Based on tensorflow-micropython-examples

The repo's structure is designed to be easily implemented.

We have two possible cmakes to use:

• src/base.cmake (Includes only tflite micro)
• src/full.cmake (Includes additional libraries like ulab)

A helper script scripts/build_and_check.sh can be used to build the firmware for a given board using ESP-IDF 5.4. The script follows the same steps as the GitHub workflow and will fetch the required ESP-IDF release if missing.

# Build firmware for the default MICROLITE board
./scripts/build_and_check.sh

# Or specify a board
./scripts/build_and_check.sh MICROLITE_S3

The script expects the submodules to be accessible over the network in order to fetch Micropython and related dependencies.

When building with ESP-IDF 5.4 the following submodule commits have been verified to work:

• micropython: 9bde12597a6980ff87ff0137a2616e6e430a1a0e
• esp-tflite-micro: 772214721682ef2d3eed09cafea777edad55541f
• esp-nn: 12129cf04b09af0023127ca7551dc1a363344211
• micropython-ulab: a05ec05351260cf48fefc347265b8d8bf29c03f1

These versions are pinned via Git submodules. After cloning the repository run git submodule update --init --recursive to fetch the correct revisions. If Git submodules cannot be fetched directly (for example in environments without full Git access), run scripts/fetch_submodules.sh which downloads the same revisions using plain HTTP archives.