二维深度图像矩阵转回图片

Posted 2020-10-15 韬小虾

labels.npy文件查看

1 import numpy as np
2 matrix = np.load(\'./V1/labels.npy\')
3 
4 print(matrix)
5 print(matrix.shape)
6 print(type(matrix))

1、print(matrix)

2、print(matrix.shape)

(480, 640, 2284)

3、print(type(matrix))

<class \'numpy.ndarray\'>

转换、压缩、裁剪图片

 1 import numpy as np
 2 import  torchvision.transforms as transforms
 3 from PIL import Image
 4 
 5 depth_dataset = np.load(\'./V1/labels.npy\')
 6 # print(dataset.shape)
 7 
 8 def __scale_width(img, target_high):
 9     ow, oh = img.size
10     if (ow == target_high):
11         return img
12     h = target_high
13     w = int(target_high * ow / oh)
14     return img.resize((w, h), Image.BICUBIC)
15 
16 for i in range(10):  #dataset.shape[2]
17     depth = depth_dataset[:, :, i]#2284
18 
19     #归一化
20     depth_min = depth.min()
21     depth_max = depth.max()
22     depth = ((depth-depth_min)/(depth_max-depth_min))*255
23 
24     #还原图片
25     images = Image.fromarray(depth)
26 
27     if images.mode != \'RGB\':
28         images = images.convert(\'RGB\')
29 
30     # images.show()
31     #压缩图片
32     scale = __scale_width(images, 256)
33     # print(scale.size)
34     # scale.show()
35 
36     #裁剪图片
37     crop = transforms.RandomCrop((256, 256))
38     crop_img = crop(scale)
39 
40     # crop_img.show()
41     #保存图片
42     crop_img.save(\'./V1Labels/%d.jpg\' %i)