Unet 架构中的最后一个卷积层期望维度错误
Last convoulutional layer in U-net architecure is expecting wrong dimention
我正尝试在 Keras 中实现 u-net，但在训练模型时出现此错误（调用 model.fit()）：
ValueError: Error when checking target: expected conv2d_302 to have shape > (None, 1, 128, 640) but got array with shape (360, 1, 128, 128)
model.summary() 的输出是：
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_19 (InputLayer) (None, 1, 128, 128) 0
__________________________________________________________________________________________________
conv2d_303 (Conv2D) (None, 32, 128, 128) 320 input_19[0][0]
__________________________________________________________________________________________________
conv2d_304 (Conv2D) (None, 32, 128, 128) 9248 conv2d_303[0][0]
__________________________________________________________________________________________________
max_pooling2d_83 (MaxPooling2D) (None, 32, 64, 64) 0 conv2d_304[0][0]
__________________________________________________________________________________________________
conv2d_305 (Conv2D) (None, 64, 64, 64) 18496 max_pooling2d_83[0][0]
__________________________________________________________________________________________________
conv2d_306 (Conv2D) (None, 64, 64, 64) 36928 conv2d_305[0][0]
__________________________________________________________________________________________________
max_pooling2d_84 (MaxPooling2D) (None, 64, 32, 32) 0 conv2d_306[0][0]
__________________________________________________________________________________________________
conv2d_307 (Conv2D) (None, 128, 32, 32) 73856 max_pooling2d_84[0][0]
__________________________________________________________________________________________________
conv2d_308 (Conv2D) (None, 128, 32, 32) 147584 conv2d_307[0][0]
__________________________________________________________________________________________________
max_pooling2d_85 (MaxPooling2D) (None, 128, 16, 16) 0 conv2d_308[0][0]
__________________________________________________________________________________________________
conv2d_309 (Conv2D) (None, 256, 16, 16) 295168 max_pooling2d_85[0][0]
__________________________________________________________________________________________________
conv2d_310 (Conv2D) (None, 256, 16, 16) 590080 conv2d_309[0][0]
__________________________________________________________________________________________________
max_pooling2d_86 (MaxPooling2D) (None, 256, 8, 8) 0 conv2d_310[0][0]
__________________________________________________________________________________________________
conv2d_311 (Conv2D) (None, 512, 8, 8) 1180160 max_pooling2d_86[0][0]
__________________________________________________________________________________________________
conv2d_312 (Conv2D) (None, 512, 8, 8) 2359808 conv2d_311[0][0]
__________________________________________________________________________________________________
conv2d_transpose_29 (Conv2DTran (None, 256, 16, 16) 524544 conv2d_312[0][0]
__________________________________________________________________________________________________
concatenate_29 (Concatenate) (None, 256, 16, 32) 0 conv2d_transpose_29[0][0]
conv2d_310[0][0]
__________________________________________________________________________________________________
conv2d_313 (Conv2D) (None, 256, 16, 32) 590080 concatenate_29[0][0]
__________________________________________________________________________________________________
conv2d_314 (Conv2D) (None, 256, 16, 32) 590080 conv2d_313[0][0]
__________________________________________________________________________________________________
conv2d_transpose_30 (Conv2DTran (None, 128, 32, 64) 131200 conv2d_314[0][0]
__________________________________________________________________________________________________
concatenate_30 (Concatenate) (None, 128, 32, 96) 0 conv2d_transpose_30[0][0]
conv2d_308[0][0]
__________________________________________________________________________________________________
conv2d_315 (Conv2D) (None, 128, 32, 96) 147584 concatenate_30[0][0]
__________________________________________________________________________________________________
conv2d_316 (Conv2D) (None, 128, 32, 96) 147584 conv2d_315[0][0]
__________________________________________________________________________________________________
conv2d_transpose_31 (Conv2DTran (None, 64, 64, 192) 32832 conv2d_316[0][0]
__________________________________________________________________________________________________
concatenate_31 (Concatenate) (None, 64, 64, 256) 0 conv2d_transpose_31[0][0]
conv2d_306[0][0]
__________________________________________________________________________________________________
conv2d_317 (Conv2D) (None, 64, 64, 256) 36928 concatenate_31[0][0]
__________________________________________________________________________________________________
conv2d_318 (Conv2D) (None, 64, 64, 256) 36928 conv2d_317[0][0]
__________________________________________________________________________________________________
conv2d_transpose_32 (Conv2DTran (None, 32, 128, 512) 8224 conv2d_318[0][0]
__________________________________________________________________________________________________
concatenate_32 (Concatenate) (None, 32, 128, 640) 0 conv2d_transpose_32[0][0]
conv2d_304[0][0]
__________________________________________________________________________________________________
conv2d_319 (Conv2D) (None, 32, 128, 640) 9248 concatenate_32[0][0]
__________________________________________________________________________________________________
conv9 (Conv2D) (None, 32, 128, 640) 9248 conv2d_319[0][0]
__________________________________________________________________________________________________
conv2d_320 (Conv2D) (None, 1, 128, 640) 33 conv9[0][0]
==================================================================================================
Total params: 6,976,161
Trainable params: 6,976,161
Non-trainable params: 0
型号代码如下：
img_rows=128
img_cols= 128
inputs = Input((1, img_rows, img_cols))
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(inputs)
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(pool1)
conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(pool2)
conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(pool3)
conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)
conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(pool4)
conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv5)
up6 = concatenate([Conv2DTranspose(256, (2, 2), strides=(2, 2), padding='same')(conv5), conv4], axis=3)
conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(up6)
conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv6)
up7 = concatenate([Conv2DTranspose(128, (2, 2), strides=(2, 2), padding='same')(conv6), conv3], axis=3)
conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(up7)
conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv7)
up8 = concatenate([Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(conv7), conv2], axis=3)
conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(up8)
conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv8)
up9 = concatenate([Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same')(conv8), conv1], axis=3)
conv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(up9)
conv9 = Conv2D(32, (3, 3), activation='relu', padding='same', name='conv9')(conv9)
conv10 = Conv2D(1, (1, 1), activation='sigmoid')(conv9)
model = Model(inputs=[inputs], outputs=[conv10])
model.compile(optimizer=Adam(lr=1e-5), loss="mean_absolute_error")
model.summary()
model.fit(X_train, y_train, batch_size=36, nb_epoch=5)
我不明白为什么倒数第二层（conv9）的输出与最后一层（conv10）的预期不同。
Keras 模型由 https://github.com/jocicmarko/ultrasound-nerve-segmentation/blob/master/train.py 提供。
已更新：添加了完整的 model.summary()。
可以肯定的是，原作者想要在 channels 维度上进行串联，而不是图像维度之一。
卷积网络中的张量可以采用以下两种格式之一：
(batch_size, width, height, channels)
或
(batch_size, channels, width, height)
在您链接的模型中使用了第一种格式，但您的模型使用了第二种格式。
您可以通过以下两种方式之一修复它：
- 将
concatente 层中的 axis=3 更改为 axis=1
或
- 在卷积层中设置
data_format="channels_last"。 data_format 的默认值，如果省略，取自 keras 配置，很可能这个值对于您和您使用的模型的作者是不同的。参见 https://keras.io/layers/convolutional/#conv2d
更新：事实上，原始模型在您链接到的文件的开头更改了data_format：
K.set_image_data_format('channels_last')
只需在模型的开头添加这一行，即可解决问题。
我正尝试在 Keras 中实现 u-net，但在训练模型时出现此错误（调用 model.fit()）：
ValueError: Error when checking target: expected conv2d_302 to have shape > (None, 1, 128, 640) but got array with shape (360, 1, 128, 128)
model.summary() 的输出是：
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_19 (InputLayer) (None, 1, 128, 128) 0
__________________________________________________________________________________________________
conv2d_303 (Conv2D) (None, 32, 128, 128) 320 input_19[0][0]
__________________________________________________________________________________________________
conv2d_304 (Conv2D) (None, 32, 128, 128) 9248 conv2d_303[0][0]
__________________________________________________________________________________________________
max_pooling2d_83 (MaxPooling2D) (None, 32, 64, 64) 0 conv2d_304[0][0]
__________________________________________________________________________________________________
conv2d_305 (Conv2D) (None, 64, 64, 64) 18496 max_pooling2d_83[0][0]
__________________________________________________________________________________________________
conv2d_306 (Conv2D) (None, 64, 64, 64) 36928 conv2d_305[0][0]
__________________________________________________________________________________________________
max_pooling2d_84 (MaxPooling2D) (None, 64, 32, 32) 0 conv2d_306[0][0]
__________________________________________________________________________________________________
conv2d_307 (Conv2D) (None, 128, 32, 32) 73856 max_pooling2d_84[0][0]
__________________________________________________________________________________________________
conv2d_308 (Conv2D) (None, 128, 32, 32) 147584 conv2d_307[0][0]
__________________________________________________________________________________________________
max_pooling2d_85 (MaxPooling2D) (None, 128, 16, 16) 0 conv2d_308[0][0]
__________________________________________________________________________________________________
conv2d_309 (Conv2D) (None, 256, 16, 16) 295168 max_pooling2d_85[0][0]
__________________________________________________________________________________________________
conv2d_310 (Conv2D) (None, 256, 16, 16) 590080 conv2d_309[0][0]
__________________________________________________________________________________________________
max_pooling2d_86 (MaxPooling2D) (None, 256, 8, 8) 0 conv2d_310[0][0]
__________________________________________________________________________________________________
conv2d_311 (Conv2D) (None, 512, 8, 8) 1180160 max_pooling2d_86[0][0]
__________________________________________________________________________________________________
conv2d_312 (Conv2D) (None, 512, 8, 8) 2359808 conv2d_311[0][0]
__________________________________________________________________________________________________
conv2d_transpose_29 (Conv2DTran (None, 256, 16, 16) 524544 conv2d_312[0][0]
__________________________________________________________________________________________________
concatenate_29 (Concatenate) (None, 256, 16, 32) 0 conv2d_transpose_29[0][0]
conv2d_310[0][0]
__________________________________________________________________________________________________
conv2d_313 (Conv2D) (None, 256, 16, 32) 590080 concatenate_29[0][0]
__________________________________________________________________________________________________
conv2d_314 (Conv2D) (None, 256, 16, 32) 590080 conv2d_313[0][0]
__________________________________________________________________________________________________
conv2d_transpose_30 (Conv2DTran (None, 128, 32, 64) 131200 conv2d_314[0][0]
__________________________________________________________________________________________________
concatenate_30 (Concatenate) (None, 128, 32, 96) 0 conv2d_transpose_30[0][0]
conv2d_308[0][0]
__________________________________________________________________________________________________
conv2d_315 (Conv2D) (None, 128, 32, 96) 147584 concatenate_30[0][0]
__________________________________________________________________________________________________
conv2d_316 (Conv2D) (None, 128, 32, 96) 147584 conv2d_315[0][0]
__________________________________________________________________________________________________
conv2d_transpose_31 (Conv2DTran (None, 64, 64, 192) 32832 conv2d_316[0][0]
__________________________________________________________________________________________________
concatenate_31 (Concatenate) (None, 64, 64, 256) 0 conv2d_transpose_31[0][0]
conv2d_306[0][0]
__________________________________________________________________________________________________
conv2d_317 (Conv2D) (None, 64, 64, 256) 36928 concatenate_31[0][0]
__________________________________________________________________________________________________
conv2d_318 (Conv2D) (None, 64, 64, 256) 36928 conv2d_317[0][0]
__________________________________________________________________________________________________
conv2d_transpose_32 (Conv2DTran (None, 32, 128, 512) 8224 conv2d_318[0][0]
__________________________________________________________________________________________________
concatenate_32 (Concatenate) (None, 32, 128, 640) 0 conv2d_transpose_32[0][0]
conv2d_304[0][0]
__________________________________________________________________________________________________
conv2d_319 (Conv2D) (None, 32, 128, 640) 9248 concatenate_32[0][0]
__________________________________________________________________________________________________
conv9 (Conv2D) (None, 32, 128, 640) 9248 conv2d_319[0][0]
__________________________________________________________________________________________________
conv2d_320 (Conv2D) (None, 1, 128, 640) 33 conv9[0][0]
==================================================================================================
Total params: 6,976,161
Trainable params: 6,976,161
Non-trainable params: 0
型号代码如下：
img_rows=128
img_cols= 128
inputs = Input((1, img_rows, img_cols))
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(inputs)
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(pool1)
conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(pool2)
conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(pool3)
conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)
conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(pool4)
conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv5)
up6 = concatenate([Conv2DTranspose(256, (2, 2), strides=(2, 2), padding='same')(conv5), conv4], axis=3)
conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(up6)
conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv6)
up7 = concatenate([Conv2DTranspose(128, (2, 2), strides=(2, 2), padding='same')(conv6), conv3], axis=3)
conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(up7)
conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv7)
up8 = concatenate([Conv2DTranspose(64, (2, 2), strides=(2, 2), padding='same')(conv7), conv2], axis=3)
conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(up8)
conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv8)
up9 = concatenate([Conv2DTranspose(32, (2, 2), strides=(2, 2), padding='same')(conv8), conv1], axis=3)
conv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(up9)
conv9 = Conv2D(32, (3, 3), activation='relu', padding='same', name='conv9')(conv9)
conv10 = Conv2D(1, (1, 1), activation='sigmoid')(conv9)
model = Model(inputs=[inputs], outputs=[conv10])
model.compile(optimizer=Adam(lr=1e-5), loss="mean_absolute_error")
model.summary()
model.fit(X_train, y_train, batch_size=36, nb_epoch=5)
我不明白为什么倒数第二层（conv9）的输出与最后一层（conv10）的预期不同。
Keras 模型由 https://github.com/jocicmarko/ultrasound-nerve-segmentation/blob/master/train.py 提供。
已更新：添加了完整的 model.summary()。
可以肯定的是，原作者想要在 channels 维度上进行串联，而不是图像维度之一。
卷积网络中的张量可以采用以下两种格式之一：
(batch_size, width, height, channels)
或
(batch_size, channels, width, height)
在您链接的模型中使用了第一种格式，但您的模型使用了第二种格式。
您可以通过以下两种方式之一修复它：
- 将
concatente层中的axis=3更改为axis=1
或
- 在卷积层中设置
data_format="channels_last"。data_format的默认值，如果省略，取自 keras 配置，很可能这个值对于您和您使用的模型的作者是不同的。参见 https://keras.io/layers/convolutional/#conv2d
更新：事实上，原始模型在您链接到的文件的开头更改了data_format：
K.set_image_data_format('channels_last')
只需在模型的开头添加这一行，即可解决问题。