Tensorflow C Api + windows 10 + vs2017（详细教程）

Tensorflow C Api+ windows 10 + vs2017 部署（详细教程）

一、获得Tensorflow c 库文件

方法一：选择合适版本，在windows下编译成c接口库文件。
方法二：在官网下载合适的版本，直接获得c接口库文件。libtensorflow-cpu-windows-x86_64-1.15.0.zip

二、配置Tensorflow库文件

1.打开VS2017，创建新项目，选择Visual C++下控制台应用。

2.打开项目的属性页，在【配置属性】->【C/C++】->【常规】->【附加包含目录】添加tensorflow的头文件路径

D:\tensorflow\include  

3.在打开的属性页里面，【配置属性】->【链接器】->【附加库目录】添加tensorflow的lib文件路径。

D:\tensorflow\lib

4.在打开的属性页里面，【配置属性】->【链接器】->【输入】->【附加依赖项】添加tensorflow的lib文件。

tensorflow.lib
kernel32.lib
user32.lib
gdi32.lib
winspool.lib
shell32.lib
ole32.lib
oleaut32.lib
uuid.lib
comdlg32.lib
advapi32.lib

5.配置完毕，代码测试程序。

std::cout << TF_Version() << std::endl;

6.运行软件
结果为1.15.0.

三、Tensorflow C Api 读取模型并检测示例

1.导入模型

TF_Graph* LoadGraph(const char* graph_path, const char* checkpoint_prefix, TF_Status* status) {
  if (graph_path == nullptr) {
    return nullptr;
  }

  auto buffer = ReadBufferFromFile(graph_path);
  if (buffer == nullptr) {
    return nullptr;
  }

  MAKE_SCOPE_EXIT(delete_status){ TF_DeleteStatus(status); };
  if (status == nullptr) {
    status = TF_NewStatus();
  } else {
    delete_status.dismiss();
  }

  auto graph = TF_NewGraph();
  auto opts = TF_NewImportGraphDefOptions();

  TF_GraphImportGraphDef(graph, buffer, opts, status);
  TF_DeleteImportGraphDefOptions(opts);
  TF_DeleteBuffer(buffer);

  if (TF_GetCode(status) != TF_OK) {
    DeleteGraph(graph);
    return nullptr;
  }
  
 auto graph = tf_utils::LoadGraph("Tensorflow.pb");  ///模型路径
 SCOPE_EXIT{ tf_utils::DeleteGraph(graph); };
 if (graph == nullptr) {
	std::cout << "Can't load graph" << std::endl;
	return 1;
	}

2.导入图像

  TF_Tensor* Mat2Tensor(cv::Mat img)
  {
	const std::vector< std::int64_t>input_dims = { 1, IMAGE_SIZE, IMAGE_SIZE, 3 };
	auto data_size = sizeof(std::uint8_t);

	for (auto i : input_dims)
	{
		data_size *= i;
	}

	cv::resize(img, img, cv::Size(IMAGE_SIZE, IMAGE_SIZE));
	cvtColor(img, img, cv::COLOR_BGR2RGB);

	TF_Tensor* image_tensor = CreateTensor(TF_UINT8, input_dims.data(), input_dims.size(), img.data, data_size);

	if (TF_TensorType(image_tensor) != TF_UINT8) {
		std::cout << "Wrong tensor type" << std::endl;
		return nullptr;
	}

	if (TF_NumDims(image_tensor) != static_cast<int>(input_dims.size())) {
		std::cout << "Wrong number of dimensions" << std::endl;
		return nullptr;
	}

	for (std::size_t i = 0; i < input_dims.size(); ++i) {
		if (TF_Dim(image_tensor, static_cast<int>(i)) != input_dims[i]) {
			std::cout << "Wrong dimension size for dim: " << i << std::endl;
			return nullptr;
		}
	}

	if (TF_TensorByteSize(image_tensor) != data_size) {
		std::cout << "Wrong tensor byte size" << std::endl;
		return nullptr;
	}

	auto tensor_data = static_cast<float*>(TF_TensorData(image_tensor));

	if (tensor_data == nullptr) {
		std::cout << "Wrong data tensor" << std::endl;
		return nullptr;
	}
	return image_tensor;
  }

  Mat srcImg = imread("test.jpg");    //使用OpenCV 读取图像
  TF_Tensor* input_image = Mat2Tensor(srcImg);  //Mat 转换成Tensor

3.输入操作名称为input_4的

auto input_op = TF_Output{TF_GraphOperationByName(graph, "input_4"), 0};
if (input_op.oper == nullptr) {
  std::cout << "Can't init input_op" << std::endl;
  return 2;
}

4.输出操作名称为output_node0

auto out_op = TF_Output{TF_GraphOperationByName(graph, "output_node0"), 0};
if (out_op.oper == nullptr) {
  std::cout << "Can't init out_op" << std::endl;
  return 3;
}

5.将模型导入到会话中。

  auto status = TF_NewStatus();
  SCOPE_EXIT{ TF_DeleteStatus(status); }; // Auto-delete on scope exit.
  auto options = TF_NewSessionOptions();
  auto sess = TF_NewSession(graph, options, status);
  TF_DeleteSessionOptions(options);

  if (TF_GetCode(status) != TF_OK) {
    return 4;
  }

6.运行会话。

  TF_SessionRun(sess,
                nullptr, // Run options.
                &input_op, &input_tensor, 1, // Input tensors, input tensor values, number of inputs.
                &out_op, &output_tensor, 1, // Output tensors, output tensor values, number of outputs.
                nullptr, 0, // Target operations, number of targets.
                nullptr, // Run metadata.
                status // Output status.
                );

  if (TF_GetCode(status) != TF_OK) {
    std::cout << "Error run session";
    return 5;
  }

  TF_CloseSession(sess, status);
  if (TF_GetCode(status) != TF_OK) {
    std::cout << "Error close session";
    return 6;
  }

  TF_DeleteSession(sess, status);
  if (TF_GetCode(status) != TF_OK) {
    std::cout << "Error delete session";
    return 7;
  }

7.获得结果

auto data = static_cast<float*>(TF_TensorData(output_tensor));
std::cout << "Output vals: " << data[0] << ", " << data[1] << ", " << data[2] << ", " << data[3] << std::endl;

本文示例代码主要参考https://github.com/Neargye/hello_tf_c_api