Repository files navigation

This repository contains sources for the libcoral C++ API, which providesconvenient functions to perform inferencing and on-device transfer learningwith TensorFlow Lite models onCoral devices.

For developer documentation, see our guide toRun inference on the Edge TPUwith C++ and check out thelibcoral API reference.

Be sure to clone this repo with submodules:

git clone --recurse-submodules https://github.com/google-coral/libcoral

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

cd libcoralgit submodule init && git submodule update

Then you can build everything usingmake command which invokesBazel internally.

For example, runmake tests to build all C++ unit tests ormake benchmarksto build all C++ benchmarks. To get the list of all available make targets runmake help. All output goes toout directory.

On Linux you can compile natively or cross-compile for 32-bit and 64-bit ARMCPUs.

To compile natively you need to install at least the following packages:

sudo apt-get install -y build-essential \                        libpython3-dev \                        libusb-1.0-0-dev \

and to cross-compile:

sudo dpkg --add-architecture armhfsudo apt-get install -y crossbuild-essential-armhf \                        libpython3-dev:armhf \                        libusb-1.0-0-dev:armhfsudo dpkg --add-architecture arm64sudo apt-get install -y crossbuild-essential-arm64 \                        libpython3-dev:arm64 \                        libusb-1.0-0-dev:arm64

Compilation or cross-compilation is done by setting CPU variable formakecommand:

make CPU=k8      tests  # Builds for x86_64 (default CPU value)make CPU=armv7a  tests  # Builds for ARMv7-A, e.g. Pi 3 or Pi 4make CPU=aarch64 tests  # Builds for ARMv8, e.g. Coral Dev Board

You need to install the following software:

1. Xcode fromhttps://developer.apple.com/xcode/
2. Xcode Command Line Tools:xcode-select --install
3. Bazel for macOS fromhttps://github.com/bazelbuild/bazel/releases
4. MacPorts fromhttps://www.macports.org/install.php
5. Ports ofpython interpreter andnumpy library:sudo port install python35 python36 python37 py35-numpy py36-numpy py37-numpy
6. Port oflibusb library:sudo port install libusb

Right after that all normalmake commands should work as usual. You can runmake tests to compile all C++ unit tests natively on macOS.

Docker allows to avoid complicated environment setup and build binaries forLinux on other operating systems without complicated setup, e.g.,

make DOCKER_IMAGE=debian:buster DOCKER_CPUS="k8 armv7a aarch64" DOCKER_TARGETS=tests docker-buildmake DOCKER_IMAGE=ubuntu:18.04 DOCKER_CPUS="k8 armv7a aarch64" DOCKER_TARGETS=tests docker-build