Node DEQUANTIZE (number 0f) failed to prepare with status 1 (TFMIC-40)

[Criminal-9527]

Checklist

• Checked the issue tracker for similar issues to ensure this is not a duplicate.
• Provided a clear description of your suggestion.
• Included any relevant context or examples.

Issue or Suggestion Description

我们在使用模型检测时出现了如下报错：
input->type == kTfLiteInt8 || Int16 || Uint8 was not true，Node DEQUANTIZE (number 0f) failed to prepare with status 1。
我们使用的idf版本是v5.2，板子是esp32s3-wroom,使用的模型是mediapipe的hand_landmark_lite.tflite模型，在netron上查看其结构如下：



我已经按照netron将模型所需op全部加入resolver

这个DEQUANTIZE量化操作的输入并不是我在手动加模型输入的时候能够决定的呀，为什么会导致这样的报错呢，难道是esp-tflite-micro的这个op与模型的同名op之间有差别吗？
完整代码有点长，我们想做实时识别手势，所以还有摄像头的好多代码，下面是只有有关模型的代码，hand_landmark_lite.tflite这个模型在电脑上是确定可以运行的

#include "tensorflow/lite/micro/micro_mutable_op_resolver.h"
#include "tensorflow/lite/micro/micro_interpreter.h"
#include "tensorflow/lite/micro/system_setup.h"
#include "tensorflow/lite/schema/schema_generated.h"
#include "tensorflow/lite/micro/micro_log.h"
#include "tensorflow/lite/c/common.h"

#include "model.h"


const tflite::Model* model = nullptr;
tflite::MicroInterpreter* interpreter = nullptr;
TfLiteTensor* input = nullptr;
TfLiteTensor* output = nullptr;
int inference_count = 0;

constexpr int kTensorArenaSize = 20 * 1024;
uint8_t tensor_arena[kTensorArenaSize];

void model_setup();
void model_handle(uvc_frame_t* frame);


void model_setup() {
  model = tflite::GetModel(g_model);
  if (model->version() != TFLITE_SCHEMA_VERSION) {
    MicroPrintf("Model provided is schema version %d not equal to supported "
                "version %d.", model->version(), TFLITE_SCHEMA_VERSION);
    return;
  }

  static tflite::MicroMutableOpResolver<8> resolver;
  if (resolver.AddAdd() != kTfLiteOk) 
    return;
  if (resolver.AddMean() != kTfLiteOk)
    return;
  if (resolver.AddLogistic() != kTfLiteOk) 
    return;
  if (resolver.AddConv2D() != kTfLiteOk) 
    return;
  if (resolver.AddFullyConnected() != kTfLiteOk) 
    return;
  if (resolver.AddDequantize() != kTfLiteOk) 
    return;
  if (resolver.AddDepthwiseConv2D() != kTfLiteOk) 
    return;
  if (resolver.AddMaxPool2D() != kTfLiteOk) 
    return;

  static tflite::MicroInterpreter static_interpreter(
      model, resolver, tensor_arena, kTensorArenaSize);
  interpreter = &static_interpreter;

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

  input = interpreter->input(0);
  output = interpreter->output(0);
}

float* normalize(uint8_t* data, int size){
  float* ret = new float[size];
  for(int i = 0; i < size; i++)
    ret[i] = (float)data[i] / 255.0f;
  return ret;
}

void model_handle(uvc_frame_t* frame) {
  input->type = kTfLiteFloat32;
  (input->data).f = normalize((uint8_t*)frame->data, frame->data_bytes);
  
  input->dims->data[0] = 1;
  input->dims->data[1] = 224;
  input->dims->data[2] = 224;
  input->dims->data[3] = 3;
  
  (input->dims)->size = 4;
  input->bytes = 224 * 224 * 3;

  TfLiteStatus invoke_status = interpreter->Invoke();
  if (invoke_status != kTfLiteOk) {
    MicroPrintf("Invoke failed");
    return;
  }

  float* landmark = new float[63];
  landmark = (output->data).f;
  for(int i = 0; i < 21; i += 3)
    MicroPrintf("x: %f, y: %f, z: %f", landmark[i], landmark[i + 1], landmark[i + 2]);
    
  delete[] landmark;
}

感谢！

[vikramdattu]

@Criminal-9527 the tflite-micro models are expected to use quantised weights (inputs/outputs and filter values) and only the scales are supposed to be float.

To make sure your model can work with tflite-micro framework, you need to quantise it to use int8. This is a simple process and can be achieved with small piece of code after you train the model.
Example can be found here: https://ai.google.dev/edge/litert/models/post_training_integer_quant#convert_using_integer-only_quantization