TensorFlow - （一）

一、使用TensorFlowlite 构建iOS自定义版本

• 由于TensorFlowlite 提供的 cocoapods 版本 支持到iOS9;所以需要使用手动编译源码，构建静态包支持iOS8以上系统；
• TensorFlow目前支持swift和OC，如果想使用直接cococapads集成；

二、在iOS使用C++ api

• 使用api 之前 需要你准备好lite版的model；model训练完成之后转换为lite即可
• 加载model的类：

class FlatBufferModel {
 public:
  // Builds a model based on a file.
  // Caller retains ownership of `error_reporter` and must ensure its lifetime
  // is longer than the FlatBufferModel instance.
  // Returns a nullptr in case of failure.
  static std::unique_ptr BuildFromFile(
      const char* filename,
      ErrorReporter* error_reporter = DefaultErrorReporter());
 
  // Builds a model based on a pre-loaded flatbuffer.
  // Caller retains ownership of the buffer and should keep it alive until
  // the returned object is destroyed. Caller also retains ownership of
  // `error_reporter` and must ensure its lifetime is longer than the
  // FlatBufferModel instance.
  // Returns a nullptr in case of failure.
  // NOTE: this does NOT validate the buffer so it should NOT be called on
  // invalid/untrusted input. Use VerifyAndBuildFromBuffer in that case
  static std::unique_ptr BuildFromBuffer(
      const char* caller_owned_buffer, size_t buffer_size,
      ErrorReporter* error_reporter = DefaultErrorReporter());
}

• C++ 加载model 的步骤

构建FlatBufferModel，并且从构建的FlatBufferModel中初始化Interpreter实例；
可以选择性地进行优化Tensor;
设置Tensor的值；
调用运行推理；
读取Tensor的输出值
使用Interpreter对象注意事项：
Tensor用整数值表示，避免使用字符串进行比较；
访问Interpreter避免在多个线程中并发访问；
Tensor的输入和输出内存分配使用AllocateTensors() 在重置大小之后进行

tflite::FlatBufferModel model(path_to_model);
 
tflite::ops::builtin::BuiltinOpResolver resolver;
std::unique_ptr interpreter;
tflite::InterpreterBuilder(*model, resolver)(&interpreter);
 
// Resize input tensors, if desired.
interpreter->AllocateTensors();
 
float* input = interpreter->typed_input_tensor(0);
// Fill `input`.
 
interpreter->Invoke();
 
float* output = interpreter->typed_output_tensor(0);