├── README.md
└── utils.py


/README.md:
--------------------------------------------------------------------------------
1 | # SSGN
2 | 
3 | Q.Zhang, Q. Yuan, J. li, X. Liu, H. Shen, and L. Zhang, "Hybrid Noise Removal in Hyperspectral Imagery With a Spatial-Spectral Gradient Network," Accepted by IEEE TGRS, in press, 2019. https://ieeexplore.ieee.org/document/8734833 
4 | 


--------------------------------------------------------------------------------
/utils.py:
--------------------------------------------------------------------------------
 1 | import math
 2 | import torch
 3 | import torch.nn as nn
 4 | import numpy as np
 5 | from skimage.measure.simple_metrics import compare_psnr
 6 | 
 7 | 
 8 | def weights_init_kaiming(m):
 9 |     classname = m.__class__.__name__
10 |     if classname.find('Conv') != -1:
11 |         nn.init.kaiming_normal(m.weight.data, a=0, mode='fan_in')
12 |     elif classname.find('Linear') != -1:
13 |         nn.init.kaiming_normal(m.weight.data, a=0, mode='fan_in')
14 |     elif classname.find('BatchNorm') != -1:
15 |         # nn.init.uniform(m.weight.data, 1.0, 0.02)
16 |         m.weight.data.normal_(mean=0, std=math.sqrt(2./9./64.)).clamp_(-0.025,0.025)
17 |         nn.init.constant(m.bias.data, 0.0)
18 | 
19 | def batch_PSNR(img, imclean, data_range):
20 |     Img = img.data.cpu().numpy().astype(np.float32)
21 |     Iclean = imclean.data.cpu().numpy().astype(np.float32)
22 |     PSNR = 0
23 |     for i in range(Img.shape[0]):
24 |         PSNR += compare_psnr(Iclean[i,:,:,:], Img[i,:,:,:], data_range=data_range)
25 |     return (PSNR/Img.shape[0])
26 | 
27 | def data_augmentation(image, mode):
28 |     out = np.transpose(image, (1,2,0))
29 |     if mode == 0:
30 |         # original
31 |         out = out
32 |     elif mode == 1:
33 |         # flip up and down
34 |         out = np.flipud(out)
35 |     elif mode == 2:
36 |         # rotate counterwise 90 degree
37 |         out = np.rot90(out)
38 |     elif mode == 3:
39 |         # rotate 90 degree and flip up and down
40 |         out = np.rot90(out)
41 |         out = np.flipud(out)
42 |     elif mode == 4:
43 |         # rotate 180 degree
44 |         out = np.rot90(out, k=2)
45 |     elif mode == 5:
46 |         # rotate 180 degree and flip
47 |         out = np.rot90(out, k=2)
48 |         out = np.flipud(out)
49 |     elif mode == 6:
50 |         # rotate 270 degree
51 |         out = np.rot90(out, k=3)
52 |     elif mode == 7:
53 |         # rotate 270 degree and flip
54 |         out = np.rot90(out, k=3)
55 |         out = np.flipud(out)
56 |     return np.transpose(out, (2,0,1))
57 | 


--------------------------------------------------------------------------------