LibTensorFlowForiOSSwift

中文文档

This is a TensorFlow demo that can be run on iOS and use swift develop. It implements a text classifier that can predict emoji from short text (like tweets).

Modified from emoji-tf-ios; use emoji_frozen.pb model from emoji-tf-ios.

how to run

• open terminal and go to project root folder

• input sh run.sh.

In the end, will automatically open Xcode

you can run the project now

about run.sh

it will compile the TensorFlow for iOS automatically

about project setting

how to import TensorFlow in ios project

• download TensorFlow to the root folder and compile

  befor compile source file, you should change flow TensorFlow Kernel files. (version <= 1.2.1)

  in run.sh, it will change them automatically

  kernel path: tensorflow/tensorflow/core/kernels

  • cwise_op_add_1.cc

    before:

     ...
     ...
     #include "tensorflow/core/kernels/cwise_ops_common.h"
    
     namespace tensorflow {
     REGISTER5(BinaryOp, CPU, "Add", functor::add, float, Eigen::half, double, int32,
               int64);
    
     #if TENSORFLOW_USE_SYCL
     ...
     ...

    change to:

     ...
     ...
     #include "tensorflow/core/kernels/cwise_ops_common.h"
    
     namespace tensorflow {
     REGISTER5(BinaryOp, CPU, "Add", functor::add, float, Eigen::half, double, int32,
               int64);
    
     // line 21 insert this code
     #if defined(__ANDROID_TYPES_SLIM__)
     REGISTER(BinaryOp, CPU, "Add", functor::add, int32);
     #endif  // __ANDROID_TYPES_SLIM__
     // insert end
    
     #if TENSORFLOW_USE_SYCL
     ...
     ...

  • cwise_op_less.cc

    before:

     ...
     ...
     #include "tensorflow/core/kernels/cwise_ops_common.h"
    
     namespace tensorflow {
     REGISTER8(BinaryOp, CPU, "Less", functor::less, float, Eigen::half, double,
               int32, int64, uint8, int8, int16);
     #if GOOGLE_CUDA
     REGISTER7(BinaryOp, GPU, "Less", functor::less, float, Eigen::half, double,
               int64, uint8, int8, int16);
     ...
     ...

    change to:

     ...
     ...
     #include "tensorflow/core/kernels/cwise_ops_common.h"
    
     namespace tensorflow {
     REGISTER8(BinaryOp, CPU, "Less", functor::less, float, Eigen::half, double,
               int32, int64, uint8, int8, int16);
     
     // line 21 insert this code
     #if defined(__ANDROID_TYPES_SLIM__)
     REGISTER(BinaryOp, CPU, "Less", functor::less, int32);
     #endif  // __ANDROID_TYPES_SLIM__
     // insert end
    
     #if GOOGLE_CUDA
     REGISTER7(BinaryOp, GPU, "Less", functor::less, float, Eigen::half, double,
               int64, uint8, int8, int16);
     ...
     ...

• about libtensorflow-core.a

  • in Other Link Flags add $(SRCROOT)/tensorflow/tensorflow/contrib/makefile/gen/lib/libtensorflow-core.a
  • in Library Search Paths add $(SRCROOT)/tensorflow/tensorflow/contrib/makefile/gen/lib

• about libprotobuf.a & libprotobuf-lite.a

  • in Build Phases | Link Binary With Libraries add libprotobuf.a & libprotobuf-lite.a (path: tensorflow/tensorflow/contrib/makefile/gen/protobuf_ios/lib/)
  • in Library Search Paths add $(SRCROOT)/tensorflow/tensorflow/contrib/makefile/gen/protobuf_ios/lib

• in Header Search Paths add flows

  • $(SRCROOT)/tensorflow/
  • $(SRCROOT)/tensorflow/tensorflow/contrib/makefile/downloads/protobuf/src/
  • $(SRCROOT)/tensorflow/tensorflow/contrib/makefile/downloads
  • $(SRCROOT)/tensorflow/tensorflow/contrib/makefile/downloads/eigen
  • $(SRCROOT)/tensorflow/tensorflow/contrib/makefile/gen/proto

• in Other Link Flags add -force_load

• in Build Phases | Link Binary With Libraries add Accelerate.framework

• in C++ Language Dialect select GNU++11 or GNU++14

• in C++ Standard Library select libc++

• Enable Bitcode set No

• remove any -all_load ，use -ObjC replace it

  Remove any use of the -all_load flag in your project. The protocol buffers libraries (full and lite versions) contain duplicate symbols, and the -all_load flag will cause these duplicates to become link errors. If you were using -all_load to avoid issues with Objective-C categories in static libraries, you may be able to replace it with the -ObjC flag.

• suppress TensorFlow warning：

  • in Other C Flags & Other C++ Flags add -isystem $(SRCROOT)/tensorflow

reference:

• compile TensorFlow:

  https://github.com/tensorflow/tensorflow/tree/master/tensorflow/contrib/makefile

• change Kernels error:

  https://github.com/h4x3rotab/emoji-tf-ios/blob/master/README.md

• import TensorFlow to iOS:

  https://github.com/tensorflow/tensorflow/blob/master/tensorflow/examples/ios/README.md

• suppress warning of TensorFlow:

  https://clang.llvm.org/docs/UsersManual.html#id27