oberluz / pycoral

Python API for ML inferencing and transfer-learning on Coral devices
https://coral.ai
Apache License 2.0
34 stars 2 forks source link

PyCoral API

This repository contains an easy-to-use Python API that helps you run inferences and perform on-device transfer learning with TensorFlow Lite models on Coral devices.

To install the prebuilt PyCoral library, see the instructions at coral.ai/software/.

Note: If you're on a Debian system, be sure to install this library from apt-get and not from pip. Using pip install is not guaranteed compatible with the other Coral libraries that you must install from apt-get. For details, see coral.ai/software/.

Documentation and examples

To learn more about how to use the PyCoral API, see our guide to Run inference on the Edge TPU with Python and check out the PyCoral API reference.

Several Python examples are available in the examples/ directory. For instructions, see the examples README.

Compilation

When building this library yourself, it's critical that you have version-matching builds of libcoral and libedgetpu—notice these are submodules of the pycoral repo, and they all share the same TENSORFLOW_COMMIT value. So just be sure if you change one, you must change them all.

For complete details about how to build all these libraries, read Build Coral for your platform. Or to build just this library, follow these steps:

  1. Clone this repo and include submodules:

    git clone --recurse-submodules https://github.com/oberluz/pycoral

    If you already cloned without the submodules. You can add them with this:

    cd pycoral
    
    git submodule init && git submodule update
  2. Run scripts/build.sh to build pybind11-based native layer for different Linux architectures. Build is Docker-based, so you need to have it installed.

  3. Run make wheel to generate Python library wheel and then pip3 install $(ls dist/*.whl) to install it