在构建和训练 3D Keras U-NET 时遇到 ValueError答案

【问题标题】：Getting a ValueError on building and training 3D Keras U-NET在构建和训练 3D Keras U-NET 时遇到 ValueError
【发布时间】：2020-02-25 08:38:15
【问题描述】：

在训练我使用 keras 为 3D Unet 构建的模型时，我得到 ValueError：conv3d_46 层的输入 0 与该层不兼容：预期 ndim=5，发现 ndim=6。收到的完整形状：[None, 2, 256, 256, 120, 4]。我的数据的形状大小为 (2, 256, 256, 120, 4)。

型号：

data = Input(shape=inp_shape)
flt=32


conv1 = Conv3D(flt, (3, 3, 3), activation='relu', padding='same')(data)
conv1 = Conv3D(flt, (3, 3, 3), activation='relu', padding='same')(conv1)
pool1 = MaxPooling3D(pool_size=(2, 2, 2))(conv1)

conv2 = Conv3D(flt*2, (3, 3, 3), activation='relu', padding='same')(pool1)
conv2 = Conv3D(flt*2, (3, 3, 3), activation='relu', padding='same')(conv2)
pool2 = MaxPooling3D(pool_size=(2, 2, 2))(conv2)

conv3 = Conv3D(flt*4, (3, 3, 3), activation='relu', padding='same')(pool2)
conv3 = Conv3D(flt*4, (3, 3, 3), activation='relu', padding='same')(conv3)
pool3 = MaxPooling3D(pool_size=(2, 2, 2))(conv3)

conv4 = Conv3D(flt*8, (3, 3, 3), activation='relu', padding='same')(pool3)
conv4 = Conv3D(flt*8, (3, 3, 3), activation='relu', padding='same')(conv4)
pool4 = MaxPooling3D(pool_size=(2, 2, 2))(conv4)

conv5 = Conv3D(flt*16, (3, 3, 3), activation='relu', padding='same')(pool4)
conv5 = Conv3D(flt*8, (3, 3, 3), activation='relu', padding='same')(conv5)

up6 = concatenate([Conv3DTranspose(flt*8, (2, 2, 2), strides=(2, 2, 2), padding='same')(conv5), conv4], axis=-1)
conv6 = Conv3D(flt*8, (3, 3, 3), activation='relu', padding='same')(up6)
conv6 = Conv3D(flt*4, (3, 3, 3), activation='relu', padding='same')(conv6)

up7 = concatenate([Conv3DTranspose(flt*4, (2, 2, 2), strides=(2, 2, 2), padding='same')(conv6), conv3], axis=-1)
conv7 = Conv3D(flt*4, (3, 3, 3), activation='relu', padding='same')(up7)
conv7 = Conv3D(flt*2, (3, 3, 3), activation='relu', padding='same')(conv7)

up8 = concatenate([Conv3DTranspose(flt*2, (2, 2, 2), strides=(2, 2, 2), padding='same')(conv7), conv2], axis=4)
conv8 = Conv3D(flt*2, (3, 3, 3), activation='relu', padding='same')(up8)
conv8 = Conv3D(flt, (3, 3, 3), activation='relu', padding='same')(conv8)

up9 = concatenate([Conv3DTranspose(flt, (2, 2, 2), strides=(2, 2, 2), padding='same')(conv8), conv1], axis=4)
conv9 = Conv3D(flt, (3, 3, 3), activation='relu', padding='same')(up9)
conv9 = Conv3D(flt, (3, 3, 3), activation='relu', padding='same')(conv9)


conv10 = Conv3D(2, (1,1,1), activation='sigmoid')(conv9)

model = Model(inputs=[data], outputs=[conv10])

训练模型的代码如下：-

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['binary_accuracy'])

【问题讨论】：

我们需要binary_crossentropy 的损失函数用于 UNet 和 sigmoid 激活。 categorical_crossentropy 用于多类分类。
我改变了，但仍然遇到同样的错误。我认为模型有问题。我想不通。@ShubhamPanchal 任务是执行分割
目标标签的最后一维为2。模型输出的最后一维为1。可能目标标签已经过一次热编码？
@ShubhamPanchal 哦，让我编辑和测试
@ShubhamPanchal 感谢更改模型运行的二维后。

标签： python tensorflow keras unity3d-unet

【解决方案1】：

目标标签的最后一维为 2。模型的输出的最后一维为 1。感谢@Shubham Panchal

【讨论】：