3D CNN 模型抛出负维度错误——维度问题

Question

我正在创建一个高度 = 128、宽度 = 128、通道 = 3 的 3D CNN 模型。3D CNN 的代码-

def get_model(width=128, height=128, depth=3):
  """
  Build a 3D convolutional neural network
  """
  inputs = tf.keras.Input((width, height, depth, 1))

  x = layers.Conv3D(filters=64, kernel_size=3, activation="relu")(inputs)
  x = layers.MaxPool3D(pool_size=2)(x)
  x = layers.BatchNormalization()(x)

  x = layers.Conv3D(filters=128, kernel_size=3, activation="relu")(x)
  x = layers.MaxPool3D(pool_size=2)(x)
  x = layers.BatchNormalization()(x)

  x = layers.Conv3D(filters=256, kernel_size=3, activation="relu")(x)
  x = layers.MaxPool3D(pool_size=2)(x)
  x = layers.BatchNormalization()(x)

  x = layers.GlobalAveragePooling3D()(x)
  x = layers.Dense(units=512, activation="relu")(x)

  x = layers.Dropout(0.3)(x)

  outputs = layers.Dense(units=4, activation='softmax')(x)

  model= keras.Model(inputs, outputs, name="3DCNN")
  return model

所以在创建模型函数后，当我尝试构建模型时，它会引发值错误

-ValueError: Negative dimension size caused by subtracting 2 from 1 for '{{node max_pooling3d_5/MaxPool3D}} = MaxPool3D[T=DT_FLOAT, data_format="NDHWC", ksize=[1, 2, 2, 2, 1], padding="VALID", strides=[1, 2, 2, 2, 1]](Placeholder)' with input shapes: [?,126,126,1,64].

构建模型的代码：- #构建模型

model = get_model(width=128, height=128, depth=3)
model.summary()

完全错误-

   InvalidArgumentError                      Traceback (most recent call last)
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py in _create_c_op(graph, node_def, inputs, control_inputs, op_def)
       1879   try:
    -> 1880     c_op = pywrap_tf_session.TF_FinishOperation(op_desc)
       1881   except errors.InvalidArgumentError as e:
    
    InvalidArgumentError: Negative dimension size caused by subtracting 2 from 1 for '{{node max_pooling3d_5/MaxPool3D}} = MaxPool3D[T=DT_FLOAT, data_format="NDHWC", ksize=[1, 2, 2, 2, 1], padding="VALID", strides=[1, 2, 2, 2, 1]](Placeholder)' with input shapes: [?,126,126,1,64].
    
    During handling of the above exception, another exception occurred:
    
    ValueError                                Traceback (most recent call last)
    14 frames
    /usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/ops.py in _create_c_op(graph, node_def, inputs, control_inputs, op_def)
       1881   except errors.InvalidArgumentError as e:
       1882     # Convert to ValueError for backwards compatibility.
    -> 1883     raise ValueError(str(e))
       1884 
       1885   return c_op
    
    ValueError: Negative dimension size caused by subtracting 2 from 1 for '{{node max_pooling3d_5/MaxPool3D}} = MaxPool3D[T=DT_FLOAT, data_format="NDHWC", ksize=[1, 2, 2, 2, 1], padding="VALID", strides=[1, 2, 2, 2, 1]](Placeholder)' with input shapes:

 [?,126,126,1,64].

这个错误是什么意思？？我的维度有问题吗？？

提前谢谢！！！！！！

Answer 1

不指定data_format 参数，Conv3D 层将输入形状视为：

batch_shape + (conv_dim1, conv_dim2, conv_dim3, channels)

您已指定为：

batch_shape + (width=128, height=128, depth=3, channels=1)

因此，您有一个数据，其形状为 (128,128,3) 并且有 1 个通道。

由于卷积操作适用于前 3 个维度，即 (128,128,3)，在第一次卷积 kernel_size=3 之后，第 3 个维度（您指定为 depth=3 的维度）收缩到 1。然后在下一层 ( MaxPooling3D) 它无法通过 2 进行池化，因为形状不适合。因此，请考虑将深度尺寸更改为更大的数字或更改kernel_size 参数。例如，输入形状可以是 (128,128,128,1) 或 kernel_size 应该更改为 (3,3,1) 之类的其他内容。

P.S：如果你有一个 RGB 图像，那么通道数是 3，最后一个维度应该设置为 3。在 3D 图像中还有另一个概念，称为深度（另一个维度），它与通道不同。所以：

• 3D 图像 RGB：(width, height, depth, 3)
• 3D图像灰度：(width, height, depth, 1)
• 2D 图像 RGB：(width, height, 3)
• 二维图像灰度：(width, height, 1)