keras.layers.UpSampling2D与tf.image.resize_nearest_neighbor

参考https://www.bookstack.cn/read/keras-docs-zh/sources-layers-convolutional.md

    keras.layers.UpSampling2D(size=(2, 2), data_format=None)

2D输入的上采样层

沿着数据的行和列分别重复size[0],size[1]次

参数:

  size: 整数,或2个整数的元组,行和列的上采样因子

data_format:字符串, 默认channels_last

输入尺寸:

  若data_format是channels_last,输入4D张量,尺寸为(batch,rows,cols,channels)

输出尺寸:

若data_format是channels_last,输出4D张量,尺寸为(batch,upsample_rows,upsampled_cols,channels)

代码实现:

import keras
import numpy as np
import tensorflow as tf
UpSampling2D = keras.layers.UpSampling2D
a=np.array([[1,2,3],[5,6,7]])
a=a.reshape((1,2,-1 ,3))#[[[[1 2 3]],变为4维向量量
                        #[[5 6 7]]]]
print(a)
ta = tf.convert_to_tensor(a)#变为张量
shape=ta.get_shape().as_list()
# print(shape[1],shape[2])
b=UpSampling2D(2)(ta)#用keras方法对a进行上采样
b2=tf.image.resize_nearest_neighbor(ta,(2*shape[1],2*shape[2]))#用tf方法对其进行上采样
with tf.Session() as sess:
    b1=sess.run(b2)#
    print(b1)#[[[[1 2 3]
              #  [1 2 3]]
              #
              # [[1 2 3]
              #  [1 2 3]]
              #
              # [[5 6 7]
              #  [5 6 7]]
              #
              # [[5 6 7]
              #  [5 6 7]]]]