[TF Lite] composing local build environment

This is obsolete document. You don't need this unless you use older versions of nnstreamer.

Description

This document is created for composing build environment about Tensorflow-lite at local PC rather than mobile device. The source codes are stored at here with MNIST tflite model. If there are missing models, please contact @helloahn .

Preparation

we need ingredients likes below.

• Tensorflow-Lite model file(.tflite) - lists
• Tensorflow Library(.a or .so) - download rpm
• gcc & g++ up to version 6

Extract Headers & Library from RPM

try below command at terminal.

rpm2cpio {file_name}.rpm | cpio -div

and then, you can get the result likes below.

├── include
│   ├── flatbuffers
│   │   ├── base.h
│   │   ├── code_generators.h
│   │   ├── flatbuffers.h
│   │   ├── flatc.h
│   │   ├── flexbuffers.h
│   │   ├── grpc.h
│   │   ├── hash.h
│   │   ├── idl.h
│   │   ├── minireflect.h
│   │   ├── reflection_generated.h
│   │   ├── reflection.h
│   │   ├── registry.h
│   │   ├── stl_emulation.h
│   │   └── util.h
│   └── tensorflow
│       └── contrib
│           └── lite
│               ├── allocation.h
│               ├── arena_planner.h
│               ├── builtin_op_data.h
│               ├── builtin_ops.h
│               ├── context.h
│               ├── error_reporter.h
│               ├── graph_info.h
│               ├── interpreter.h
│               ├── kernels
│               │   ├── activation_functor.h
│               │   ├── gemm_support.h
│               │   ├── kernel_util.h
│               │   ├── op_macros.h
│               │   ├── padding.h
│               │   ├── register.h
│               │   └── test_util.h
│               ├── memory_planner.h
│               ├── model.h
│               ├── nnapi_delegate.h
│               ├── optional_debug_tools.h
│               ├── schema
│               │   └── schema_generated.h
│               ├── simple_memory_arena.h
│               ├── string.h
│               ├── string_util.h
│               ├── tools
│               │   ├── gen_op_registration.h
│               │   ├── mutable_op_resolver.h
│               │   └── verifier.h
│               └── version.h
└── lib64
    ├── libtensorflow-lite.a
    └── pkgconfig
        └── tensorflow-lite.pc

Link Libraries with CMakeLists.txt

• Set C++ Version

To 'make' project with Tensorflow-lite, we have to set C++ version up to 11. In my case, I used version 14.

set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")

• Set include_directories

To include headers(tensorflow, flatbuffers) We have to set include path at CMakeLists.txt likes below.

include_directories(
        ${PATH_TO_HEADERS_DIR}
)

• Set target_link_libraries

To link with libraries, the required libraries should be defined at the CMakeLists.txt file. I defined below three libraries with target_link_libraries.

target_link_libraries( output 
        ${PROJECT_SOURCE_DIR}/lib64/libtensorflow-lite.a
        pthread
        dl
)

pthread and dl are also linked because Tensorflow-lite requires those libraries.

For more information

• Tensorflow Lite API Document