boards/evkmimx8mq/demo_apps/hello_world_tflite/hello_world_tflite.cc

#include "tensorflow/lite/micro/all_ops_resolver.h"
#include "tensorflow/lite/micro/micro_error_reporter.h"
#include "tensorflow/lite/micro/micro_interpreter.h"
#include "tensorflow/lite/schema/schema_generated.h"
#include "tensorflow/lite/version.h"
#include "model.h"

namespace {
    tflite::ErrorReporter *error_reporter = nullptr;
    const tflite::Model *model = nullptr;
    tflite::MicroInterpreter *interpreter = nullptr;

    TfLiteTensor *input = nullptr;
    TfLiteTensor *output = nullptr;
    int inference_count = 0;
    const int kInferencesPerCycle = 1000;
    const float kXrange = 2.f * 3.14159265359f;

    const int kModelArenaSize = 4096;
    const int kExtraArenaSize = 4096;
    const int kTensorArenaSize = kModelArenaSize + kExtraArenaSize;
    uint8_t tensor_arena[kTensorArenaSize] __attribute__((aligned(16)));
}  // namespace

void HandleOutput(tflite::ErrorReporter *error_reporter, float x_val, float y_val) {
    int brightness = (int)(127.5f * (y_val + 1));
    TF_LITE_REPORT_ERROR(error_reporter, "%d", brightness);
}

extern "C" void hello_world_tflite_setup(void) {
    static tflite::MicroErrorReporter micro_error_reporter;
    error_reporter = &micro_error_reporter;
    TF_LITE_REPORT_ERROR(error_reporter, "Hello from TFLite micro!");

    model = tflite::GetModel(g_model);
    if (model->version() != TFLITE_SCHEMA_VERSION) {
        TF_LITE_REPORT_ERROR(error_reporter,
            "Model schema version is %d, supported is %d",
            model->version(), TFLITE_SCHEMA_VERSION);
        return;
    }

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

    TfLiteStatus allocate_status = interpreter->AllocateTensors();
    if (allocate_status != kTfLiteOk) {
        TF_LITE_REPORT_ERROR(error_reporter, "AllocateTensors failed");
        return;
    }

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

    TF_LITE_REPORT_ERROR(error_reporter, "setup() complete.");
}

extern "C" void hello_world_tflite_loop(void) {
    float position = static_cast<float>(inference_count) /
                     static_cast<float>(kInferencesPerCycle);
    float x_val = position * kXrange;

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

    TfLiteStatus invoke_status = interpreter->Invoke();
    if (invoke_status != kTfLiteOk) {
        TF_LITE_REPORT_ERROR(error_reporter, "Invoke failed on x_val: %f",
                                             static_cast<double>(x_val));
        return;
    }

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

    HandleOutput(error_reporter, x_val, y_val);

    inference_count += 1;
    if (inference_count >= kInferencesPerCycle) {
        inference_count = 0;
    }
}