2020-11-21

tf.space_to_depth 函数

space_to_depth(input, block_size, data_format=“NHWC”, name=None):

将 height和 width 维度的值重新排列到depth维度。block_size表示输入的块大小。block_size能整除
height 和 width 。

data_format :
“NHWC”: [ batch, height, width, channels ]
“NCHW”: [ batch, channels, height, width ]
“NCHW_VECT_C”: qint8 [ batch, channels / 4, height, width, 4 ]

输入形状为 [n, h, w, c], data_format = “NHWC” 和block_size = bs，则
输出形状为 [n, h/bs, w/bs, c*bs*bs], data_format = “NHWC”

例如, 对于一个给定输入，形状为 [1, 2, 2, 1], data_format = “NHWC” 和block_size = 2:

x = [[[[1], [2]],
      [[3], [4]]]]

这个运算会输出一个张量，形状为 [1, 1, 1, 4]:

[[[[1, 2, 3, 4]]]]

相似的, 输入形状为 [1 4 4 1], block size 为 2:

x = [[[[1],   [2],  [5],  [6]],
      [[3],   [4],  [7],  [8]],
      [[9],  [10], [13],  [14]],
      [[11], [12], [15],  [16]]]]

输出形状 [1 2 2 4]:

x = [[[[1, 2, 3, 4],
       [5, 6, 7, 8]],
      [[9, 10, 11, 12],
       [13, 14, 15, 16]]]]