tensorflow tf.pad解析

tf.pad的作用是填充

它的表达式如下：

pad(    tensor,    paddings,    mode='CONSTANT',    name=None)

tensor：要填充的张量

padings ：也是一个张量，代表每一维填充多少行/列，但是有一个要求它的rank一定要和tensor的rank是一

mode ：可以取三个值，分别是"CONSTANT" ,"REFLECT","SYMMETRIC"
              mode="CONSTANT" 是填充0
              mode="REFLECT"是映射填充，上下（1维）填充顺序和paddings是相反的，左右（零维）顺序补齐
              mode="SYMMETRIC"是对称填充，上下（1维）填充顺序是和paddings相同的，左右（零维）对称补齐

本例tensor的rank=4，即4维向量tensor.shape=[b,h,w,c]，则padings=[a=[a1,a2],b=[b1,b2],c=[c1,c2],d=[d1,d2]]同样也是4维

填充后的行踪为[b+a1+a2,h+b1+b2,w+c1+c2,c+d1+d2](相应维度增加)

import tensorflow as tf

input=np.ones([1,5,5,1])
print(input.shape)
kernel_size=3
pad_begin = (kernel_size - 1) // 2
pad_end = kernel_size - 1 - pad_begin
inputs=tf.pad(input,[[0,0],[pad_begin,pad_end],[pad_begin,pad_end],[0,0]])
print(pad_begin)
print(pad_end)
print(inputs)
with tf.Session() as sess:
    inputs=sess.run(inputs)
    print(inputs)

[

[

[0]

[0]

[0]

[0]

[0]

[0]

[0]

]

[

[

[1]

[1]

[1]

[1]

[1]

]

[

[0]

[1]

[1]

[1]

[1]

[1]

[0]

]

[

[1]

[1]

[1]

[1]

[1]

]

[

[0]

[1]

[1]

[1]

[1]

[1]

[0]

]

[

[1]

[1]

[1]

[1]

[1]

]

[

[0]

[1]

[1]

[1]

[1]

[1]

[0]

]

[

[1]

[1]

[1]

[1]

[1]

]

[

[0]

[1]

[1]

[1]

[1]

[1]

[0]

]

[

[1]

[1]

[1]

[1]

[1]

]

]

[

[0]

[1]

[1]

[1]

[1]

[1]

[0]

]

[

[0]

[0]

[0]

[0]

[0]

[0]

[0]

]

]