How can I use Tensorflow in tizen studio? - tensorflow

Question

I want to use tensorflow in tizen.
Tizen studio's architecture is x86.
But tensorflow is only 64-bit.
How can I use tensorflow in tizen studio??

Answer 1

Download packages for target device
tensorflow
python
db4
mkdir tmp
cd tmp
BASE_URL='http://download.tizen.org/snapshots/tizen/5.0-base/latest/repos/standard/packages/armv7l/'
wget -O - $BASE_URL 2>/dev/null | egrep 'db4-[0-9]' | awk -F'"' '{print $2}' | xargs -i wget $BASE_URL{}
wget -O - $BASE_URL 2>/dev/null | egrep 'python-[0-9]' | awk -F'"' '{print $2}' | xargs -i wget $BASE_URL{}
UNIFIED_URL='http://download.tizen.org/snapshots/tizen/5.0-unified/latest/repos/standard/packages/armv7l/'
wget -O - $UNIFIED_URL 2>/dev/null | grep 'tensorflow-v' | awk -F'"' '{print $2}' | xargs -i wget $UNIFIED_URL{}
Install packages on target device
sdb root on; sdb shell 'mount -o remount,rw /'
sdb push *.rpm /tmp
sdb shell 'cd /tmp; rpm -ivh --force db4*.rpm; rpm -ivh --force *python*.rpm; rpm -ivh --force tensorflow*.rpm'
cd ..
rm -r tmp
Download packages needed for building
tensorflow
tensorflow-devel
libpython-2.7
libgfortran
libgomp
mkdir tmp
cd tmp
BASE_URL='http://download.tizen.org/snapshots/tizen/5.0-base/latest/repos/standard/packages/armv7l/'
wget -O - $BASE_URL 2>/dev/null | grep 'libpython-2.7' | awk -F'"' '{print $2}' | xargs -i wget $BASE_URL{}
wget -O - $BASE_URL 2>/dev/null | grep 'libgfortran-' | awk -F'"' '{print $2}' | xargs -i wget $BASE_URL{}
wget -O - $BASE_URL 2>/dev/null | grep 'libgomp-' | awk -F'"' '{print $2}' | xargs -i wget $BASE_URL{}
UNIFIED_URL='http://download.tizen.org/snapshots/tizen/5.0-unified/latest/repos/standard/packages/armv7l/'
wget -O - $UNIFIED_URL 2>/dev/null | grep 'tensorflow' | grep -v 'lite' | awk -F'"' '{print $2}' | xargs -i wget $UNIFIED_URL{}
Extract header and so files and selectively import to project lib folder
ls *.rpm | xargs -i bash -c "rpm2cpio {} | cpio -idmv"
cd usr/lib/
ln -s libpython2.7.so.1.0 libpython2.7.so
ln -s libgfortran.so.3 libgfortran.so
cp libgfortran.so* libgomp.so.1* libpython2.7.so* libpywrap_tensorflow_internal.so ../../../lib/
cd ../include
cp -r tensorflow ../../../inc/
cd ../../..
rm -r tmp
Set up include path in Tizen Studio project
screenshot
Set up library path and add libraries to Tizen Studio project
screenshot
Note that if you want to build the project for emulator, you have to use library files for x86 architecture, so you would have to replace "armv7l" in all the URLs above with "i686"

Answer 2

Tizen studio is just development tool.
Tizen supports x86-64(AMD64) and you can use the tensorflow on Tizen.
If you want to use Tizen with tensorflow, you need to install some packages(tensorflow, python,...) yourself.

Answer 3

If you can afford using Tensorflow-Lite, you can use it directly in Tizen-latest. Soon, with Tizen 5.5 M2, there will be Machine Learning APIs that allows app developers to directly plugin .tflite models as well as their own native models (as .so files or as functions) without worrying about importing external libraries.
Anyway, if you want to do it right now, use the Tizen-latest (Tizen:Unified project in build.tizen.org), use tensorflow-lite-dev.rpm package directly or use nnstreamer-capi package, which gives you some easy-to-use interfaces. Note that nnstreamer (https://github.com/nnsuite/nnstreamer ) is going to be the main engine for Tizen-Machine-Learning (backend of its Machine Learning API sets). Although nnstreamer is compatible with Tensorflow (non-lite), Caffe2, PyTorch, ROS, and so on, they are not included in Tizen by default.