【Silicon Labs 开发套件评测】基于深度学习的tensorflow应用代码分析

北方

【Silicon Labs 开发套件评测】基于深度学习的tensorflow应用代码分析 [复制链接]

 

基于深度学习的tensorflow应用代码分析

1. Silab这个版本的SDK最有趣的更新就是把深度学习框架tensorflow引入嵌入式开发中。本次还提供了一个范例程序，tensorflow_lite_micro_helloworld，非常值得分析和对比一下。

2. 按照前述流程创建范例程序，然后编译并下载调试。

代码的过程是按照正弦函数输出波形，对应在PWM控制板载LED的亮度逐步变换，串口输出x,y的对应数值如下

 

3. 代码简析

整个逻辑仍然采用了标准的框架分析方法，在main.c启动任务app.c，在app.c中才启动最终的核心代码。这里是在tensorflow子目录项目下，

main.c

int main(void)
{
  // Initialize Silicon Labs device, system, service(s) and protocol stack(s).
  // Note that if the kernel is present, processing task(s) will be created by
  // this call.
  sl_system_init();

  // Initialize the application. For example, create periodic timer(s) or
  // task(s) if the kernel is present.
  app_init();

#if defined(SL_CATALOG_KERNEL_PRESENT)
  // Start the kernel. Task(s) created in app_init() will start running.
  sl_system_kernel_start();
#else // SL_CATALOG_KERNEL_PRESENT
  while (1) {
    // Do not remove this call: Silicon Labs components process action routine
    // must be called from the super loop.
    sl_system_process_action();

    app_process_action();

#if defined(SL_CATALOG_POWER_MANAGER_PRESENT)
    // Let the CPU go to sleep if the system allows it.
    sl_power_manager_sleep();
#endif
  }
#endif // SL_CATALOG_KERNEL_PRESENT
}

app.c

void app_init(void)
{
#if defined(SL_CATALOG_MVP_PRESENT)
  sli_mvp_init_t init = { .use_dma = false };
  sli_mvp_init(&init);
#endif
  tensorflow_lite_micro_helloworld_init();
}

/***************************************************************************//**
 * App ticking function.
 ******************************************************************************/
void app_process_action(void)
{
  tensorflow_lite_micro_helloworld_process_action();
}

void tensorflow_lite_micro_helloworld_init(void)
{
  sl_pwm_start(&sl_pwm_led0);
  setup();
}

/***************************************************************************//**
 * Ticking function.
 ******************************************************************************/
void tensorflow_lite_micro_helloworld_process_action(void)
{
  // Delay between model inferences to simplify visible output
  sl_sleeptimer_delay_millisecond(100);
  loop();
}

核心代码都在main_functions.c

// The name of this function is important for Arduino compatibility.
void setup() {
  // Set up logging. Google style is to avoid globals or statics because of
  // lifetime uncertainty, but since this has a trivial destructor it's okay.
  // NOLINTNEXTLINE(runtime-global-variables)
  static tflite::MicroErrorReporter micro_error_reporter;
  error_reporter = µ_error_reporter;

  // Map the model into a usable data structure. This doesn't involve any
  // copying or parsing, it's a very lightweight operation.
  model = tflite::GetModel(g_model);
  if (model->version() != TFLITE_SCHEMA_VERSION) {
    TF_LITE_REPORT_ERROR(error_reporter,
                         "Model provided is schema version %d not equal "
                         "to supported version %d.",
                         model->version(), TFLITE_SCHEMA_VERSION);
    return;
  }

  // This pulls in all the operation implementations we need.
  // NOLINTNEXTLINE(runtime-global-variables)
  static tflite::AllOpsResolver resolver;

  // Build an interpreter to run the model with.
  static tflite::MicroInterpreter static_interpreter(
      model, resolver, tensor_arena, kTensorArenaSize, error_reporter);
  interpreter = &static_interpreter;

  // Allocate memory from the tensor_arena for the model's tensors.
  TfLiteStatus allocate_status = interpreter->AllocateTensors();
  if (allocate_status != kTfLiteOk) {
    TF_LITE_REPORT_ERROR(error_reporter, "AllocateTensors() failed");
    return;
  }

  // Obtain pointers to the model's input and output tensors.
  input = interpreter->input(0);
  output = interpreter->output(0);

  // Keep track of how many inferences we have performed.
  inference_count = 0;
}

// The name of this function is important for Arduino compatibility.
void loop() {
  // Calculate an x value to feed into the model. We compare the current
  // inference_count to the number of inferences per cycle to determine
  // our position within the range of possible x values the model was
  // trained on, and use this to calculate a value.
  float position = static_cast<float>(inference_count) /
                   static_cast<float>(kInferencesPerCycle);
  float x_val = position * kXrange;

  // Place our calculated x value in the model's input tensor
  input->data.f[0] = x_val;

  // Run inference, and report any error
  TfLiteStatus invoke_status = interpreter->Invoke();
  if (invoke_status != kTfLiteOk) {
    TF_LITE_REPORT_ERROR(error_reporter, "Invoke failed on x_val: %f\n",
                         static_cast<double>(x_val));
    return;
  }

  // Read the predicted y value from the model's output tensor
  float y_val = output->data.f[0];

  // Output the results. A custom HandleOutput function can be implemented
  // for each supported hardware target.
  HandleOutput(error_reporter, x_val, y_val);

  // Increment the inference_counter, and reset it if we have reached
  // the total number per cycle
  inference_count += 1;
  if (inference_count >= kInferencesPerCycle) inference_count = 0;
}

这里其实直接调用了tensorflow的API

const tflite::Model* model = nullptr;
tflite::MicroInterpreter* interpreter = nullptr;
TfLiteTensor* input = nullptr;
TfLiteTensor* output = nullptr;

首先读取计算出的模型

model = tflite::GetModel(g_model);

然后，导入输入数据，

input->data.f[0] = x_val;

启动推理

TfLiteStatus invoke_status = interpreter->Invoke();

直接读出输出，

 float y_val = output->data.f[0];

按照x,y的数值进行串口输出和led亮度控制，

HandleOutput(error_reporter, x_val, y_val);

 

 

4. 深度计算的开发应用流程。

4.1 先读取一下model的构型

#include "tensorflow/lite/micro/examples/hello_world/model.h"

// Keep model aligned to 8 bytes to guarantee aligned 64-bit accesses.
alignas(8) const unsigned char g_model[] = {
    0x1c, 0x00, 0x00, 0x00, 0x54, 0x46, 0x4c, 0x33, 0x00, 0x00, 0x12, 0x00,
    0x1c, 0x00, 0x04, 0x00, 0x08, 0x00, 0x0c, 0x00, 0x10, 0x00, 0x14, 0x00,
    0x00, 0x00, 0x18, 0x00, 0x12, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
    0x60, 0x09, 0x00, 0x00, 0xa8, 0x02, 0x00, 0x00, 0x90, 0x02, 0x00, 0x00,
    0x3c, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 
........
}

这个是谷歌提供的tensorflow-lite的标准模式，通常的计算方式，是生成一组深度学习模型的参数数据，但是tensorlow-lite压缩成整形数据，但是对于计算精度损失不大。压缩工具可以从谷歌网站上下载和使用。

4.2 这个首先，需要离线训练模型，搭建深度学习的模型，然后用谷歌的AI引擎训练，收敛后压缩模型，并尝试嵌入到开发板的代码中，不少模型是数M起步，一般的嵌入式开发用不了。

这个过程还是挺痛苦的，这个范例已经完成了这个过程，直接把model模型转换成了数据格式，方便评测。

 

5. 深度学习的移植开发。

虽然这个范例程序使用起来很容易，但是，深度学习的移植是非常困难的，至少模型的收敛和剪裁就需要非常多的经验。学习曲线比较陡。在轻型的系统上，计算性能有限，速度也比较慢，如果能用常规的算法，比深度学习的循环计算效率要高不少。

不过，这个也是一个非常有意义的范例，可以拓展项目开发能力和想象力。

 

 

Jacktang

直接把model模型转换成了数据格式这样就行，基本上可以测试一些数据进行说明了

freebsder

期待北方兄多搞一下tf呀，向你学习一下！