Qt on Android with Tensorflow as 3rd party lib

[LeoChen]

Hello:

Currently I tried to embed tensorflow c++ lib into my app, I somehow managed to done the job on desktop and iOS, but failed on Android. The problem is the link problem as the following:

/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/lib/libtensorflow-core.a(net.o):net.cc:function tensorflow::internal::PickUnusedPortOrDie(): error: undefined reference to 'rand'
/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/lib/libtensorflow-core.a(subprocess.o):subprocess.cc:function tensorflow::SubProcess::Communicate(std::string const*, std::string*, std::string*) [clone .part.5]: error: undefined reference to 'sigemptyset'
/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/protobuf/lib/libprotobuf.a(strutil.o):strutil.cc:function google::protobuf::safe_strtof(char const*, float*): error: undefined reference to 'strtof'
/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/protobuf/lib/libprotobuf-lite.a(strutil.o):strutil.cc:function google::protobuf::safe_strtof(char const*, float*): error: undefined reference to 'strtof'
collect2: error: ld returned 1 exit status
make: *** [libCrossDemo.so] Error 1
11:57:57: The process "/usr/bin/make" exited with code 2.
Error while building/deploying project CrossDemo (kit: Android for armeabi-v7a (GCC 4.9, Qt 5.8.0))
When executing step "Make"

I think this problem is because I did not reference to the std lib correctly. Could someone help me with this problem? thank you.

====
OS: macOS 10.12.3
SDK: android-21
NDK: r13b

[SGaist]

Hi and welcome to devnet,

How did you configure your project to link to tensorflow ?

[LeoChen]

Hello:

My configuration for android is as the following:

android: {

CONFIG += c++11

LIBS += -L/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/lib -ltensorflow-core
INCLUDEPATH += /Users/chen/Documents/Git/tensorflow \
        #/Users/chen/Documents/Git/tensorflow/bazel-out/local-fastbuild/genfiles
DEPENDPATH += /Users/chen/Documents/Git/tensorflow \
        #/Users/chen/Documents/Git/tensorflow/bazel-out/local-fastbuild/genfiles

LIBS += -L/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/protobuf/lib -lprotobuf -lprotobuf-lite
#LIBS += -L/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_ios/protobuf_ios/lib -lprotobuf-lite -lprotobuf

INCLUDEPATH += /Users/chen/Documents/Git/tensorflow/ \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads/protobuf/src \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads/eigen \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/proto

DEPENDPATH += /Users/chen/Documents/Git/tensorflow/ \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads/protobuf/src \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads/eigen \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/proto

PROBOF_PATH = /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/protobuf/include
INCLUDEPATH += $$PROTOBUF_PATH
DEPENDPATH += $$PROTOBUF_PATH

# libc.a
LIBS += -L/Users/chen/Library/Android/android-ndk-r13b/platforms/android-21/arch-arm64/usr/lib/libc.a

QMAKE_LFLAGS += -lz -lm -llog -landroid -lgnustl_static
QMAKE_LFLAGS += -Wl,--allow-multiple-definition -Wl,--whole-archive
QMAKE_CXXFLAGS += -std=c++11 -fno-rtti -fno-exceptions \
                     -fpic -O2 -mfpu=neon -DTF_LEAN_BINARY  \
                     -DGOOGLE_PROTOBUF_NO_RTTI \
                     -DGOOGLE_PROTOBUF_NO_STATIC_INITIALIZER
#QMAKE_LFLAGS += -llog -lz
#CONFIG+= static
#QMAKE_LFLAGS += -static -static-libgcc

#QMAKE_LFLAGS += -force_load /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_android/lib/libtensorflow-core.a
}

Comparably with the iOS configuration, which kind of works fine:

 ios: {
LIBS += -L/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_ios/lib -ltensorflow-core
LIBS += -L/Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_ios/protobuf_ios/lib -lprotobuf-lite -lprotobuf

INCLUDEPATH += /Users/chen/Documents/Git/tensorflow/ \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads_ios/protobuf/src \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads_ios \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads_ios/eigen \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_ios/proto

DEPENDPATH += /Users/chen/Documents/Git/tensorflow/ \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads_ios/protobuf/src \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads_ios \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/downloads_ios/eigen \
                /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_ios/proto

CONFIG -= bitcode
QMAKE_LFLAGS += -force_load /Users/chen/Documents/Git/tensorflow/tensorflow/contrib/makefile/gen_ios/lib/libtensorflow-core.a

LIBS += -framework Accelerate

# Copy models into bundle
APP_PB_FILES.files = $$PWD/Models/graph_binary.pb
APP_PB_FILES.path = Models
QMAKE_BUNDLE_DATA += APP_PB_FILES
}

[SGaist]

Which version of the SDK/ndk are you using ?

[LeoChen]

The versions of SDK/NDK are as the following:

SDK: android-21
NDK: r13b

Officially from Google

[ekkescorner]

@LeoChen you should use NDK r10e (https://wiki.qt.io/Qt_5.8_Tools_and_Versions)

[LeoChen]

Thank you for your information, I tried but still get the same error.

[LeoChen]

@ekkescorner Mysteriously solved after several failure, thanks a lot！

[GTDev]

You can also find a guide how to build and integrate TensorFlow with Qt and V-Play here:
https://v-play.net/cross-platform-development/machine-learning-add-image-classification-for-ios-and-android-with-qt-and-tensorflow

The example project of this guide includes Image Classification and Object Recognition features based on neural network models, which you can control from QML:

Tensorflow provides the machine learning framework, whereas V-Play and Qt facilitate the app deployment to multiple platforms: desktop and mobile.