Not getting the same result when using .tflite in C and Python. about tensorflow

[sawantkumar]

Hi @saad-koukous

Can you provide more details like the code for handling input and output in both cases ( in python and c) and also the model execution code?

[saad-koukous]

Hi @sawantkumar, Thank you for your response. However, I can't share the .tflite model and data.
but i can share with you how i'm allocating input tensor and output tensor.
interprater method:

// Allocate tensors
    if (TfLiteInterpreterAllocateTensors(interpreter) != kTfLiteOk) {
        fprintf(stderr, "Failed to allocate tensors\n");
        return 1;
    }
const TfLiteTensor* input_tensor = TfLiteInterpreterGetInputTensor(interpreter, 0);
//and here i'm looping on a file to load data in the input_tensor

Signature Runner method

// Allocate tensors
    if (TfLiteSignatureRunnerAllocateTensors(signature_runner) != kTfLiteOk) {
        fprintf(stderr, "Failed to allocate tensors\n");
        return 1;
    const char *signature_inputName = TfLiteSignatureRunnerGetInputName(signature_runner, 0);
    printf("Signature input name: %s\n", signature_inputName);
    }
    TfLiteTensor *input_tensor = TfLiteSignatureRunnerGetInputTensor(signature_runner, signature_inputName);
    //and here i'm looping on a file to load data in the input_tensor

python

interpreter.allocate_tensors()
signature = interpreter.get_signature_runner('signature_name')
infe = serving_default(input=input) #same input of the c program
output = infe['output']

[sawantkumar]

Hi @saad-koukous ,

Can you provide me with some more details like what kind of model this is, an object detection or some nlp model, what do you mean when you say you don't get the same output? Also if you can let me what specific layers are being used and what activation functions are being used in your tflite model, it will be easier for me to debug the issue.

[saad-koukous]

Hi @sawantkumar,
The model is designed for image segmentation, and when I input the same data into the TensorFlow Lite model using both Python and the C API, I'm not getting the same results at all; in fact, the results are significantly different.
the input is a matrix where elements are float32.