1、 将一维变为三维张量
import torch
a = torch.Tensor([0,1,0,1,0,1,0,0,1,0,1,1,0,0,1,0,1,1])
res1 = a.view((3,3,2))
print("res1:",res1)
res1 = res1[:,:,1]
print("res1_",res1)
输出:
res1: tensor([[[0., 1.],
[0., 1.],
[0., 1.]],
[[0., 0.],
[1., 0.],
[1., 1.]],
[[0., 0.],
[1., 0.],
[1., 1.]]])
res1_: tensor([[1., 1., 1.],
[0., 0., 1.],
[0., 0., 1.]])
Process finished with exit code 0
从上面可以看到一维和变形后对应的关系,后面就是根据这个来区分通道的。
本来一维数据的[0,2,4,6,8..] 对应于第一个通道,而[1,3,5,7,9,...]对应了第二个通道
2、将数据压缩成一维
#把数据压成一维
b = torch.Tensor([[[1,2,3],[4,5,6],[7,8,9]],[[1,2,3],[4,5,6],[7,8,9]]])
res2 = b.ravel()
print("res2:",res2)
输出:
res2: tensor([1., 2., 3., 4., 5., 6., 7., 8., 9., 1., 2., 3., 4., 5., 6., 7., 8., 9.])
3、 增加数据维度
c = np.zeros((3,3) + (2, ))
print("c_shape",c.shape)
print("c",c)
输出:
c_shape (3, 3, 2)
c [[[0. 0.]
[0. 0.]
[0. 0.]]
[[0. 0.]
[0. 0.]
[0. 0.]]
[[0. 0.]
[0. 0.]
[0. 0.]]]
Process finished with exit code 0
4、arange
#一维数组,返回给定间隔内的等间距值。
d = np.arange(10)
print("d:",d)
输出:
d: [0 1 2 3 4 5 6 7 8 9]
5、onehot
import numpy as np
import cv2
# 将标记图(每个像素值代该位置像素点的类别)转换为onehot编码
def onehot(data, n):
print("data_shape:",data.shape)
print("data_shape:",data.shape + (n, ))
buf = np.zeros(data.shape + (n, )) #增加数据维度 例如(9,9) -> (9,9,2)
print("arange:",np.arange(data.size)*n)
print("data_ravel:",data.ravel())
nmsk = np.arange(data.size)*n + data.ravel() #返回等间距[0,2,4,6,8,...] + [0,0,0,0,1,1,1]
print("nmsk:",nmsk)
print("nmsk_shape:",nmsk.shape)
print("buf_ravel_shape:",buf.ravel().shape)
print("nmsk-1:",nmsk-1)
buf.ravel()[nmsk-1] = 1
print("buf_ravel:",buf.ravel())
print("buf:",buf)
return buf
if __name__ == '__main__':
imgB = cv2.imread('./00000.png' , 0)
imgB = cv2.resize(imgB, (9, 9))
imgB = imgB / 255
imgB = imgB.astype('uint8')
imgB = onehot(imgB, 2)
cv2.imshow('test',imgB[:,:,0])
cv2.waitKey(0)
cv2.destroyWindow('test')
输出:
data_shape: (9, 9)
data_shape: (9, 9, 2)
arange: [ 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34
36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70
72 74 76 78 80 82 84 86 88 90 92 94 96 98 100 102 104 106
108 110 112 114 116 118 120 122 124 126 128 130 132 134 136 138 140 142
144 146 148 150 152 154 156 158 160]
data_ravel: [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 0 0 0 1 1 1 1 1 0
0 0 0 1 1 1 1 1 0 0 0 1 1 1 1 1 0 0 0 0 1 1 1 1 1 0 0 0 0 1 1 1 1 1 1 0 0
1 1 1 1 1 1 1]
nmsk: [ 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34
36 38 40 42 44 47 49 51 52 54 56 58 60 63 65 67 69 71
72 74 76 78 81 83 85 87 89 90 92 94 97 99 101 103 105 106
108 110 112 115 117 119 121 123 124 126 128 130 133 135 137 139 141 143
144 146 149 151 153 155 157 159 161]
nmsk_shape: (81,)
buf_ravel_shape: (162,)
nmsk-1: [ -1 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33
35 37 39 41 43 46 48 50 51 53 55 57 59 62 64 66 68 70
71 73 75 77 80 82 84 86 88 89 91 93 96 98 100 102 104 105
107 109 111 114 116 118 120 122 123 125 127 129 132 134 136 138 140 142
143 145 148 150 152 154 156 158 160]
buf_ravel: [0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1.
0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 0. 1. 0.
1. 0. 1. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 1.
0. 1. 0. 1. 0. 1. 0. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 1. 0. 1. 0. 1. 0. 0.
1. 0. 1. 0. 1. 0. 1. 0. 1. 1. 0. 1. 0. 1. 0. 1. 0. 0. 1. 0. 1. 0. 1. 0.
1. 0. 1. 1. 0. 1. 0. 1. 0. 1. 0. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 1.
0. 1. 0. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 0. 1. 1.]
buf: [[[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]]
[[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]]
[[0. 1.]
[0. 1.]
[0. 1.]
[0. 1.]
[0. 0.]
[1. 0.]
[1. 0.]
[1. 1.]
[0. 1.]]
[[0. 1.]
[0. 1.]
[0. 1.]
[0. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 1.]]
[[0. 1.]
[0. 1.]
[0. 1.]
[0. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 1.]]
[[0. 1.]
[0. 1.]
[0. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 1.]
[0. 1.]]
[[0. 1.]
[0. 1.]
[0. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 1.]
[0. 1.]]
[[0. 1.]
[0. 1.]
[0. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 1.]]
[[0. 1.]
[0. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 0.]
[1. 1.]]]
nmsk的作用在于:标记需要变为1的地方,因为是两个通道,在第一个通道为1的,在第二个通道为0,反之亦然。nmsk = np.arange(data.size)*n + data.ravel()中,[0,2,4,6,8,...] + [0,0,0,1,1] = [0,2,4,7,9],我们知道[ 0,2,4,6,8]指的是第一个通道的数据,[,1,3,5,7,9]指的是第二个通道的数据。
nmsk-1 = [-1,1,3,6,8] ,分为两部分 第一部分 、原来通过+1的部分[6,8] -> [6,8],然后-1操作后,第一个通道中原来是1的地方被保留了,因为数据默认为0,所以只要保留1的地方就行。 第二部分,原来+0的部分[0,2,4] -> [-1,1,3],通过-1操作后,原来第一个通道为0的部分,变到第二个通道为1.
我们开始用0填充的数据,只需要知道哪里是1就行了,剩下的[0,2,4,5,7]默认都为0。