看錯誤內容：

/data/projects/LipReadingSDKGPU/lib/cwlibs/libLipReading.so: undefined reference to `tensorflow::Status::ToString[abi:cxx11]() const‘
/data/projects/LipReadingSDKGPU/lib/cwlibs/libLipReading.so: undefined reference to `tensorflow::ReadBinaryProto(tensorflow::Env*, std::__cxx11::basic_string<char, std::char_traits<char
 
>, std::allocator<char> > const&, google::protobuf::MessageLite*)‘
/data/projects/LipReadingSDKGPU/lib/cwlibs/libLipReading.so: undefined reference to `google::protobuf::internal::fixed_address_empty_string[abi:cxx11]‘
/data/projects/LipReadingSDKGPU/lib/cwlibs/libLipReading.so: undefined reference to `tensorflow::internal
 
::CheckOpMessageBuilder::NewString[abi:cxx11]()‘
/data/projects/LipReadingSDKGPU/lib/cwlibs/libLipReading.so: undefined reference to `tensorflow::strings::StrCat[abi:cxx11](tensorflow::strings::AlphaNum const&)‘

stackoverflow上一個友人的解釋：

I suspect this is a C++ ABI issue. The ABI for std::string has changed in
 
 GCC 5 (related to C++11 requirements, but it applies even if you aren‘t using C++11). See:

https://gcc.gnu.org/gcc-5/changes.html#libstdcxx

If libprotobuf was built with GCC 4.x or prior, but your app is built with GCC 5, then you will see problems, because libprotobuf uses std::string in its interface. You have two options:

1.Rebuild libprotobuf with GCC 5 (but now any apps built with GCC 4 won‘t work with the new version of libprotobuf).
2.Build you app with -D_GLIBCXX_USE_CXX11_ABI=0 as described at the above link. This will force GCC to use the old ABI version.

這個錯誤是在我把gcc4.8編的工程遷移到ubuntu16.04(gcc5.4)上編譯時候發生的。這是C++ ABI一個錯誤，gcc4升gcc5時，std::string庫接口做了遷移，而我工程中用了三方庫tensorflow和protobuf，protobuf中用到了std::string。已經編好的tensorflow是用gcc4.8編的，gcc5上鏈接，接口對不上。兩種解決方案：一種用gcc5重新編譯三方庫和工程，另一種就是在工程的編譯選項（而不是鏈接選項）上加上-D_GLIBCXX_USE_CXX11_ABI=0參數，強制使用舊接口，問題解決。

更詳細的解釋可以參考：http://www.itkeyword.com/doc/8367047421581508x309/linking-problems-due-to-symbols-with-abicxx11

c++ - Linking problems due to symbols with abi::cxx11?