Problem Statement: I would like to use Golang TensorFlow to run my model in Raspberry PI 3. To use TensorFlow go, I need to have libtensorflow.s
I found solution:
Couple of things to take care.
***For Tensorflow < v1.5***
Follow instruction here: https://blog.meinside.pe.kr/TensorFlow-and-Go-on-Raspberry-Pi/
***For Tensorflow v1.11.0 - v1.12.0***
Install bazel:
1. Download bazel from link: https://github.com/bazelbuild/bazel/releases
for specific architecture.
2. Install bazel
chmod +x bazel-<version>-installer-linux-<arch>.sh
./bazel-<version>-installer-linux-<arch>.sh --user
Install protobuf:
1. Build/Install protobuf from source.
https://github.com/protocolbuffers/protobuf/blob/master/src/README.md
Build Tensorflow:
1. Git clone
git clone https://github.com/tensorflow/tensorflow.git
2. Check and remove build from specific tensorflow version
go to: tensorflow/tensorflow/tools/ci_build/install/*_toolchain.sh [specific to your linux type and arch]
remove the bazel build and installation like as follows
look for the line: **rm -rf /usr/local/bin/bazel** and remove
and remove **bazel** from line **apt-get install -y ...**
then run
CI_DOCKER_EXTRA_PARAMS="-e CI_BUILD_PYTHON=python3 -e CROSSTOOL_PYTHON_INCLUDE_PATH=/usr/include/python3.6" tensorflow/tools/ci_build/ci_build.sh PI tensorflow/tools/ci_build/pi/build_raspberry_pi.sh
or
tensorflow/tools/ci_build/ci_build.sh PI tensorflow/tools/ci_build/pi/build_raspberry_pi.sh
or
tensorflow/tools/ci_build/ci_build.sh PI tensorflow/tools/ci_build/pi/build_raspberry_pi.sh PI_ONE
Output:
Output can be found here:
output-artifacts
output-artifacts/tensorflow-1.12.0-cp27-none-linux_armv7l.whl
output-artifacts/libtensorflow_framework.so
output-artifacts/benchmark_model
output-artifacts/libtensorflow.so
Dependency Versions:
Tensorflow Bazel protoc
<1.5 0.5.0 3.0 <any>
1.11.0-1.12.0 0.15.2 3.x <any>
>1.12.0 new vers <any>