了解如何在 keras 模型中正确使用 keras.layers.Normalization per image

问：

我想将归一化预处理层合并到我的 keras 模型中，以规范化每个图像。这些图像是只有一个通道的 3D 荧光显微镜图像，因此 （batch_size，x，y，z，1）。相关：我认为我还应该在任何增强层之后应用这种归一化，因为模型应该预测的任何数据也会首先归一化，对吗？

如果我想在整个训练集中进行归一化，我可以用

InSize = xTrain.shape[1:]
inputs = keras.Input(shape = InSize + (1,), name = "InputLayer")
normalizer = keras.layers.Normalization(axis = None, name = 'NormLayer')
normalizer.adapt(xTrain)

x = normalizer(inputs)

模型的其余部分如下。 是训练数据。但是，当我更改为 .我需要调整normalizer.adapt，但我不确定如何调整。我试过了xTrainaxis = Noneaxis = 0

normalizer.adapt(xTrain,batch_size = batch_size)

但这产生了错误

/mnt/e/FLYWORKS Dropbox/Tamiki Komatsuzaki/jarenson/Code/Classifier/CanNonCan_v1.0.1.ipynb Cell 18 line 2
     22 # thresholder = tf.keras.layers.ThresholdedReLU(theta = Threshold, name = 'ThresholdLayer')
     23 # thresholder = tf.keras.layers.ReLU(max_value = max_value, negative_slope = 0.0, threshold = Threshold)
     24 print(xTrain.shape)
---> 25 normalizer.adapt(xTrain,batch_size = batch_size)
     27 inputs = keras.Input(shape = InSize + (1,), name = "InputLayer")
     28 # x = thresholder(inputs)
     29 # normalizer.adapt(x,batch_size = batch_size)

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/layers/preprocessing/normalization.py:286, in Normalization.adapt(self, data, batch_size, steps)
    240 def adapt(self, data, batch_size=None, steps=None):
    241     """Computes the mean and variance of values in a dataset.
    242 
    243     Calling `adapt()` on a `Normalization` layer is an alternative to
   (...)
    284           argument is not supported with array inputs.
    285     """
--> 286     super().adapt(data, batch_size=batch_size, steps=steps)

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/base_preprocessing_layer.py:258, in PreprocessingLayer.adapt(self, data, batch_size, steps)
    256 with data_handler.catch_stop_iteration():
    257     for _ in data_handler.steps():
--> 258         self._adapt_function(iterator)
    259         if data_handler.should_sync:
    260             context.async_wait()

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/tensorflow/python/util/traceback_utils.py:153, in filter_traceback.<locals>.error_handler(*args, **kwargs)
    151 except Exception as e:
    152   filtered_tb = _process_traceback_frames(e.__traceback__)
--> 153   raise e.with_traceback(filtered_tb) from None
    154 finally:
    155   del filtered_tb

File /tmp/__autograph_generated_filerl2staau.py:10, in outer_factory.<locals>.inner_factory.<locals>.tf__adapt_step(iterator)
      8 with ag__.FunctionScope('adapt_step', 'fscope', ag__.ConversionOptions(recursive=True, user_requested=True, optional_features=(), internal_convert_user_code=True)) as fscope:
      9     data = ag__.converted_call(ag__.ld(next), (ag__.ld(iterator),), None, fscope)
---> 10     ag__.converted_call(ag__.ld(self)._adapt_maybe_build, (ag__.ld(data),), None, fscope)
     11     ag__.converted_call(ag__.ld(self).update_state, (ag__.ld(data),), None, fscope)

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/base_preprocessing_layer.py:295, in PreprocessingLayer._adapt_maybe_build(self, data)
    292 if batch_input_shape is None:
    293     # Set the number of dimensions.
    294     self._batch_input_shape = data_shape_nones
--> 295 self.build(data_shape)
    296 self.built = True

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/layers/preprocessing/normalization.py:187, in Normalization.build(self, input_shape)
    185 for d in self._keep_axis:
    186     if input_shape[d] is None:
--> 187         raise ValueError(
    188             "All `axis` values to be kept must have known shape. "
    189             "Got axis: {}, "
    190             "input shape: {}, with unknown axis at index: {}".format(
    191                 self.axis, input_shape, d
    192             )
    193         )
    194 # Axes to be reduced.
    195 self._reduce_axis = [d for d in range(ndim) if d not in self._keep_axis]

ValueError: in user code:

    File "/home/jgarenson/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/base_preprocessing_layer.py", line 122, in adapt_step  *
        self._adapt_maybe_build(data)
    File "/home/jgarenson/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/base_preprocessing_layer.py", line 295, in _adapt_maybe_build  **
        self.build(data_shape)
    File "/home/jgarenson/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/layers/preprocessing/normalization.py", line 187, in build
        raise ValueError(

    ValueError: All `axis` values to be kept must have known shape. Got axis: (0,), input shape: [None, 27, 54, 198], with unknown axis at index: 0

我尝试将代码更改为

inputs = keras.Input(shape = InSize + (1,), name = "InputLayer")
normalizer = keras.layers.Normalization(axis = None, name = 'NormLayer')
normalizer.adapt(xTrain)

x = normalizer(inputs)

但这会产生错误：

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/layers/preprocessing/normalization.py:286, in Normalization.adapt(self, data, batch_size, steps)
    240 def adapt(self, data, batch_size=None, steps=None):
    241     """Computes the mean and variance of values in a dataset.
    242 
    243     Calling `adapt()` on a `Normalization` layer is an alternative to
   (...)
    284           argument is not supported with array inputs.
    285     """
--> 286     super().adapt(data, batch_size=batch_size, steps=steps)

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/base_preprocessing_layer.py:246, in PreprocessingLayer.adapt(self, data, batch_size, steps)
    244 if self.built:
    245     self.reset_state()
--> 246 data_handler = data_adapter.DataHandler(
    247     data,
    248     batch_size=batch_size,
    249     steps_per_epoch=steps,
    250     epochs=1,
    251     steps_per_execution=self._steps_per_execution,
    252     distribute=False,
    253 )
    254 self._adapt_function = self.make_adapt_function()
    255 for _, iterator in data_handler.enumerate_epochs():

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/data_adapter.py:1259, in DataHandler.__init__(self, x, y, sample_weight, batch_size, steps_per_epoch, initial_epoch, epochs, shuffle, class_weight, max_queue_size, workers, use_multiprocessing, model, steps_per_execution, distribute)
   1256     self._steps_per_execution = steps_per_execution
   1258 adapter_cls = select_data_adapter(x, y)
-> 1259 self._adapter = adapter_cls(
   1260     x,
   1261     y,
   1262     batch_size=batch_size,
   1263     steps=steps_per_epoch,
   1264     epochs=epochs - initial_epoch,
   1265     sample_weights=sample_weight,
   1266     shuffle=shuffle,
   1267     max_queue_size=max_queue_size,
   1268     workers=workers,
   1269     use_multiprocessing=use_multiprocessing,
   1270     distribution_strategy=tf.distribute.get_strategy(),
   1271     model=model,
   1272 )
   1274 strategy = tf.distribute.get_strategy()
   1276 self._current_step = 0

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/data_adapter.py:471, in GenericArrayLikeDataAdapter.__init__(self, *args, **kwargs)
    462 def __init__(self, *args, **kwargs):
    463     logging.warning(
    464         "Keras is training/fitting/evaluating on array-like data. Keras "
    465         "may not be optimized for this format, so if your input data "
   (...)
    468         "load a Dataset instead."
    469     )
--> 471     super().__init__(*args, **kwargs)

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/data_adapter.py:257, in TensorLikeDataAdapter.__init__(self, x, y, sample_weights, sample_weight_modes, batch_size, epochs, steps, shuffle, **kwargs)
    251 (sample_weights, _, _) = training_utils.handle_partial_sample_weights(
    252     y, sample_weights, sample_weight_modes, check_all_flat=True
    253 )
    255 inputs = pack_x_y_sample_weight(x, y, sample_weights)
--> 257 num_samples = set(
    258     int(i.shape[0]) for i in tf.nest.flatten(inputs)
    259 ).pop()
    260 _check_data_cardinality(inputs)
    262 # If batch_size is not passed but steps is, calculate from the input
    263 # data.  Default to 32 for backwards compat.

File ~/PythonEnv/Classifier1.0/env/lib/python3.10/site-packages/keras/engine/data_adapter.py:258, in <genexpr>(.0)
    251 (sample_weights, _, _) = training_utils.handle_partial_sample_weights(
    252     y, sample_weights, sample_weight_modes, check_all_flat=True
    253 )
    255 inputs = pack_x_y_sample_weight(x, y, sample_weights)
    257 num_samples = set(
--> 258     int(i.shape[0]) for i in tf.nest.flatten(inputs)
    259 ).pop()
    260 _check_data_cardinality(inputs)
    262 # If batch_size is not passed but steps is, calculate from the input
    263 # data.  Default to 32 for backwards compat.

TypeError: int() argument must be a string, a bytes-like object or a real number, not 'NoneType'

最后，我不完全确定在这种情况下应该如何应用.adapt方法。有没有人经历过这个问题？