diffusion/metrics.py at main · dptmf35/diffusion · GitHub

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
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
import numpy as np
from skimage.metrics import structural_similarity as SSIM

def tensor_back_to_unMinMax(input_image, min, max):
  image = input_image * (max - min) + min
  return image

def Peak_Signal_to_Noise_Rate_3D(arr1, arr2, size_average=True, PIXEL_MAX=1.0):
  '''
  :param arr1:
    Format-[NDHW], OriImage [0,1]
  :param arr2:
    Format-[NDHW], ComparedImage [0,1]
  :return:
    Format-None if size_average else [N]
  '''
  print(arr1.ndim, arr2.ndim)
  assert (isinstance(arr1, np.ndarray)) and (isinstance(arr2, np.ndarray))
  assert (arr1.ndim == 4) and (arr2.ndim == 4)
  arr1 = arr1.astype(np.float64)
  arr2 = arr2.astype(np.float64)
  eps = 1e-10
  se = np.power(arr1 - arr2, 2)
  mse = se.mean(axis=1).mean(axis=1).mean(axis=1)
  zero_mse = np.where(mse == 0)
  mse[zero_mse] = eps
  psnr = 20 * np.log10(PIXEL_MAX / np.sqrt(mse))
  # #zero mse, return 100
  psnr[zero_mse] = 100

  if size_average:
    return psnr.mean()
  else:
    return psnr

def MAE(arr1, arr2, size_average=True):
  '''
  :param arr1:
    Format-[NDHW], OriImage
  :param arr2:
    Format-[NDHW], ComparedImage
  :return:
    Format-None if size_average else [N]
  '''
  assert (isinstance(arr1, np.ndarray)) and (isinstance(arr2, np.ndarray))
  assert (arr1.ndim == 4) and (arr2.ndim == 4)
  arr1 = arr1.astype(np.float64)
  arr2 = arr2.astype(np.float64)
  if size_average:
    return np.abs(arr1 - arr2).mean()
  else:
    return np.abs(arr1 - arr2).mean(1).mean(1).mean(1)

def MSE(arr1, arr2, size_average=True):
  '''
  :param arr1:
    Format-[NDHW], OriImage
  :param arr2:
    Format-[NDHW], ComparedImage
  :return:
    Format-None if size_average else [N]
  '''
  assert (isinstance(arr1, np.ndarray)) and (isinstance(arr2, np.ndarray))
  assert (arr1.ndim == 4) and (arr2.ndim == 4)
  arr1 = arr1.astype(np.float64)
  arr2 = arr2.astype(np.float64)
  if size_average:
    return np.power(arr1 - arr2, 2).mean()
  else:
    return np.power(arr1 - arr2, 2).mean(1).mean(1).mean(1)

def Peak_Signal_to_Noise_Rate(arr1, arr2, size_average=True, PIXEL_MAX=1.0):
  '''
  :param arr1:
    Format-[NDHW], OriImage [0,1]
  :param arr2:
    Format-[NDHW], ComparedImage [0,1]
  :return:
    Format-None if size_average else [N]
  '''
  assert (isinstance(arr1, np.ndarray)) and (isinstance(arr2, np.ndarray))
  assert (arr1.ndim == 4) and (arr2.ndim == 4)
  arr1 = arr1.astype(np.float64)
  arr2 = arr2.astype(np.float64)
  eps = 1e-10
  se = np.power(arr1 - arr2, 2)
  # Depth
  mse_d = se.mean(axis=2, keepdims=True).mean(axis=3, keepdims=True).squeeze(3).squeeze(2)
  zero_mse = np.where(mse_d==0)
  mse_d[zero_mse] = eps
  psnr_d = 20 * np.log10(PIXEL_MAX / np.sqrt(mse_d))
  # #zero mse, return 100
  psnr_d[zero_mse] = 100
  psnr_d = psnr_d.mean(1)

  # Height
  mse_h = se.mean(axis=1, keepdims=True).mean(axis=3, keepdims=True).squeeze(3).squeeze(1)
  zero_mse = np.where(mse_h == 0)
  mse_h[zero_mse] = eps
  psnr_h = 20 * np.log10(PIXEL_MAX / np.sqrt(mse_h))
  # #zero mse, return 100
  psnr_h[zero_mse] = 100
  psnr_h = psnr_h.mean(1)

  # Width
  mse_w = se.mean(axis=1, keepdims=True).mean(axis=2, keepdims=True).squeeze(2).squeeze(1)
  zero_mse = np.where(mse_w == 0)
  mse_w[zero_mse] = eps
  psnr_w = 20 * np.log10(PIXEL_MAX / np.sqrt(mse_w))
  # #zero mse, return 100
  psnr_w[zero_mse] = 100
  psnr_w = psnr_w.mean(1)

  psnr_avg = (psnr_h + psnr_d + psnr_w) / 3
  if size_average:
    return [psnr_d.mean(), psnr_h.mean(), psnr_w.mean(), psnr_avg.mean()]
  else:
    return [psnr_d, psnr_h, psnr_w, psnr_avg]

def Structural_Similarity(arr1, arr2, size_average=True, PIXEL_MAX=1.0):
  '''
  :param arr1:
    Format-[NDHW], OriImage [0,1]
  :param arr2:
    Format-[NDHW], ComparedImage [0,1]
  :return:
    Format-None if size_average else [N]
  '''
  assert (isinstance(arr1, np.ndarray)) and (isinstance(arr2, np.ndarray))
  assert (arr1.ndim == 4) and (arr2.ndim == 4)
  arr1 = arr1.astype(np.float64)
  arr2 = arr2.astype(np.float64)

  N = arr1.shape[0]
  # Depth
  arr1_d = np.transpose(arr1, (0, 2, 3, 1))
  arr2_d = np.transpose(arr2, (0, 2, 3, 1))
  ssim_d = []
  for i in range(N):
    ssim = SSIM(arr1_d[i], arr2_d[i], data_range=PIXEL_MAX, multichannel=True)
    ssim_d.append(ssim)
  ssim_d = np.asarray(ssim_d, dtype=np.float64)

  # Height
  arr1_h = np.transpose(arr1, (0, 1, 3, 2))
  arr2_h = np.transpose(arr2, (0, 1, 3, 2))
  ssim_h = []
  for i in range(N):
    ssim = SSIM(arr1_h[i], arr2_h[i], data_range=PIXEL_MAX, multichannel=True)
    ssim_h.append(ssim)
  ssim_h = np.asarray(ssim_h, dtype=np.float64)

  # Width
  # arr1_w = np.transpose(arr1, (0, 1, 2, 3))
  # arr2_w = np.transpose(arr2, (0, 1, 2, 3))
  ssim_w = []
  for i in range(N):
    ssim = SSIM(arr1[i], arr2[i], data_range=PIXEL_MAX, multichannel=True)
    ssim_w.append(ssim)
  ssim_w = np.asarray(ssim_w, dtype=np.float64)

  ssim_avg = (ssim_d + ssim_h + ssim_w) / 3

  if size_average:
    return [ssim_d.mean(), ssim_h.mean(), ssim_w.mean(), ssim_avg.mean()]
  else:
    return [ssim_d, ssim_h, ssim_w, ssim_avg]