-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmetrics.py
155 lines (128 loc) · 6.27 KB
/
metrics.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
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
import numpy as np
import os
from os.path import join
import cv2
import torch
import math
from tqdm import tqdm
import lpips
import glob
from skimage.metrics import structural_similarity as ssim
import argparse
from options.base_options import str2bool
def calc_psnr_np(sr, hr, range=255.):
diff = (sr.astype(np.float32) - hr.astype(np.float32)) / range
total_mse = np.power(diff, 2)
total_psnr = -10 * math.log10(total_mse.mean())
return total_psnr
def calc_psnr_corner(sr, hr, range=255.):
diff = (sr.astype(np.float32) - hr.astype(np.float32)) / range
total_mse = np.power(diff, 2).mean()
return total_mse
def lpips_norm(img):
img = img[:, :, :, np.newaxis].transpose((3, 2, 0, 1))
img = img / (255. / 2.) - 1
return torch.Tensor(img).to(device)
def calc_lpips(x_mask_out, x_canon, loss_fn_alex_1):
lpips_mask_out = lpips_norm(x_mask_out)
lpips_canon = lpips_norm(x_canon)
LPIPS_1 = loss_fn_alex_1(lpips_mask_out, lpips_canon)
return LPIPS_1.detach().cpu() #, LPIPS_1.detach().cpu()
def crop_part(out, ref, scale):
s = scale
H, W, _ = out.shape
c_H, c_W = (H - H//s)//2, (W - W//s)//2
top = [out[:, 0:c_W,...], ref[:, 0:c_W,...]]
left = [out[:c_H, c_W:c_W+W//s,...], ref[:c_H, c_W:c_W+W//s,...]]
right = [out[c_H+H//s:, c_W:c_W+W//s, ...], ref[c_H+H//s:, c_W:c_W+W//s, ...]]
bottom = [out[:, c_W+W//s:,...], ref[:, c_W+W//s:,...]]
center = [out[c_H:c_H+H//s, c_W:c_W+W//s,...], ref[c_H:c_H+H//s, c_W:c_W+W//s,...]]
return top, left, right, bottom, center
def calc_metrics(out, ref, scale, loss_fn_alex_1):
parts = crop_part(out, ref, scale)
psnr_corner = 0.0
ssim_corner = 0.0
lpips_corner = 0.0
areas = 0.0
for part in parts[:-1]:
total_psnr = calc_psnr_corner(part[0], part[1])
SSIM = ssim(part[0], part[1], win_size=11, data_range=255, multichannel=True, gaussian_weights=True)
LPIPS_1 = calc_lpips(part[0], part[1], loss_fn_alex_1)
area = part[0].shape[0] * part[0].shape[1]
psnr_corner += total_psnr * area
ssim_corner += SSIM
lpips_corner += LPIPS_1
areas += area
psnr_corner, ssim_corner, lpips_corner = - 10 * math.log10(psnr_corner / areas), ssim_corner / 4, lpips_corner / 4
center_psnr = calc_psnr_np(parts[-1][0], parts[-1][1])
center_SSIM = ssim(parts[-1][0], parts[-1][1], win_size=11, data_range=255, multichannel=True, gaussian_weights=True)
center_lpips = calc_lpips(parts[-1][0], parts[-1][1], loss_fn_alex_1)
total_psnr = calc_psnr_np(out, ref)
total_ssim = ssim(out, ref, win_size=11, data_range=255, multichannel=True, gaussian_weights=True)
total_lpips = calc_lpips(out, ref, loss_fn_alex_1)
return [psnr_corner, center_psnr, total_psnr, ssim_corner, center_SSIM, total_ssim, lpips_corner, center_lpips, total_lpips]
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Test for argparse')
parser.add_argument('--name', '-n', help='name 属性，必要参数')
parser.add_argument('--device', default="0")
parser.add_argument('--load_iter', default="401")
parser.add_argument('--dataroot', type=str, default='')
parser.add_argument('--camera', type=str, default='')
args = parser.parse_args()
print(args)
args.device = "cuda:" + args.device
device = torch.device(args.device if torch.cuda.is_available() else "cpu")
loss_fn_alex_1 = lpips.LPIPS(net='alex', version='0.1').to(device)
files = [
'./ckpt/' + args.name + '/'
]
camera = args.camera
scale = 4
if camera == 'nikon':
ori_target = args.dataroot + '/test_HR/'
elif camera == 'iphone':
ori_target = args.dataroot + '/test/HR/'
for file in files:
log_dir = '%s/log_metrics_%s.txt' % (file, args.load_iter)
f = open(log_dir, 'a')
f.write('\n=============%s=============\n' % (camera))
print('\n=============%s=============\n' % (camera))
names = []
if camera == 'nikon':
for file_name in os.listdir(ori_target):
if file_name.startswith('DSC_'):
names.append(file_name)
elif camera == 'iphone':
for video in os.listdir(ori_target):
for file_name in os.listdir(ori_target + video):
names.append(video + '/' + file_name)
names = sorted(names)
ori_metrics = np.zeros([len(names), 9])
i = 0
for name in tqdm(names):
out = cv2.imread(file + 'sr_output_' + args.load_iter + '/' + name)[..., ::-1]
hr = cv2.imread(ori_target + name)[..., ::-1]
ori_metrics[i] = calc_metrics(out, hr, scale, loss_fn_alex_1)
f.write('name: %s, \n corner_psnr: %.2f, \t center_psnr: %.2f, \t total_psnr: %.2f, \
\n corner_SSIM: %.4f, \t center_SSIM: %.4f, \t total_SSIM: %.4f, \
\n corner_LPIPS: %.3f, \t center_LPIPS: %.3f, \t total_LPIPS: %.3f \t \n' \
% (name, ori_metrics[i][0], ori_metrics[i][1], ori_metrics[i][2], ori_metrics[i][3], ori_metrics[i][4],
ori_metrics[i][5], ori_metrics[i][6], ori_metrics[i][7], ori_metrics[i][8]))
i = i + 1
metrics_mean = np.mean(ori_metrics, axis=0)
f.write('\n camera: %s ====== \
\n corner_psnr: %.2f, \t center_psnr: %.2f, \t total_psnr: %.2f, \
\n corner_SSIM: %.4f, \t center_SSIM: %.4f, \t total_SSIM: %.4f, \
\n corner_LPIPS: %.3f, \t center_LPIPS: %.3f, \t total_LPIPS: %.3f \t \n' \
% (camera, metrics_mean[0], metrics_mean[1], metrics_mean[2],
metrics_mean[3], metrics_mean[4], metrics_mean[5],
metrics_mean[6], metrics_mean[7], metrics_mean[8]))
print('\n camera: %s ====== \
\n corner_psnr: %.2f, \t center_psnr: %.2f, \t total_psnr: %.2f, \
\n corner_SSIM: %.4f, \t center_SSIM: %.4f, \t total_SSIM: %.4f, \
\n corner_LPIPS: %.3f, \t center_LPIPS: %.3f, \t total_LPIPS: %.3f \t \n' \
% (camera, metrics_mean[0], metrics_mean[1], metrics_mean[2],
metrics_mean[3], metrics_mean[4], metrics_mean[5],
metrics_mean[6], metrics_mean[7], metrics_mean[8]))
f.flush()
f.close()