-
Notifications
You must be signed in to change notification settings - Fork 1
/
evaluation_metrics.py
205 lines (188 loc) · 9.74 KB
/
evaluation_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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
import lpips, os, cv2, math
from skimage.metrics import structural_similarity
import numpy as np
import torch
class Quantitation(object):
def __init__(self):
self.lpips_vgg = lpips.LPIPS(net='vgg', spatial=False).cuda()
self.lpips_alex = lpips.LPIPS(net='alex', spatial=False).cuda()
@staticmethod
def MSE(x, y):
"""
@:keyword: smaller, better
"""
return np.sqrt(np.mean((x.astype(np.float) - y.astype(np.float))**2))
@staticmethod
def PSNR(x, y, factor=255.0):
"""
@:keyword: higher, better
"""
mse = np.sqrt(np.mean((x.astype(np.float) - y.astype(np.float)) ** 2))
return 10 * math.log10(factor ** 2 / mse)
def LPIPS(self, x, y):
"""
@:keyword: smaller, better
"""
x_torch = torch.tensor(x).permute(2, 0, 1).unsqueeze(dim=0).cuda()
y_torch = torch.tensor(y).permute(2, 0, 1).unsqueeze(dim=0).cuda()
with torch.no_grad():
lpips_score_vgg = self.lpips_vgg(x_torch, y_torch).squeeze().item()
lpips_score_alex = self.lpips_alex(x_torch, y_torch).squeeze().item()
return lpips_score_vgg, lpips_score_alex
@staticmethod
def SSIM( x, y):
"""
@:keyword: higher, better
"""
return structural_similarity(x, y, multichannel=True)
def ASR(self):
"""
@:keyword: We consider images as being successfully reconstructed if SSIM ≥ 0.6
@:keyword: smaller, better. percentage
"""
pass
def DLG_GRNN():
evaluator = Quantitation()
method = 'GRNN' # DLG GRNN
network = 'res18' # lenet res20 res18
# datasets = ['mnist', 'cifar10', 'cifar100']
datasets = ['cifar10', 'cifar100', 'imagenet']
for dataset in datasets:
root_paht = f'./{method}/{network}'
mse = []
psnr = []
lpips_vgg = []
lpips_alex = []
ssim = []
mse_kl = []
psnr_kl = []
lpips_vgg_kl = []
lpips_alex_kl = []
ssim_kl = []
for d in os.listdir(root_paht):
if f"-{dataset}-" in d:
if d.endswith('-no_kl'):
print(d)
model_path = f"{root_paht}/{d}/saved_img/"
for idx in range(20):
try:
true_img_name = os.listdir(f"{model_path}/true_data/exp-{str(idx).zfill(6)}/")[0]
fake_img_name = os.listdir(f"{model_path}/fake_data/exp-{str(idx).zfill(6)}/")[0]
true_img = cv2.imread(f"{model_path}/true_data/exp-{str(idx).zfill(6)}/{true_img_name}")
fake_img = cv2.imread(f"{model_path}/fake_data/exp-{str(idx).zfill(6)}/{fake_img_name}")
except FileNotFoundError:
true_img_name = os.listdir(f"{model_path}/true_data/exp{str(idx).zfill(4)}/")[0]
fake_img_name = os.listdir(f"{model_path}/fake_data/exp{str(idx).zfill(4)}/")[0]
true_img = cv2.imread(f"{model_path}/true_data/exp{str(idx).zfill(4)}/{true_img_name}")
fake_img = cv2.imread(f"{model_path}/fake_data/exp{str(idx).zfill(4)}/{fake_img_name}")
mse.append(evaluator.MSE(true_img, fake_img))
psnr.append(evaluator.PSNR(true_img, fake_img))
lpips_score = evaluator.LPIPS(true_img, fake_img)
lpips_vgg.append(lpips_score[0])
lpips_alex.append(lpips_score[1])
ssim.append(evaluator.SSIM(true_img, fake_img))
elif d.endswith('-kl'):
print(d)
model_path = f"{root_paht}/{d}/saved_img/"
for idx in range(20):
try:
true_img_name = os.listdir(f"{model_path}/true_data/exp-{str(idx).zfill(6)}/")[0]
fake_img_name = os.listdir(f"{model_path}/fake_data/exp-{str(idx).zfill(6)}/")[0]
true_img = cv2.imread(f"{model_path}/true_data/exp-{str(idx).zfill(6)}/{true_img_name}")
fake_img = cv2.imread(f"{model_path}/fake_data/exp-{str(idx).zfill(6)}/{fake_img_name}")
except FileNotFoundError:
true_img_name = os.listdir(f"{model_path}/true_data/exp{str(idx).zfill(4)}/")[0]
fake_img_name = os.listdir(f"{model_path}/fake_data/exp{str(idx).zfill(4)}/")[0]
true_img = cv2.imread(f"{model_path}/true_data/exp{str(idx).zfill(4)}/{true_img_name}")
fake_img = cv2.imread(f"{model_path}/fake_data/exp{str(idx).zfill(4)}/{fake_img_name}")
mse_kl.append(evaluator.MSE(true_img, fake_img))
psnr_kl.append(evaluator.PSNR(true_img, fake_img))
lpips_score = evaluator.LPIPS(true_img, fake_img)
lpips_vgg_kl.append(lpips_score[0])
lpips_alex_kl.append(lpips_score[1])
ssim_kl.append(evaluator.SSIM(true_img, fake_img))
print(f"{method} {network} {dataset}")
mse_mean = format(np.round(np.mean(mse), 2), '.2f')
psnr_mean = format(np.round(np.mean(psnr), 2), '.2f')
lpips_vgg_mean = format(np.round(np.mean(lpips_vgg), 2), '.2f')
lpips_alex_mean = format(np.round(np.mean(lpips_alex), 2), '.2f')
ssim_mean = format(np.round(np.mean(ssim), 3), '.3f')
mse_kl_mean = format(np.round(np.mean(mse_kl), 2), '.2f')
psnr_kl_mean = format(np.round(np.mean(psnr_kl), 2), '.2f')
lpips_vgg_kl_mean = format(np.round(np.mean(lpips_vgg_kl), 2), '.2f')
lpips_alex_kl_mean = format(np.round(np.mean(lpips_alex_kl), 2), '.2f')
ssim_kl_mean = format(np.round(np.mean(ssim_kl), 3), '.3f')
print(f"{mse_mean} & {mse_kl_mean} & {psnr_mean} & {psnr_kl_mean} & {lpips_vgg_mean} & {lpips_vgg_kl_mean} & {lpips_alex_mean} & {lpips_alex_kl_mean} & {ssim_mean} & {ssim_kl_mean}")
def GGL():
root_paht = f'./GGL/'
evaluator = Quantitation()
mse = []
psnr = []
lpips_vgg = []
lpips_alex = []
ssim = []
mse_kl = []
psnr_kl = []
lpips_vgg_kl = []
lpips_alex_kl = []
ssim_kl = []
for d in os.listdir(root_paht):
if d.endswith('-no_kl'):
print(d)
model_path = f"{root_paht}/{d}/"
for idx in range(99,34099, 1000):
for i in range(5):
true_img = cv2.imread(f"{model_path}/{idx}_true_{i}.png")
fake_img = cv2.imread(f"{model_path}/{idx}_out_{i}.png")
mse.append(evaluator.MSE(true_img, fake_img))
psnr.append(evaluator.PSNR(true_img, fake_img))
lpips_score = evaluator.LPIPS(true_img, fake_img)
lpips_vgg.append(lpips_score[0])
lpips_alex.append(lpips_score[1])
ssim.append(evaluator.SSIM(true_img, fake_img))
elif d.endswith('-kl'):
print(d)
model_path = f"{root_paht}/{d}"
for idx in range(99,37099, 1000):
for i in range(5):
true_img = cv2.imread(f"{model_path}/{idx}_true_{i}.png")
fake_img = cv2.imread(f"{model_path}/{idx}_out_{i}.png")
mse_kl.append(evaluator.MSE(true_img, fake_img))
psnr_kl.append(evaluator.PSNR(true_img, fake_img))
lpips_score = evaluator.LPIPS(true_img, fake_img)
lpips_vgg_kl.append(lpips_score[0])
lpips_alex_kl.append(lpips_score[1])
ssim_kl.append(evaluator.SSIM(true_img, fake_img))
mse_mean = format(np.round(np.mean(mse), 2), '.2f')
psnr_mean = format(np.round(np.mean(psnr), 2), '.2f')
lpips_vgg_mean = format(np.round(np.mean(lpips_vgg), 2), '.2f')
lpips_alex_mean = format(np.round(np.mean(lpips_alex), 2), '.2f')
ssim_mean = format(np.round(np.mean(ssim), 3), '.3f')
mse_kl_mean = format(np.round(np.mean(mse_kl), 2), '.2f')
psnr_kl_mean = format(np.round(np.mean(psnr_kl), 2), '.2f')
lpips_vgg_kl_mean = format(np.round(np.mean(lpips_vgg_kl), 2), '.2f')
lpips_alex_kl_mean = format(np.round(np.mean(lpips_alex_kl), 2), '.2f')
ssim_kl_mean = format(np.round(np.mean(ssim_kl), 3), '.3f')
print(f"{mse_mean} & {mse_kl_mean} & {psnr_mean} & {psnr_kl_mean} & {lpips_vgg_mean} & {lpips_vgg_kl_mean} & {lpips_alex_mean} & {lpips_alex_kl_mean} & {ssim_mean} & {ssim_kl_mean}")
def white_black():
evaluator = Quantitation()
white_32 = np.ones((32,32,3), dtype=np.uint8)*255
black_32 = np.zeros((32,32,3), dtype=np.uint8)
white_256 = np.ones((256,256,3), dtype=np.uint8)*255
black_256 = np.zeros((256,256,3), dtype=np.uint8)
mse_32 = evaluator.MSE(white_32, black_32)
mse_256 = evaluator.MSE(white_256, black_256)
psnr_32 = evaluator.PSNR(white_32, black_32)
psnr_256 = evaluator.PSNR(white_256, black_256)
lpips_vgg_32 = evaluator.LPIPS(white_32, black_32)[0]
lpips_vgg_256 = evaluator.LPIPS(white_256, black_256)[0]
lpips_alex_32 = evaluator.LPIPS(white_32, black_32)[1]
lpips_alex_256 = evaluator.LPIPS(white_256, black_256)[1]
ssim_32 = evaluator.SSIM(white_32, black_32)
ssim_256 = evaluator.SSIM(white_256, black_256)
print(f"32: {format(mse_32, '.2f')} & {format(psnr_32, '.2f')} & {format(lpips_vgg_32, '.2f')} & {format(lpips_alex_32, '.2f')} & {format(ssim_32, '.3f')}")
print(f"256: {format(mse_256, '.2f')} & {format(psnr_256, '.2f')} & {format(lpips_vgg_256, '.2f')} & {format(lpips_alex_256, '.2f')} & {format(ssim_256, '.3f')}")
if __name__=='__main__':
# DLG_GRNN()
# GGL()
white_black()