Is there an equivalent of tf.lite.Interpreter.get_input_details for C++?

Question

In TensorFlow lite's Python API, there are methods to retrieve details concerning the input and output tensors, called tf.lite.Interpreter.get_input_details and tf.lite.Interpreter.get_output_details . I am after the 'quantization' entry in these dicts, which contain a value for mean and standard deviation each.

When deploying these models on another platform, I use the C++ API to set up an interpreter. To prepare the inputs for my model, I also need mean and standard deviation.

I was unable to find equivalents of these functions in C++. Do they exist?

Answer 1

You can get the quantization parameter out of TfLiteTensor structure, like...

// Get the input tensor indices.
const std::vector<int>& inputs = interpreter->inputs();
// Get the tensor object of the 1st input.
TfLiteTensor* tensor = interpreter->tensor(input[0]);
const TfLiteQuantizationParams& params = tensor->param

Answer 2

您还可以使用PrintInterpreterState()获取相同的信息，实际上非常PrintInterpreterState() ：

tflite::PrintInterpreterState(interpreter.get());