/ubit_uTensor_demo

Primary LanguageC++

Background Reading

Simple Neural Network on MCU

An End-to-End Tutorial Running Convolution Neural Network on MCU with uTensor

Set up

We typically use MacOS for development. You will need to install utensor-cli (aka. the code generator) and mbed-cli.

For utensor-cli, I recommend doing the following:

  • use pyenv to manage your python versions
  • use pipenv pipenv install -d --skip-lock to install utensor-cli
  • switch utensor-cli to the correct branch, see the section below.

Code Generation

Using the code-generator ubit branch

utensor-cli convert mnist_for_mc.pb --transform-methods='dropout|>linear_reoder|>quantize|>conv_pool|>FakeGatherV2|>inline|>biasAdd|>remove_id_op|>refcnt' --output-nodes mul_6

Build Instruction

mbed import https://github.com/neil-tan/ubit_uTensor_demo
cd ubit_uTensor_demo
# connect your board
mbed compile -m auto -t GCC_ARM --profile=uTensor/build_profile/release.json -f

TODO

  • Patch the main.cpp to use the correct dimensions and tensor-names
  • Verify the output
  • Switch to uBit toolchain, Yotta