向 tensorflow.python.keras.models.Model 添加一个新层

Question

我正在使用 Python 3.7.7。和张量流 2.3.0。我想从 U-Net 网络中提取编码器并添加它GlobalAveragePooling2D。我使用函数式 API 来定义 U-Net：inputs = Input(shape=img_shape)conv1 = Conv2D(64, (5, 5), activation='relu', padding='same', data_format="channels_last", name='conv1_1')(inputs)conv1 = Conv2D(64, (5, 5), activation='relu', padding='same', data_format="channels_last", name='conv1_2')(conv1)pool1 = MaxPooling2D(pool_size=(2, 2), data_format="channels_last", name='pool1')(conv1)conv2 = Conv2D(96, (3, 3), activation='relu', padding='same', data_format="channels_last", name='conv2_1')(pool1)conv2 = Conv2D(96, (3, 3), activation='relu', padding='same', data_format="channels_last", name='conv2_2')(conv2)pool2 = MaxPooling2D(pool_size=(2, 2), data_format="channels_last", name='pool2')(conv2)conv3 = Conv2D(128, (3, 3), activation='relu', padding='same', data_format="channels_last", name='conv3_1')(pool2)conv3 = Conv2D(128, (3, 3), activation='relu', padding='same', data_format="channels_last", name='conv3_2')(conv3)pool3 = MaxPooling2D(pool_size=(2, 2), data_format="channels_last", name='pool3')(conv3)conv4 = Conv2D(256, (3, 3), activation='relu', padding='same', data_format="channels_last", name='conv4_1')(pool3)conv4 = Conv2D(256, (4, 4), activation='relu', padding='same', data_format="channels_last", name='conv4_2')(conv4)pool4 = MaxPooling2D(pool_size=(2, 2), data_format="channels_last", name='pool4')(conv4)conv5 = Conv2D(512, (3, 3), activation='relu', padding='same', data_format="channels_last", name='conv5_1')(pool4)conv5 = Conv2D(512, (3, 3), activation='relu', padding='same', data_format="channels_last", name='conv5_2')(conv5)我这样做是因为 U-Net 已经经过预训练，所以我需要从预训练模型中获取编码器。但我收到以下错误：AttributeError: 'Model' object has no attribute 'shape'在行：encoder_output = GlobalAveragePooling2D()(encoder_input)我也尝试过，但没有成功：encoder_output = GlobalAveragePooling2D()(encoder_input.get_layer('conv5_2'))和：encoder_output = GlobalAveragePooling2D()(encoder_input.layers[-1].output)如何将GlobalAveragePooling2D图层添加到old_model？

Answer 1

我找到了一种方法。

from tensorflow.python.keras.models import Model

from tensorflow.python.keras.layers import GlobalAveragePooling2D

encoder_input = Model(inputs=old_model.layers[0].input, outputs=old_model.layers[14].output)

encoder_output = GlobalAveragePooling2D()(encoder_input.layers[-1].output)

encoder = Model(encoder_input.input, encoder_output, name='encoder')

summary新模型 ( ) 的是encoder：

Model: "encoder"

_________________________________________________________________

Layer (type)                 Output Shape              Param #   

=================================================================

input_1 (InputLayer)         [(None, 200, 200, 1)]     0         

_________________________________________________________________

conv1_1 (Conv2D)             (None, 200, 200, 64)      1664      

_________________________________________________________________

conv1_2 (Conv2D)             (None, 200, 200, 64)      102464    

_________________________________________________________________

pool1 (MaxPooling2D)         (None, 100, 100, 64)      0         

_________________________________________________________________

conv2_1 (Conv2D)             (None, 100, 100, 96)      55392     

_________________________________________________________________

conv2_2 (Conv2D)             (None, 100, 100, 96)      83040     

_________________________________________________________________

pool2 (MaxPooling2D)         (None, 50, 50, 96)        0         

_________________________________________________________________

conv3_1 (Conv2D)             (None, 50, 50, 128)       110720    

_________________________________________________________________

conv3_2 (Conv2D)             (None, 50, 50, 128)       147584    

_________________________________________________________________

pool3 (MaxPooling2D)         (None, 25, 25, 128)       0         

_________________________________________________________________

conv4_1 (Conv2D)             (None, 25, 25, 256)       295168    

_________________________________________________________________

conv4_2 (Conv2D)             (None, 25, 25, 256)       1048832   

_________________________________________________________________

pool4 (MaxPooling2D)         (None, 12, 12, 256)       0         

_________________________________________________________________

conv5_1 (Conv2D)             (None, 12, 12, 512)       1180160   

_________________________________________________________________

conv5_2 (Conv2D)             (None, 12, 12, 512)       2359808   

_________________________________________________________________

global_average_pooling2d (Gl (None, 512)               0         

=================================================================

Total params: 5,384,832

Trainable params: 5,384,832

Non-trainable params: 0

_________________________________________________________________

None

我希望以下输出形状是正确的：

_________________________________________________________________

global_average_pooling2d (Gl (None, 512)               0         

=================================================================

但是在Transfer learning & fine-tuning中有类似的输出形状。