime MNIST classifier.

Setup

The easiest way to test this is via the Docker devcontainer.

1. Install Docker. I highly recommend Orbstack for MacOS users as it currently uses fewer resources.
2. Make sure Docker/Orbstack is running
3. Install VSCode.
4. Install Dev Containers Extension
5. git clone https://github.com/britannio/pico-mnist && code pico-mnist/

After following the above instructions open your project in Visual Studio Code and run the >Dev Containers: Reopen in Container command. If the instructions are followed correctly Visual Studio Code should also automatically suggest opening the repository in container mode when the project is loaded.

Notes to self

PICO_SDK_PATH should be set as an environment variable. It should point to the root of the repo https://github.com/raspberrypi/pico-sdk.git once cloned.

chmod u+x compile.sh to make compile.sh executable.

./compile.sh to produce pico_vision.uf2

CLANG=1 python3 tinygrad_compile_mnist.py to train the model and convert it to C.