尝试使用TFlite运行推理时出错

Question

我已经用我的数据自定义训练了ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8_3，并导出了模型，我能够进行推理。我正在使用tensorflow 2.6，现在我想将模型转换为tflite，以便在移动设备上运行它。我已经使用以下代码将模型转换为tflite，代码运行并成功保存了tflite模型。

    converter=tf.lite.TFLiteConverter.from_saved_model(PATH_TO_SAVED_MODE1L)
    converter.optimizations=[tf.lite.Optimize.DEFAULT]
    converter.target_spec.supported_ops=[tf.lite.OpsSet.TFLITE_BUILTINS,tf.lite.OpsSet.SELECT_TF_OPS]
    tflite_model=converter.convert()
    with open('C:/Users/CX/Desktop/model.tflite', 'wb') as f:
      f.write(tflite_model)

现在，我尝试使用下面的代码使用一些图像进行测试：

input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
#interpreter.allocate_tensors()
path=r"C:\t3"
for file in pathlib.Path(path).iterdir():

   img = cv2.imread(r"{}".format(file.resolve()))
   print(np.shape(img))
   img = img.reshape(1,img.shape[0],img.shape[1],3) 
   print(np.shape(img))
   img = img.astype(np.uint32)

   vv= np.array(img, dtype=np.uint8)
   interpreter.set_tensor(input_details[0]['shape_signature'],img)
   interpreter.invoke()

我得到了这个错误：

TypeError: SetTensor(): incompatible function arguments. The following argument types are supported:
    1. (self: tensorflow.lite.python.interpreter_wrapper._pywrap_tensorflow_interpreter_wrapper.InterpreterWrapper, arg0: int, arg1: handle) -> object

Invoked with: <tensorflow.lite.python.interpreter_wrapper._pywrap_tensorflow_interpreter_wrapper.InterpreterWrapper object at 0x000001E4D88E38F0>, array([ 1, -1, -1,  3])

下面是interpreter.get_input_details()函数的详细信息：

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

=和interpreter.get_output_details()

[{'name': 'StatefulPartitionedCall:6', 'index': 462, 'shape': array([    1, 12804,     4]), 'shape_signature': array([    1, 12804,     4]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'StatefulPartitionedCall:0', 'index': 1594, 'shape': array([1, 1]), 'shape_signature': array([ 1, -1]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'StatefulPartitionedCall:5', 'index': 1559, 'shape': array([1]), 'shape_signature': array([1]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'StatefulPartitionedCall:7', 'index': 482, 'shape': array([    1, 12804,    56]), 'shape_signature': array([    1, 12804,    56]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'StatefulPartitionedCall:1', 'index': 1647, 'shape': array([1, 1, 1]), 'shape_signature': array([ 1, -1, -1]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'StatefulPartitionedCall:2', 'index': 1629, 'shape': array([1, 1]), 'shape_signature': array([ 1, -1]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'StatefulPartitionedCall:4', 'index': 1576, 'shape': array([1, 1]), 'shape_signature': array([ 1, -1]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}, {'name': 'StatefulPartitionedCall:3', 'index': 1612, 'shape': array([1, 1, 1]), 'shape_signature': array([ 1, -1, -1]), 'dtype': <class 'numpy.float32'>, 'quantization': (0.0, 0), 'quantization_parameters': {'scales': array([], dtype=float32), 'zero_points': array([], dtype=int32), 'quantized_dimension': 0}, 'sparsity_parameters': {}}]

任何帮助都是非常感谢的

Answer 1

看起来您需要调整输入张量的大小，并且set_tensor中的给定形状不应该具有-1值而不是固定的形状值。