tf.transpose(a, perm = None, name = 'transpose')

a是一個張量（Tensor），實際上就是一個數組。

perm是a置換的維度

name是操作的名稱

最後返回一個轉置的張量

影象處理時資料集中儲存資料的形式為：[channel,image_height,image_width]，在intel GPU加速的情況下，因為GPU對於影象的處理比較多，希望在訪問同一個channel的畫素是連續的，一般儲存選用NCHW【參考連結：NCHW和NHWC】。而在tensorflow中使用CNN時我們需要將其轉化為[image_height,image_width,channel]的形式，這個時候我們可以使用tf.transpose函式，即tf.transpose(a.[1,2,0])

如果perm沒有給定，那麼預設是perm是=[n-1,n-2,n-3.....,0]，其中rank(a)=n

對於二維輸入資料，預設perm就是常規的矩陣轉置操作。

tf.transpose的第二個引數perm=[0,1,2],0代表三維陣列的高（即為二維陣列的個數），1代表二維陣列的行，2代表二維陣列的列。
tf.transpose(x, perm=[1,0,2])代表將三位陣列的高和行進行轉置。

wiki文件如下：

tf.transpose(a, perm=None, name='transpose') 
 
Transposes a. Permutes the dimensions according to perm.
 
The returned tensor's dimension i will correspond to the input dimension perm[i]. If perm is not given, it is set to (n-1...0), where n is the rank of the input tensor. Hence by default, this operation performs a regular matrix transpose on 2-D input Tensors.
For example:
# 'x' is [[1 2 3]
#         [4 5 6]]
tf.transpose(x) ==> [[1 4]
                     [2 5]
                     [3 6]]
# Equivalently
tf.transpose(x perm=[1, 0]) ==> [[1 4]
                                 [2 5]
                                 [3 6]]
# 'perm' is more useful for n-dimensional tensors, for n > 2
# 'x' is   [[[1  2  3]
#            [4  5  6]]
#           [[7  8  9]
#            [10 11 12]]]
# Take the transpose of the matrices in dimension-0
tf.transpose(b, perm=[0, 2, 1]) ==> [[[1  4]
                                      [2  5]
                                      [3  6]]
                                     [[7 10]
                                      [8 11]
                                      [9 12]]]
Args: 
•a: A Tensor.
•perm: A permutation of the dimensions of a.
•name: A name for the operation (optional).
Returns: 
A transposed Tensor.
 

測試程式碼如下：

import tensorflow as tf
 
#x = tf.constant([[1, 2 ,3],[4, 5, 6]])
x = [[[1,2,3,4],[5,6,7,8],[9,10,11,12]],[[21,22,23,24],[25,26,27,28],[29,30,31,32]]]
#a=tf.constant(x)
a=tf.transpose(x, [0, 1, 2])
b=tf.transpose(x, [0, 2, 1])
c=tf.transpose(x, [1, 0, 2])
d=tf.transpose(x, [1, 2, 0])
e=tf.transpose(x, [2, 1, 0])
f=tf.transpose(x, [2, 0, 1])
 
# 'perm' is more useful for n-dimensional tensors, for n > 2
# 'x' is   [[[1  2  3]
#            [4  5  6]]
#           [[7  8  9]
#            [10 11 12]]]
# Take the transpose of the matrices in dimension-0
#tf.transpose(b, perm=[0, 2, 1])
with tf.Session() as sess:
    print ('---------------')
    print (sess.run(a))
    print ('---------------')
    print (sess.run(b))
    print ('---------------')
    print (sess.run(c))
    print ('---------------')
    print (sess.run(d))
    print ('---------------')
    print (sess.run(e))
    print ('---------------')
    print (sess.run(f))
    print ('---------------')
 

測試結果如下：

---------------
[[[ 1  2  3  4]
  [ 5  6  7  8]
  [ 9 10 11 12]]
 
 [[21 22 23 24]
  [25 26 27 28]
  [29 30 31 32]]]
---------------
[[[ 1  5  9]
  [ 2  6 10]
  [ 3  7 11]
  [ 4  8 12]]
 
 [[21 25 29]
  [22 26 30]
  [23 27 31]
  [24 28 32]]]
---------------
[[[ 1  2  3  4]
  [21 22 23 24]]
 
 [[ 5  6  7  8]
  [25 26 27 28]]
 
 [[ 9 10 11 12]
  [29 30 31 32]]]
---------------
[[[ 1 21]
  [ 2 22]
  [ 3 23]
  [ 4 24]]
 
 [[ 5 25]
  [ 6 26]
  [ 7 27]
  [ 8 28]]
 
 [[ 9 29]
  [10 30]
  [11 31]
  [12 32]]]
---------------
[[[ 1 21]
  [ 5 25]
  [ 9 29]]
 
 [[ 2 22]
  [ 6 26]
  [10 30]]
 
 [[ 3 23]
  [ 7 27]
  [11 31]]
 
 [[ 4 24]
  [ 8 28]
  [12 32]]]
---------------
[[[ 1  5  9]
  [21 25 29]]
 
 [[ 2  6 10]
  [22 26 30]]
 
 [[ 3  7 11]
  [23 27 31]]
 
 [[ 4  8 12]
  [24 28 32]]]
---------------