Question

我正在寻找一种张量流python方法来放大（调整大小）张量，使每个特征映射中的每个元素沿着两个轴加倍，例如：

 ([[1, 2, 3],
   [4, 5, 6],
   [7, 8, 9]])

=＆GT;

 ([[1, 1, 2, 2, 3, 3],
   [1, 1, 2, 2, 3, 3],
   [4, 4, 5, 5, 6, 6],
   [4, 4, 5, 5, 6, 6],
   [7, 7, 8, 8, 9, 9],
   [7, 7, 8, 8, 9, 9]])

我看到tf.tile和tf.pad，但我无法弄清楚如何使用此方法来获得该结果。

感谢任何提示！

更新：

感谢sygi提供的有用提示，这是一个独立于形状的解决方案，使用python3内核在jupyter笔记本中工作：

import tensorflow as tf
import numpy as np

i = np.array([[1, 2, 3],
              [4, 5, 6],
              [7, 8, 9]])

j = np.array([[1,  2,  3,  4],
              [5,  6,  7,  8],
              [9, 10, 11, 12]])

k = np.array([[1, 2],
              [3, 4],
              [5, 6]])

a = tf.placeholder(tf.int64, shape=(None, None))
a_shape = tf.shape(a)

b = tf.reshape(a, [a_shape[0], a_shape[1], 1])
c = tf.tile(b, [1, 1, 2])
d = tf.reshape(c, [a_shape[0], a_shape[1]*2])

e = tf.reshape(d, [a_shape[0], a_shape[1]*2, 1])
f = tf.tile(e, [1, 1, 2])
g = tf.transpose(f, [0, 2, 1])

h = tf.reshape(g, [a_shape[0]*2, a_shape[1]*2])

session = tf.InteractiveSession()
session.run(tf.initialize_all_variables())

print(h.eval(feed_dict={a: i}))
print(h.eval(feed_dict={a: j}))
print(h.eval(feed_dict={a: k}))

session.close()

结果

[[1 1 2 2 3 3]
 [1 1 2 2 3 3]
 [4 4 5 5 6 6]
 [4 4 5 5 6 6]
 [7 7 8 8 9 9]
 [7 7 8 8 9 9]]

[[ 1  1  2  2  3  3  4  4]
 [ 1  1  2  2  3  3  4  4]
 [ 5  5  6  6  7  7  8  8]
 [ 5  5  6  6  7  7  8  8]
 [ 9  9 10 10 11 11 12 12]
 [ 9  9 10 10 11 11 12 12]]

[[1 1 2 2]
 [1 1 2 2]
 [3 3 4 4]
 [3 3 4 4]
 [5 5 6 6]
 [5 5 6 6]]

Answer 1

a = tf.convert_to_tensor([[1, 2, 3],
                          [4, 5, 6],
                          [7, 8, 9]])
b = tf.reshape(a, [3, 3, 1])
c = tf.tile(b, [1, 1, 2])
d = tf.reshape(c, [3, 6])
print(d.eval())
array([[1, 1, 2, 2, 3, 3],
       [4, 4, 5, 5, 6, 6],
       [7, 7, 8, 8, 9, 9]], dtype=int32)

e = tf.reshape(d, [3, 6, 2])
f = tf.tile(e, [1, 1, 2])
g = tf.transpose(f, [0, 2, 1])
print(g.eval())
array([[[1, 1, 2, 2, 3, 3],
        [1, 1, 2, 2, 3, 3]],

       [[4, 4, 5, 5, 6, 6],
        [4, 4, 5, 5, 6, 6]],

       [[7, 7, 8, 8, 9, 9],
        [7, 7, 8, 8, 9, 9]]], dtype=int32)

h = tf.reshape(g, [6, 6])
print(h.eval())
array([[1, 1, 2, 2, 3, 3],
       [1, 1, 2, 2, 3, 3],
       [4, 4, 5, 5, 6, 6],
       [4, 4, 5, 5, 6, 6],
       [7, 7, 8, 8, 9, 9],
       [7, 7, 8, 8, 9, 9]], dtype=int32)

您可以使用以下方式获得a张量的形状（如果已定义）：

shape = a.get_shape().as_list()

Answer 2

只需将tf.image.ResizeMethod与最近邻插值一起使用

array = tf.image.resize_images(old_array, (old_size*2, old_size*2),
                               method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)

方法的输入必须是形状为[批处理，高度，宽度，通道]的4-D张量或形状为[高度，宽度，通道]的3-D张量。

https://www.tensorflow.org/api_docs/python/tf/image/resize

在张量流中扩大张量

2 个答案: