tf. keras. layers. MaxPooling? D function

Maximum pool layer It is usually used to downsample data , Remove redundant information 、 Compress features .

1、tf.keras.layers.MaxPooling1D function

The function prototype

tf.keras.layers.MaxPool1D(pool_size=2,
						  strides=None,
						  padding='valid',
						  data_format='channels_last',
						  **kwargs
)

Function description

Parameters pool_size Indicates the size of the pooled window , Parameters strides Represents the moving step size , The default is None, Represents and pool_size It's the same size .

Be careful ,padding The meaning of parameters and the meaning of convolution layer padding The meaning of parameters is a little different , When padding by VALID, That is, when there is no filling , The size of the output tensor is ( The size of the input tensor -pool_size + 1 )/ strides, When padding by SAME when , The size of the output tensor is The size of the input tensor / strides.

2、tf.keras.layers.MaxPooling2D function

The function prototype

tf.keras.layers.MaxPool2D(pool_size=(2, 2),
						  strides=None,
						  padding='valid',
						  data_format=None,
						  **kwargs
)

Function description

The process of two-dimensional maximum pooling is shown in the figure above . Where the size of the input matrix is 4×4,pool_size by 2×2,strides by 2,padding The method is no filling , The size of the output matrix is 2×2. It can be seen that , The pooled input matrix , Only the higher part of the eigenvalue is retained . in other words , The pooling operation reduces the sampling rate of the image .

Parameters here pool_size It's a tuple (2, 2), Indicates that the size of the pooled window is 2×2.

3、tf.keras.layers.MaxPooling3D function

The function prototype

tf.keras.layers.MaxPool3D(pool_size=(2, 2, 2),
						  strides=None,
						  padding='valid',
						  data_format=None,
						  **kwargs
)

Function description

The usage scene of 3D pool layer is generally multi frame （ single / Multichannel ） Of frame data （ Video frame ）, And the output is also multi frame , Sequential pair continuous k The entire channel of the frame performs the pooling operation at the same time ;

The meaning of each parameter of the function is the same as MaxPooling2D、MaxPooling1D The meaning of each parameter is similar .