How can I save a TFLite model with a fixed input shape after using interpreter.resize_input_tensor?

Question

I am trying to run inference using the TFLite model of multipose-lightning-tflite-float16 distributed by movenet.

https://www.kaggle.com/models/google/movenet/tfLite

However, this model cannot be used unless you specify the height and width of the input tensor by calling resize_input_tensor before running allocate_tensor. I would like to use a model with a fixed input shape, so is there a way to save the model after running resize_input_tensor, or to fix the input tensor without resize_input_shape?

    # -->> https://www.kaggle.com/models/google/movenet/tfLite
    TFLITE_FILE_PATH = "/home/debian/sandbox/movenet/python/tflite/1.tflite"
    interpreter = tf.lite.Interpreter(TFLITE_FILE_PATH)

    input_details = interpreter.get_input_details()
    output_details = interpreter.get_output_details()
    print(input_details)
    print(output_details)

    is_dynamic_shape_model = input_details[0]['shape_signature'][2] == -1
    if is_dynamic_shape_model:
        input_tensor_index = input_details[0]['index']
        input_shape = input_image.shape
        interpreter.resize_tensor_input(input_tensor_index, input_shape, strict=True)

    ## -->> I WANT TO SAVE MODEL HERE.

    interpreter.allocate_tensors()
    interpreter.set_tensor(input_details[0]['index'], input_image.numpy())

    interpreter.invoke()

    keypoints_with_scores = interpreter.get_tensor(output_details[0]['index'])
    keypoints_with_scores = np.squeeze(keypoints_with_scores)

[{'name': 'serving_default_input:0', 'index': 0, 'shape': array([1, 1, 1, 3], dtype=int32), 'shape_signature': array([ 1, -1, -1,  3], dtype=int32), 'dtype': <class 'numpy.uint8'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]
[{'name': 'StatefulPartitionedCall:0', 'index': 536, 'shape': array([ 1,  6, 56], dtype=int32), 'shape_signature': array([ 1,  6, 56], dtype=int32), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

Nimantha · Accepted Answer · 2025-05-18 08:54:05Z

0

I found out that there is no function to save the model after allocating_tensor.

edited May 18 at 8:54

Nimantha

6,5276 gold badges32 silver badges78 bronze badges

answered Apr 9 at 19:25

Shinobu HUYUGIRI

315 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

How can I save a TFLite model with a fixed input shape after using interpreter.resize_input_tensor?

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related