class UnetEncoderBlock(tf.keras.layers.Layer):
def __init__(
self,
num_filters,
use_pooling=True
):
super().__init__()
self.num_filters = num_filters
self.use_pooling = use_pooling
def build(self, input_shape):
self.conv_1 = tf.keras.layers.Conv2D(
self.num_filters,
(3, 3),
kernel_initializer='he_normal',
padding='same',
use_bias=False
)
self.activation_1 = tf.keras.layers.ReLU()
self.conv_2 = tf.keras.layers.Conv2D(
self.num_filters,
(3, 3),
kernel_initializer='he_normal',
padding='same',
use_bias=False
)
self.activation_2 = tf.keras.layers.ReLU()
self.batch_norm = tf.keras.layers.BatchNormalization()
self.pooling = tf.keras.layers.MaxPooling2D((2, 2))
self.dropout = tf.keras.layers.Dropout(0.1)
super().build(input_shape)
def call(self, inputs, *args, **kwargs):
conv = self.conv_1(inputs)
conv = self.activation_1(conv)
conv = self.conv_2(conv)
conv = self.activation_2(conv)
if self.use_pooling:
pool = self.pooling(conv)
pool = self.dropout(pool)
else:
pool = None
return conv, pool
class UnetDecoderBlock(tf.keras.layers.Layer):
def __init__(self, num_filters, concatenate_with=None):
super().__init__()
self.num_filters = num_filters
self.concatenate_with = concatenate_with if concatenate_with else []
def build(self, input_shape):
self.conv_transpose = tf.keras.layers.Conv2DTranspose(
self.num_filters,
(3, 3),
padding='same',
strides=(2, 2),
kernel_initializer='he_normal',
use_bias=False
)
self.weights_concat = tf.keras.layers.Concatenate()
self.dropout = tf.keras.layers.Dropout(0.1)
self.conv_1 = tf.keras.layers.Conv2D(
self.num_filters,
(3, 3),
kernel_initializer='he_normal',
padding='same',
use_bias=False
)
self.activation_1 = tf.keras.layers.ReLU()
self.conv_2 = tf.keras.layers.Conv2D(
self.num_filters,
(3, 3),
kernel_initializer='he_normal',
padding='same',
use_bias=False
)
self.activation_2 = tf.keras.layers.ReLU()
self.batch_norm = tf.keras.layers.BatchNormalization()
super().build(input_shape)
def call(self, inputs, *args, **kwargs):
unfl = self.conv_transpose(inputs)
unfl = self.weights_concat([unfl, *self.concatenate_with])
unfl = self.dropout(unfl)
conv = self.conv_1(unfl)
conv = self.activation_1(conv)
conv = self.conv_2(conv)
conv = self.activation_2(conv)
conv = self.batch_norm(conv)
return convmodel_input = tf.keras.layers.Input(input_shape)
conv_1, pool_1 = UnetEncoderBlock(16)(model_input) # 256x256 -> 128x128
conv_2, pool_2 = UnetEncoderBlock(32)(pool_1) # 128x128 -> 64x64
conv_3, pool_3 = UnetEncoderBlock(64)(pool_2) # 64x64 -> 32x32
conv_4, pool_4 = UnetEncoderBlock(128)(pool_3) # 32x32 -> 16x16
conv_5, _ = UnetEncoderBlock(256, use_pooling=False)(pool_4)
conv_6 = UnetDecoderBlock(128, concatenate_with=[conv_4])(conv_5) # 16x16 -> 32x32
conv_7 = UnetDecoderBlock(64, concatenate_with=[conv_3])(conv_6) # 32x32 -> 64x64
conv_8 = UnetDecoderBlock(32, concatenate_with=[conv_2])(conv_7) # 64x64 -> 128x128
conv_9 = UnetDecoderBlock(16, concatenate_with=[conv_1])(conv_8) # 128x128 -> 256x256
model_output = tf.keras.layers.Conv2D(1, (1, 1))(conv_9)
model_output = tf.keras.layers.Activation('sigmoid')(model_output)
model = tf.keras.models.Model(inputs=[model_input], outputs=[model_output], name='Unet')TypeError: Could not build a TypeSpec for KerasTensor(type_spec=TensorSpec(shape=(None, 32, 32, 128), dtype=tf.float32, name=None), name='batch_normalization/FusedBatchNormV3:0', description="created by layer 'batch_normalization'") of unsupported type <class 'keras.engine.keras_tensor.KerasTensor'>.Traceback (most recent call last):
File "C:\Users\yahry\PycharmProjects\lesa_niisva\tests\tests.py", line 48, in <module>
model = build_unet()
File "C:\Users\yahry\PycharmProjects\lesa_niisva\lesa\models\models.py", line 263, in build_unet
conv_6 = UnetDecoderBlock(128, concatenate_with=[conv_4])(conv_5) # 16x16 -> 32x32
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\utils\traceback_utils.py", line 61, in error_handler
return fn(*args, **kwargs)
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\engine\base_layer.py", line 1058, in __call__
return self._functional_construction_call(
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\engine\base_layer.py", line 2572, in _functional_construction_call
outputs = self._keras_tensor_symbolic_call(
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\engine\base_layer.py", line 2419, in _keras_tensor_symbolic_call
return self._infer_output_signature(
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\engine\base_layer.py", line 2484, in _infer_output_signature
outputs = tf.nest.map_structure(
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\tensorflow\python\util\nest.py", line 917, in map_structure
structure[0], [func(*x) for x in entries],
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\tensorflow\python\util\nest.py", line 917, in <listcomp>
structure[0], [func(*x) for x in entries],
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\engine\keras_tensor.py", line 663, in keras_tensor_from_tensor
out = keras_tensor_cls.from_tensor(tensor)
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\keras\engine\keras_tensor.py", line 199, in from_tensor
type_spec = tf.type_spec_from_value(tensor)
File "C:\Users\yahry\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\tensorflow\python\framework\type_spec.py", line 958, in type_spec_from_value
raise TypeError(f"Could not build a TypeSpec for {value} of "
TypeError: Could not build a TypeSpec for KerasTensor(type_spec=TensorSpec(shape=(None, 32, 32, 128), dtype=tf.float32, name=None), name='batch_normalization/FusedBatchNormV3:0', description="created by layer 'batch_normalization'") of unsupported type <class 'keras.engine.keras_tensor.KerasTensor'>.
Простой
