tensorflow-tf.nn.conv2d卷積運算(2)

tf.nn.conv2d(
    input,
    filter,
    strides,
    padding,
    use_cudnn_on_gpu=True,
    data_format='NHWC',
    dilations=[1, 1, 1, 1],
    name=None
)

計算給定4d輸入和過濾核張量的二維卷積。

給定形狀[batch，in_height, in_width, in_channels]的輸入張量和形狀[filter_height, filter_width, in_channels, out_channels]的篩選/核張量，此op執行如下操作:
將過濾核壓扁到一個二維矩陣，形狀爲[filter_height filter_width in_channels, output_channels]。
從輸入張量中提取圖像塊，形成一個形狀的虛擬張量[batch, out_height, out_width, filter_height filter_width in_channels]。
對於每個patch，右乘濾波器矩陣和圖像patch向量。

output[b, i, j, k] =
    sum_{di, dj, q} input[b, strides[1] * i + di, strides[2] * j + dj, q] *
                    filter[di, dj, q, k]

必須有strides(步長)[0]=strides[3]= 1。對於相同水平和頂點的最常見情況，stride = [1, stride, stride, 1]。

Args:
input:一個張量。必須是以下類型之一:half，bfloat16, float32, float64。一個四維張量。dimension順序是根據data_format的值來解釋的。
filter: 必須具有與輸入相同的類型。形狀的4維張量[filter_height, filter_width, in_channels, out_channels]。
strides: int型列表。長度爲4的一維張量。每個輸入維度的滑動窗口的跨步。順序是根據data_format的值來解釋的。
padding: 來自:“SAME”、“VALID”的字符串。要使用的填充算法的類型。

use_cudnn_on_gpu: 可選bool,默認爲True.
data_format: 一個可選的字符串:“NHWC”、“NCHW”。默認爲“NHWC”。指定輸入和輸出數據的數據格式。使用默認格式“NHWC”，數據按以下順序存儲:[批處理、高度、寬度、通道]。或者，格式可以是“NCHW”，數據存儲順序爲:[批處理，通道，高度，寬度]。
dilations:int的可選列表。默認爲[1,1,1,1]。長度爲4的一維張量。每個輸入維度的膨脹係數。如果設置爲k > 1，則該維度上的每個過濾器元素之間將有k-1跳過單元格。維度順序由data_format的值決定，詳細信息請參閱上面的內容。批次的膨脹和深度尺寸必須爲1。
name: 可選 名字

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Tue Oct  2 13:23:27 2018

@author: myhaspl
@email:[email protected]
tf.nn.conv2d
"""

import tensorflow as tf

g=tf.Graph()

with g.as_default():
    x=tf.constant([[
            [[1.,2.],[3.,4.],[5.,6.]],
            [[10.,20.],[30.,40.],[50.,60.]],
            ]])
    kernel=tf.constant([[[[10.],[2.]]]])
    y=tf.nn.conv2d(x,kernel,strides=[1,1,1,1],padding="SAME")

with tf.Session(graph=g) as sess:
    print sess.run(x)
    print sess.run(kernel)
    print kernel.get_shape()
    print x.get_shape()
    print sess.run(y)

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

  [[10. 20.]
   [30. 40.]
   [50. 60.]]]]
[[[[10.]
   [ 2.]]]]
(1, 1, 2, 1)
(1, 2, 3, 2)
[[[[ 14.]
   [ 38.]
   [ 62.]]

  [[140.]
   [380.]
   [620.]]]]

import tensorflow as tf

a = tf.constant([1,1,1,0,0,0,1,1,1,0,0,0,1,1,1,0,0,1,1,0,0,1,1,0,0],dtype=tf.float32,shape=[1,5,5,1])
b = tf.constant([1,0,1,0,1,0,1,0,1],dtype=tf.float32,shape=[3,3,1,1])
c = tf.nn.conv2d(a,b,strides=[1, 2, 2, 1],padding='VALID')
d = tf.nn.conv2d(a,b,strides=[1, 2, 2, 1],padding='SAME')
with tf.Session() as sess:
    print ("c shape:")
    print (c.shape)
    print ("c value:")
    print (sess.run(c))
    print ("d shape:")
    print (d.shape)
    print ("d value:")
    print (sess.run(d))

conv2d(
input,
filter,
strides,
padding,
use_cudnn_on_gpu=True,
data_format='NHWC',
name=None
)

c shape:
(1, 2, 2, 1)
c value:
[[[[ 4.]
   [ 4.]]

  [[ 2.]
   [ 4.]]]]
d shape:
(1, 3, 3, 1)
d value:
[[[[ 2.]
   [ 3.]
   [ 1.]]

  [[ 1.]
   [ 4.]
   [ 3.]]

  [[ 0.]
   [ 2.]
   [ 1.]]]]

padding爲VALID，採用丟棄的方式。

padding爲SAME,採用的是補全的方式,補0

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Tue Oct  2 13:23:27 2018

@author: myhaspl
@email:[email protected]
tf.nn.conv2d
"""

import tensorflow as tf

g=tf.Graph()

with g.as_default():
    x=tf.constant([
               [
                    [[1.],[2.],[11.]],
                    [[3.],[4.],[22.]],
                    [[5.],[6.],[33.]]
               ],
               [
                    [[10.],[20.],[44.]],
                    [[30.],[40.],[55.]],
                    [[50.],[60.],[66.]]                    
               ]
            ])#2*3*3*1
    kernel=tf.constant(
            [
                   [[[2.]],[[3.]]]
            ]
            )#2*2*1
    y=tf.nn.conv2d(x,kernel,strides=[1,1,1,1],padding="SAME")

with tf.Session(graph=g) as sess:
    print x.get_shape()
    print kernel.get_shape()
    print sess.run(y)    
    print y.get_shape()

padding="SAME"
$12+23=8$
$22+311=37$
$112+03=22$

(2, 3, 3, 1)
(1, 2, 1, 1)
[[[[  8.]
   [ 37.]
   [ 22.]]

  [[ 18.]
   [ 74.]
   [ 44.]]

  [[ 28.]
   [111.]
   [ 66.]]]

 [[[ 80.]
   [172.]
   [ 88.]]

  [[180.]
   [245.]
   [110.]]

  [[280.]
   [318.]
   [132.]]]]
(2, 3, 3, 1)