tf.keras.layers.AveragePooling2D

2D平均池化层

keras.layers.AveragePooling2D(pool_size=(2, 2), 
                              strides=None, 
                              padding='valid', 
                              data_format=None)

---

参数详解

• pool_size: 池化窗口大小
• strides: 池化步长，默认值等于 pool_size
• padding: 'VALID' 或 'SAME'，'VALID'表示无填充，'SAME'表示用0填充
• data_format: 表示输入张量的维度顺序，默认为 [batch, height, width, channel]

---

示例

from tensorflow.keras.layers import AveragePooling2D
import tensorflow as tf
import numpy as np

# 定义一个平均池化层，用0填充
pool1 = AveragePooling2D(pool_size=(2, 2),
                         strides=None,
                         padding='SAME',
                         data_format=None)

# 定义一个平均池化层，不填充
pool2 = AveragePooling2D(pool_size=(2, 2),
                         strides=None,
                         padding='VALID',
                         data_format=None)

# 生成一个维度为[64, 101, 101, 3]的矩阵
x = np.random.random((64, 101, 101, 3))

# 转成tensor类型,第一个维度64表示batch
# numpy中的数据类型和tensorflow中的数据类型完全兼容,所以这一步可以省略
x = tf.convert_to_tensor(x)
print(x.shape) # [64, 101, 101, 3]

# 进行平均池化
y1 = pool1(x)
print(y1.shape) # [64, 51, 51, 3]

# 进行平均池化
y2 = pool2(x)
print(y2.shape) # [64, 50, 50, 3]