Tensor转为numpy
np.array(Tensor)
numpy转换为Tensor
torch.Tensor(numpy.darray)
PIL.Image.Image转换成numpy
np.array(PIL.Image.Image)
numpy 转换成PIL.Image.Image
Image.fromarray(numpy.ndarray)
首先需要保证numpy.ndarray 转换成np.uint8型
numpy.astype(np.uint8),像素值[0,255]。
同时灰度图像保证numpy.shape为(H,W),不能出现channels
这里需要np.squeeze()。彩色图象保证numpy.shape为(H,W,3)
之后Image.fromarray(numpy.ndarray)
PIL.Image.Image转换成Tensor
torchvision.transfrom
1
2
3
4
5
|
img = Image. open ( '00381fa010_940422.tif' ).convert( 'L' )
import torchvision.transforms as transforms trans = transforms.Compose([transforms.ToTensor()])
a = trans(img)
|
Tensor转化成PIL.Image.Image
先转换成numpy,再转换成PIL.Image.Image
灰度图像
1
2
3
4
5
6
7
8
9
10
11
12
13
|
img = Image. open ( '00381fa010_940422.tif' ).convert( 'L' )
import torchvision.transforms as transforms
trans = transforms.Compose([transforms.ToTensor()])
a = trans(img)
b = np.array(a) #b.shape (1,64,64)
maxi = b. max ()
b = b * 255. / maxi
b = b.transpose( 1 , 2 , 0 ).astype(np.uint8)
b = np.squeeze(b,axis = 2 )
xx = Image.fromarray(b)
xx
|
彩色图象
1
2
3
4
5
6
7
8
9
10
|
img2 = Image. open ( '00381fa010_940422.tif' ).convert( 'RGB' )
import torchvision.transforms as transforms
trans = transforms.Compose([transforms.ToTensor()])
a = trans(img2)
a = np.array(a)
maxi = a. max ()
a = a / maxi * 255
a = a.transpose( 1 , 2 , 0 ).astype(np.uint8)
b = Image.fromarray(a)
b
|
python-opencv
1
2
3
4
5
6
|
import cv2
a = cv2.imread( '00381fa010_940422.tif' ) #a.shape (64,64,3)
cv2.imwrite( 'asd.jpg' ,a)
Image.fromarray(a)
b = cv2.imread( '00381fa010_940422.tif' , 0 ) #b.shape (64,64)
Image.fromarray(b)
|
cv2.imread()返回numpy.darray, 读取灰度图像之后shape为(64,64),RGB图像的shape为(64,64,3),可直接用Image.fromarray()转换成Image。
cv写图像时,灰度图像shape可以为(H,W)或(H,W,1)。彩色图像(H,W,3)
要从numpy.ndarray得到PIL.Image.Image,灰度图的shape必须为(H,W),彩色为(H,W,3)
对于Variable类型不能直接转换成numpy.ndarray,需要用.data转换
np.array(a.data)
以上这篇python、PyTorch图像读取与numpy转换实例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。
原文链接:https://blog.csdn.net/yskyskyer123/article/details/80707038