池化与采样

Outline

• Pooling

• upsample

• ReLU

Reduce Dim

subsample

Max/Avg pooling

• stride = 2

Strides

• stride = 1

For instance

import tensorflow as tf
from tensorflow.keras import layers

x = tf.random.normal([1, 14, 14, 4])
x.shape

TensorShape([1, 14, 14, 4])

pool = layers.MaxPool2D(2, strides=2)
out = pool(x)
out.shape

TensorShape([1, 7, 7, 4])

pool = layers.MaxPool2D(3, strides=2)
out = pool(x)
out.shape

TensorShape([1, 6, 6, 4])

out = tf.nn.max_pool2d(x, 2, strides=2, padding='VALID')
out.shape

TensorShape([1, 7, 7, 4])

upsample

• nearest

• bilinear

UpSampling2D

x = tf.random.normal([1, 7, 7, 4])
x.shape

TensorShape([1, 7, 7, 4])

layer = layers.UpSampling2D(size=3)
out = layer(x)
out.shape

TensorShape([1, 21, 21, 4])

layer = layers.UpSampling2D(size=2)
out = layer(x)
out.shape

TensorShape([1, 14, 14, 4])

ReLu

x = tf.random.normal([2,3])
x

tf.nn.relu(x)
x

layers.ReLU()(x)