尝试在tf.data.Dataset上实现CLAHE算法时出错，AttributeError：'Tensor'对象没有属性'ndim'

如何解决尝试在tf.data.Dataset上实现CLAHE算法时出错，AttributeError：'Tensor'对象没有属性'ndim'

我试图在我的模型上实现Lamda Keras图层，以对图像进行预处理。在这种情况下，我决定使用CLAHE算法对图像进行预处理

我使用此预处理功能来实现CLAHE（对比度受限的自适应直方图均衡化）

from skimage import exposure

def AHE(img):

    img_adapteq = exposure.equalize_adapthist(img,clip_limit=0.03)
    return img_adapteq

在将数据集加载到tensorflow时，我使用了以下代码：

test_gen =  tf.keras.preprocessing.image_dataset_from_directory(
    DS_DIR,image_size  = (IMG_HEIGHT,IMG_WIDTH),shuffle = True,)

DS_DIR只是一个字符串，它指示目录的路径，该目录包含两类文件夹，这些文件夹包含将用于测试模型的图像。

这是我的预处理层的代码：

preprocessing_layer = Sequential([
    tf.keras.layers.Lambda(function = AHE),tf.keras.layers.experimental.preprocessing.Rescaling(1/255)
],name = 'preprocessing_layer')

然后使用完整模型：

full_model = Sequential([
    tf.keras.layers.InputLayer(input_shape = (456,456)),preprocessing_layer,model
])

“模型”只是微调的Xception模型。

运行此代码后，出现以下错误：

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-55-643e72c4dacc> in <module>
      2     tf.keras.layers.InputLayer(input_shape = (456,3     preprocessing_layer,----> 4     model
      5 ])

/opt/conda/lib/python3.7/site-packages/tensorflow/python/training/tracking/base.py in _method_wrapper(self,*args,**kwargs)
    455     self._self_setattr_tracking = False  # pylint: disable=protected-access
    456     try:
--> 457       result = method(self,**kwargs)
    458     finally:
    459       self._self_setattr_tracking = previous_value  # pylint: disable=protected-access

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/sequential.py in __init__(self,layers,name)
    140         layers = [layers]
    141       for layer in layers:
--> 142         self.add(layer)
    143 
    144   @property

/opt/conda/lib/python3.7/site-packages/tensorflow/python/training/tracking/base.py in _method_wrapper(self,**kwargs)
    458     finally:
    459       self._self_setattr_tracking = previous_value  # pylint: disable=protected-access

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/sequential.py in add(self,layer)
    219       # If the model is being built continuously on top of an input layer:
    220       # refresh its output.
--> 221       output_tensor = layer(self.outputs[0])
    222       if len(nest.flatten(output_tensor)) != 1:
    223         raise ValueError(SINGLE_LAYER_OUTPUT_ERROR_MSG)

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py in __call__(self,**kwargs)
    924     if _in_functional_construction_mode(self,inputs,args,kwargs,input_list):
    925       return self._functional_construction_call(inputs,--> 926                                                 input_list)
    927 
    928     # Maintains info about the `Layer.call` stack.

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py in _functional_construction_call(self,input_list)
   1115           try:
   1116             with ops.enable_auto_cast_variables(self._compute_dtype_object):
-> 1117               outputs = call_fn(cast_inputs,**kwargs)
   1118 
   1119           except errors.OperatorNotAllowedInGraphError as e:

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/sequential.py in call(self,training,mask)
    384         kwargs['training'] = training
    385 
--> 386       outputs = layer(inputs,**kwargs)
    387 
    388       if len(nest.flatten(outputs)) != 1:

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py in __call__(self,**kwargs)
   1118 
   1119           except errors.OperatorNotAllowedInGraphError as e:

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/layers/core.py in call(self,mask,training)
    901     with backprop.GradientTape(watch_accessed_variables=True) as tape,\
    902         variable_scope.variable_creator_scope(_variable_creator):
--> 903       result = self.function(inputs,**kwargs)
    904     self._check_variables(created_variables,tape.watched_variables())
    905     return result

<ipython-input-53-828c66f442bc> in AHE(img)
      1 def AHE(img):
      2 
----> 3     img_adapteq = exposure.equalize_adapthist(img,clip_limit=0.03)
      4     return img_adapteq

/opt/conda/lib/python3.7/site-packages/skimage/color/adapt_rgb.py in image_filter_adapted(image,**kwargs)
     35         @functools.wraps(image_filter)
     36         def image_filter_adapted(image,**kwargs):
---> 37             if is_rgb_like(image):
     38                 return apply_to_rgb(image_filter,image,**kwargs)
     39             else:

/opt/conda/lib/python3.7/site-packages/skimage/color/adapt_rgb.py in is_rgb_like(image)
     17     shape is fragile.
     18     """
---> 19     return (image.ndim == 3) and (image.shape[2] in (3,4))
     20 
     21 

AttributeError: 'Tensor' object has no attribute 'ndim'

如何解决此问题并在完整模型中实现CLAHE算法？