激情久久久_欧美视频区_成人av免费_不卡视频一二三区_欧美精品在欧美一区二区少妇_欧美一区二区三区的

腳本之家，腳本語言編程技術及教程分享平臺！
分類導航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服務器之家 - 腳本之家 - Python - 淺談pytorch中torch.max和F.softmax函數的維度解釋

淺談pytorch中torch.max和F.softmax函數的維度解釋

2020-06-28 11:11Jasminexjf Python

這篇文章主要介紹了淺談pytorch中torch.max和F.softmax函數的維度解釋，具有很好的參考價值，希望對大家有所幫助。一起跟隨小編過來看看吧

在利用torch.max函數和F.Ssoftmax函數時，對應該設置什么維度，總是有點懵，遂總結一下：

首先看看二維tensor的函數的例子：

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
import torch
import torch.nn.functional as F
 
input = torch.randn(3,4)
print(input)
tensor([[-0.5526, -0.0194, 2.1469, -0.2567],
    [-0.3337, -0.9229, 0.0376, -0.0801],
    [ 1.4721, 0.1181, -2.6214, 1.7721]])
 
b = F.softmax(input,dim=0) # 按列SoftMax,列和為1
print(b)
tensor([[0.1018, 0.3918, 0.8851, 0.1021],
    [0.1268, 0.1587, 0.1074, 0.1218],
    [0.7714, 0.4495, 0.0075, 0.7762]])
 
c = F.softmax(input,dim=1)  # 按行SoftMax,行和為1
print(c)
tensor([[0.0529, 0.0901, 0.7860, 0.0710],
    [0.2329, 0.1292, 0.3377, 0.3002],
    [0.3810, 0.0984, 0.0064, 0.5143]])
 
d = torch.max(input,dim=0)  # 按列取max,
print(d)
torch.return_types.max(
values=tensor([1.4721, 0.1181, 2.1469, 1.7721]),
indices=tensor([2, 2, 0, 2]))
 
e = torch.max(input,dim=1)  # 按行取max，
print(e)
torch.return_types.max(
values=tensor([2.1469, 0.0376, 1.7721]),
indices=tensor([2, 2, 3]))

下面看看三維tensor解釋例子：

函數softmax輸出的是所給矩陣的概率分布；

b輸出的是在dim=0維上的概率分布，b[0][5][6]+b[1][5][6]+b[2][5][6]=1

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
a=torch.rand(3,16,20)
b=F.softmax(a,dim=0)
c=F.softmax(a,dim=1)
d=F.softmax(a,dim=2)
 
In [1]: import torch as t
In [2]: import torch.nn.functional as F
In [4]: a=t.Tensor(3,4,5)
In [5]: b=F.softmax(a,dim=0)
In [6]: c=F.softmax(a,dim=1)
In [7]: d=F.softmax(a,dim=2)
 
In [8]: a
Out[8]:
tensor([[[-0.1581, 0.0000, 0.0000, 0.0000, -0.0344],
 
     [ 0.0000, -0.0344, 0.0000, -0.0344, 0.0000],
     [-0.0344, 0.0000, -0.0344, 0.0000, -0.0344],
     [ 0.0000, -0.0344, 0.0000, -0.0344, 0.0000]],
 
    [[-0.0344, 0.0000, -0.0344, 0.0000, -0.0344],
     [ 0.0000, -0.0344, 0.0000, -0.0344, 0.0000],
     [-0.0344, 0.0000, -0.0344, 0.0000, -0.0344],
     [ 0.0000, -0.0344, 0.0000, -0.0344, 0.0000]],
 
 
    [[-0.0344, 0.0000, -0.0344, 0.0000, -0.0344],
     [ 0.0000, -0.0344, 0.0000, -0.0344, 0.0000],
     [-0.0344, 0.0000, -0.0344, 0.0000, -0.0344],
     [ 0.0000, -0.0344, 0.0000, -0.0344, 0.0000]]])
 
In [9]: b
Out[9]:
 
tensor([[[0.3064, 0.3333, 0.3410, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333]],
 
    [[0.3468, 0.3333, 0.3295, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333]],
 
    [[0.3468, 0.3333, 0.3295, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333],
     [0.3333, 0.3333, 0.3333, 0.3333, 0.3333]]])
 
 
In [10]: b.sum()
Out[10]: tensor(20.0000)
 
In [11]: b[0][0][0]+b[1][0][0]+b[2][0][0]
Out[11]: tensor(1.0000)
 
In [12]: c.sum()
Out[12]: tensor(15.)
 
In [13]: c
Out[13]:
tensor([[[0.2235, 0.2543, 0.2521, 0.2543, 0.2457],
 
     [0.2618, 0.2457, 0.2521, 0.2457, 0.2543],
 
     [0.2529, 0.2543, 0.2436, 0.2543, 0.2457],
 
     [0.2618, 0.2457, 0.2521, 0.2457, 0.2543]],
 
 
    [[0.2457, 0.2543, 0.2457, 0.2543, 0.2457],
 
     [0.2543, 0.2457, 0.2543, 0.2457, 0.2543],
 
     [0.2457, 0.2543, 0.2457, 0.2543, 0.2457],
 
     [0.2543, 0.2457, 0.2543, 0.2457, 0.2543]],
 
 
    [[0.2457, 0.2543, 0.2457, 0.2543, 0.2457],
 
     [0.2543, 0.2457, 0.2543, 0.2457, 0.2543],
 
     [0.2457, 0.2543, 0.2457, 0.2543, 0.2457],
 
     [0.2543, 0.2457, 0.2543, 0.2457, 0.2543]]])
 
In [14]: n=t.rand(3,4)
 
In [15]: n
Out[15]:
 
tensor([[0.2769, 0.3475, 0.8914, 0.6845],
    [0.9251, 0.3976, 0.8690, 0.4510],
    [0.8249, 0.1157, 0.3075, 0.3799]])
 
In [16]: m=t.argmax(n,dim=0)
 
In [17]: m
Out[17]: tensor([1, 1, 0, 0])
 
In [18]: p=t.argmax(n,dim=1)
 
In [19]: p
Out[19]: tensor([2, 0, 0])
 
In [20]: d.sum()
Out[20]: tensor(12.0000)
 
In [22]: d
Out[22]:
 
tensor([[[0.1771, 0.2075, 0.2075, 0.2075, 0.2005],
 
     [0.2027, 0.1959, 0.2027, 0.1959, 0.2027],
 
     [0.1972, 0.2041, 0.1972, 0.2041, 0.1972],
 
     [0.2027, 0.1959, 0.2027, 0.1959, 0.2027]],
 
 
    [[0.1972, 0.2041, 0.1972, 0.2041, 0.1972],
 
     [0.2027, 0.1959, 0.2027, 0.1959, 0.2027],
 
     [0.1972, 0.2041, 0.1972, 0.2041, 0.1972],
 
     [0.2027, 0.1959, 0.2027, 0.1959, 0.2027]],
 
 
    [[0.1972, 0.2041, 0.1972, 0.2041, 0.1972],
 
     [0.2027, 0.1959, 0.2027, 0.1959, 0.2027],
 
     [0.1972, 0.2041, 0.1972, 0.2041, 0.1972],
 
     [0.2027, 0.1959, 0.2027, 0.1959, 0.2027]]])
 
In [23]: d[0][0].sum()
Out[23]: tensor(1.)

補充知識：多分類問題torch.nn.Softmax的使用

為什么談論這個問題呢？是因為我在工作的過程中遇到了語義分割預測輸出特征圖個數為16，也就是所謂的16分類問題。

因為每個通道的像素的值的大小代表了像素屬于該通道的類的大小，為了在一張圖上用不同的顏色顯示出來，我不得不學習了torch.nn.Softmax的使用。

首先看一個簡答的例子，倘若輸出為(3, 4, 4)，也就是3張4x4的特征圖。

?
1
2
3
import torch
img = torch.rand((3,4,4))
print(img)

輸出為：

?
1
2
3
4
5
6
7
8
9
10
11
12
tensor([[[0.0413, 0.8728, 0.8926, 0.0693],
     [0.4072, 0.0302, 0.9248, 0.6676],
     [0.4699, 0.9197, 0.3333, 0.4809],
     [0.3877, 0.7673, 0.6132, 0.5203]],
    [[0.4940, 0.7996, 0.5513, 0.8016],
     [0.1157, 0.8323, 0.9944, 0.2127],
     [0.3055, 0.4343, 0.8123, 0.3184],
     [0.8246, 0.6731, 0.3229, 0.1730]],
    [[0.0661, 0.1905, 0.4490, 0.7484],
     [0.4013, 0.1468, 0.2145, 0.8838],
     [0.0083, 0.5029, 0.0141, 0.8998],
     [0.8673, 0.2308, 0.8808, 0.0532]]])

我們可以看到共三張特征圖，每張特征圖上對應的值越大，說明屬于該特征圖對應類的概率越大。

?
1
2
3
4
import torch.nn as nn
sogtmax = nn.Softmax(dim=0)
img = sogtmax(img)
print(img)

輸出為：

?
1
2
3
4
5
6
7
8
9
10
11
12
tensor([[[0.2780, 0.4107, 0.4251, 0.1979],
     [0.3648, 0.2297, 0.3901, 0.3477],
     [0.4035, 0.4396, 0.2993, 0.2967],
     [0.2402, 0.4008, 0.3273, 0.4285]],
    [[0.4371, 0.3817, 0.3022, 0.4117],
     [0.2726, 0.5122, 0.4182, 0.2206],
     [0.3423, 0.2706, 0.4832, 0.2522],
     [0.3718, 0.3648, 0.2449, 0.3028]],
    [[0.2849, 0.2076, 0.2728, 0.3904],
     [0.3627, 0.2581, 0.1917, 0.4317],
     [0.2543, 0.2898, 0.2175, 0.4511],
     [0.3880, 0.2344, 0.4278, 0.2686]]])

可以看到，上面的代碼對每張特征圖對應位置的像素值進行Softmax函數處理， 圖中標紅位置加和=1，同理，標藍位置加和=1。

我們看到Softmax函數會對原特征圖每個像素的值在對應維度(這里dim=0，也就是第一維)上進行計算，將其處理到0～1之間，并且大小固定不變。

print(torch.max(img,0))

輸出為：

?
1
2
3
4
5
6
7
8
9
torch.return_types.max(
values=tensor([[0.4371, 0.4107, 0.4251, 0.4117],
    [0.3648, 0.5122, 0.4182, 0.4317],
    [0.4035, 0.4396, 0.4832, 0.4511],
    [0.3880, 0.4008, 0.4278, 0.4285]]),
indices=tensor([[1, 0, 0, 1],
    [0, 1, 1, 2],
    [0, 0, 1, 2],
    [2, 0, 2, 0]]))

可以看到這里3x4x4變成了1x4x4，而且對應位置上的值為像素對應每個通道上的最大值，并且indices是對應的分類。

清楚理解了上面的流程，那么我們就容易處理了。

看具體案例，這里輸出output的大小為：16x416x416.

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
output = torch.tensor(output)
 
sm = nn.Softmax(dim=0)
output = sm(output)
 
mask = torch.max(output,0).indices.numpy()
 
# 因為要轉化為RGB彩色圖，所以增加一維
rgb_img = np.zeros((output.shape[1], output.shape[2], 3))
for i in range(len(mask)):
  for j in range(len(mask[0])):
    if mask[i][j] == 0:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 255
      rgb_img[i][j][2] = 255
    if mask[i][j] == 1:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 180
      rgb_img[i][j][2] = 0
    if mask[i][j] == 2:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 180
      rgb_img[i][j][2] = 180
    if mask[i][j] == 3:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 180
      rgb_img[i][j][2] = 255
    if mask[i][j] == 4:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 255
      rgb_img[i][j][2] = 180
    if mask[i][j] == 5:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 255
      rgb_img[i][j][2] = 0
    if mask[i][j] == 6:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 0
      rgb_img[i][j][2] = 180
    if mask[i][j] == 7:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 0
      rgb_img[i][j][2] = 255
    if mask[i][j] == 8:
      rgb_img[i][j][0] = 255
      rgb_img[i][j][1] = 0
      rgb_img[i][j][2] = 0
    if mask[i][j] == 9:
      rgb_img[i][j][0] = 180
      rgb_img[i][j][1] = 0
      rgb_img[i][j][2] = 0
    if mask[i][j] == 10:
      rgb_img[i][j][0] = 180
      rgb_img[i][j][1] = 255
      rgb_img[i][j][2] = 255
    if mask[i][j] == 11:
      rgb_img[i][j][0] = 180
      rgb_img[i][j][1] = 0
      rgb_img[i][j][2] = 180
    if mask[i][j] == 12:
      rgb_img[i][j][0] = 180
      rgb_img[i][j][1] = 0
      rgb_img[i][j][2] = 255
    if mask[i][j] == 13:
      rgb_img[i][j][0] = 180
      rgb_img[i][j][1] = 255
      rgb_img[i][j][2] = 180
    if mask[i][j] == 14:
      rgb_img[i][j][0] = 0
      rgb_img[i][j][1] = 180
      rgb_img[i][j][2] = 255
    if mask[i][j] == 15:
      rgb_img[i][j][0] = 0
      rgb_img[i][j][1] = 0
      rgb_img[i][j][2] = 0
 
cv2.imwrite('output.jpg', rgb_img)

最后保存得到的圖為：

淺談pytorch中torch.max和F.softmax函數的維度解釋

以上這篇淺談pytorch中torch.max和F.softmax函數的維度解釋就是小編分享給大家的全部內容了，希望能給大家一個參考，也希望大家多多支持服務器之家。

原文鏈接：https://blog.csdn.net/Jasminexjf/article/details/90402990

延伸 · 閱讀

精彩推薦
主站蜘蛛池模板： 国产69精品久久99不卡免费版 | 久久av喷吹av高潮av懂色 | 91丨九色丨国产在线观看 | av91肉丝一区二区电影 | 精品国产一区在线 | 国产一区二区亚洲 | av播放在线| 激情小说激情电影 | 性欧美视频在线观看 | 久久精品视频首页 | 农村少妇吞精夜夜爽视频 | 久在线播放 | 黄污在线看 | 欧美亚洲一级 | 国产超碰人人爽人人做人人爱 | 极品美女一级毛片 | 成人黄色网战 | 久久精品com| 亚洲免费网站 | 欧美精品久久久久久久久久 | 国产亚洲精品久久午夜玫瑰园 | 视频在线亚洲 | 羞羞视频2023 | 日韩一级片一区二区三区 | 欧美福利视频一区二区三区 | 久久精品4 | 中国女人内谢69xxxx天美 | www日韩在线 | 精品一区二区三区免费毛片爱 | 亚洲天堂午夜 | 亚洲网在线 | 亚洲综合视频网 | 午夜精品福利影院 | a一级黄色大片 | 午夜视频福利 | 亚州欧美视频 | 精品国产96亚洲一区二区三区 | 国产午夜亚洲精品理论片大丰影院 | 日本欧美一区 | 久久久久久久久久性 | 爽妇网国产精品 |