-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathloss.py
133 lines (100 loc) · 4.5 KB
/
loss.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
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
import torch
import torch.nn as nn
import torch.nn.functional as F
from math import exp
import pdb
def normailize(in_):
in_ = F.relu(in_)
max_ = in_.max(3)[0].max(2)[0].unsqueeze(2).unsqueeze(3).expand_as(in_)
min_ = in_.min(3)[0].min(2)[0].unsqueeze(2).unsqueeze(3).expand_as(in_)
in_ = in_ - min_
return in_.div(max_-min_+1e-16)
def NSS(density_pred, gt):
density_pred = normailize(density_pred)
mean, std = torch.std_mean(density_pred, dim=(2, 3), keepdim=True)
density_pred = (density_pred - mean) / (std + 1e-18)
score = density_pred*gt
return score.mean(dim=(2, 3)).sum()
def SSIM(density_pred, density):
density_pred = normailize(density_pred)
density = normailize(density)
_, _, h, w = density.shape
x, sigma_x = torch.var_mean(density_pred, dim=(2, 3), keepdim=True)
y, sigma_y = torch.var_mean(density, dim=(2, 3), keepdim=True)
sigma_xy = ((density_pred-x)*(density-y)).sum(dim=(2, 3), keepdim=True) / (h*w-1)
alpha = 4 * x * y * sigma_xy
beta = (x*x + y*y)*(sigma_x + sigma_y)
score = alpha / (beta + 1e-16)
return score.sum()
def gaussian(window_size, sigma):
gauss = torch.Tensor([exp(-(x - window_size//2)**2/float(2*sigma**2)) for x in range(window_size)])
gauss = gauss/gauss.sum()
return gauss
def create_window(window_size, channel):
_1D_window = gaussian(window_size, 1.5).unsqueeze(1)
_2D_window = _1D_window.mm(_1D_window.t()).float().unsqueeze(0).unsqueeze(0)
window = _2D_window.expand(channel, 1, window_size, window_size).contiguous()
return window
def _ssim(img1, img2, window, window_size, channel):
mu1 = F.conv2d(img1, window, padding=window_size//2, groups=channel)
mu2 = F.conv2d(img2, window, padding=window_size//2, groups=channel)
mu1_sq = mu1.pow(2)
mu2_sq = mu2.pow(2)
mu1_mu2 = mu1*mu2
sigma1_sq = F.conv2d(img1*img1, window, padding=window_size//2, groups=channel) - mu1_sq
sigma2_sq = F.conv2d(img2*img2, window, padding=window_size//2, groups=channel) - mu2_sq
sigma12 = F.conv2d(img1*img2, window, padding=window_size//2, groups=channel) - mu1_mu2
C1 = 0.01**2
C2 = 0.03**2
ssim_map = ((2*mu1_mu2 + C1)*(2*sigma12 + C2))/((mu1_sq + mu2_sq + C1)*(sigma1_sq + sigma2_sq + C2))
return ssim_map.mean(dim=(2, 3)).sum()
# class SSIM(torch.nn.Module):
# def __init__(self, window_size=11, size_average=True):
# super(SSIM, self).__init__()
# self.window_size = window_size
# self.size_average = size_average
# self.channel = 1
# self.window = create_window(window_size, self.channel)
# def forward(self, img1, img2):
# (_, channel, _, _) = img1.size()
# if channel == self.channel and self.window.data.type() == img1.data.type():
# window = self.window
# else:
# window = create_window(self.window_size, channel)
# if img1.is_cuda:
# window = window.cuda(img1.get_device())
# window = window.type_as(img1)
# self.window = window
# self.channel = channel
# return _ssim(img1, img2, window, self.window_size, channel)
def ssim(img1, img2, window_size=15):
(_, channel, _, _) = img1.size()
window = create_window(window_size, channel)
if img1.is_cuda:
window = window.cuda(img1.get_device())
window = window.type_as(img1)
return _ssim(img1, img2, window, window_size, channel)
class PL(nn.Module):
# pyramid loss
def __init__(self):
super(PL, self).__init__()
self.pool1 = nn.AvgPool2d(kernel_size=3)
self.pool2 = nn.AvgPool2d(kernel_size=5)
self.pool3 = nn.AvgPool2d(kernel_size=7)
def forward(self, pred, label):
assert pred.size() == label.size()
n, t, c, h, w = pred.shape
pred = pred.reshape((-1, c, h, w))
label = label.reshape((-1, c, h, w))
N = n*t
loss0 = torch.sum(torch.abs(pred - label)) / N
pred1 = self.pool1(pred)*9
label1 = self.pool1(label)*9
loss1 = torch.sum(torch.abs(pred1 - label1)) / N
pred2 = self.pool2(pred)*25
label2 = self.pool2(label)*25
loss2 = torch.sum(torch.abs(pred2 - label2)) / N
pred3 = self.pool3(pred)*49
label3 = self.pool3(label)*49
loss3 = torch.sum(torch.abs(pred3 - label3)) / N
return (loss0 + loss1 + loss2 + loss3)/4