/
evaluation.py
101 lines (82 loc) · 3.24 KB
/
evaluation.py
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
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
#import lpips
class PSNR(object):
def __init__(self, des="Peak Signal to Noise Ratio"):
self.des = des
def __repr__(self):
return "PSNR"
def __call__(self, y_pred, y_true, dim=1, threshold=None):
"""
args:
y_true : 4-d ndarray in [batch_size, channels, img_rows, img_cols]
y_pred : 4-d ndarray in [batch_size, channels, img_rows, img_cols]
threshold : [0.0, 1.0]
return PSNR, larger the better
"""
if threshold:
y_pred = _binarize(y_pred, threshold)
mse = torch.mean((y_pred - y_true) ** 2)
return 10 * torch.log10(1 / mse)
class SSIM(object):
'''
modified from https://github.com/jorge-pessoa/pytorch-msssim
'''
def __init__(self, des="structural similarity index"):
self.des = des
def __repr__(self):
return "SSIM"
def gaussian(self, w_size, sigma):
gauss = torch.Tensor([math.exp(-(x - w_size//2)**2/float(2*sigma**2)) for x in range(w_size)])
return gauss/gauss.sum()
def create_window(self, w_size, channel=1):
_1D_window = self.gaussian(w_size, 1.5).unsqueeze(1)
_2D_window = _1D_window.mm(_1D_window.t()).float().unsqueeze(0).unsqueeze(0)
window = _2D_window.expand(channel, 1, w_size, w_size).contiguous()
return window
def __call__(self, y_pred, y_true, w_size=11, size_average=True, full=False):
"""
args:
y_true : 4-d ndarray in [batch_size, channels, img_rows, img_cols]
y_pred : 4-d ndarray in [batch_size, channels, img_rows, img_cols]
w_size : int, default 11
size_average : boolean, default True
full : boolean, default False
return ssim, larger the better
"""
# Value range can be different from 255. Other common ranges are 1 (sigmoid) and 2 (tanh).
if torch.max(y_pred) > 128:
max_val = 255
else:
max_val = 1
if torch.min(y_pred) < -0.5:
min_val = -1
else:
min_val = 0
L = max_val - min_val
padd = 0
(_, channel, height, width) = y_pred.size()
window = self.create_window(w_size, channel=channel).to(y_pred.device)
mu1 = F.conv2d(y_pred, window, padding=padd, groups=channel)
mu2 = F.conv2d(y_true, window, padding=padd, groups=channel)
mu1_sq = mu1.pow(2)
mu2_sq = mu2.pow(2)
mu1_mu2 = mu1 * mu2
sigma1_sq = F.conv2d(y_pred * y_pred, window, padding=padd, groups=channel) - mu1_sq
sigma2_sq = F.conv2d(y_true * y_true, window, padding=padd, groups=channel) - mu2_sq
sigma12 = F.conv2d(y_pred * y_true, window, padding=padd, groups=channel) - mu1_mu2
C1 = (0.01 * L) ** 2
C2 = (0.03 * L) ** 2
v1 = 2.0 * sigma12 + C2
v2 = sigma1_sq + sigma2_sq + C2
cs = torch.mean(v1 / v2) # contrast sensitivity
ssim_map = ((2 * mu1_mu2 + C1) * v1) / ((mu1_sq + mu2_sq + C1) * v2)
if size_average:
ret = ssim_map.mean()
else:
ret = ssim_map.mean(1).mean(1).mean(1)
if full:
return ret, cs
return ret