forked from gondimribeiro/adv-attacks-vae
-
Notifications
You must be signed in to change notification settings - Fork 0
/
svhn_utils.py
99 lines (72 loc) · 2.48 KB
/
svhn_utils.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
import matplotlib
matplotlib.use("Agg")
import math
import matplotlib.pyplot as plt
import numpy as np
from scipy.io import loadmat
data_mean = 115.11177966923525
data_std = 50.819267906232888
img_dim = 32
max_pixel = 255.
n_chan = 3
def show(img, normalize, i, title="", num_rows=-3, path=None):
img = img.copy().reshape(img_dim, img_dim, 3)
if normalize:
img *= data_std
img += data_mean
img /= 255.
img = np.clip(img, 0., 1.)
if num_rows < 0:
plt.subplot(2, -num_rows, i)
else:
plt.subplot(num_rows, 2, i)
plt.imshow(img)
plt.title(title)
plt.axis("off")
def reconstruct(cvae, batch, path=None, normalize=True):
plt.figure(figsize=(8, 30))
num_imgs = batch.shape[0]
batch_rec = cvae.batch_reconstruct(batch)
for i in range(num_imgs):
show(batch[i], normalize, 2 * i + 1, 'original', num_imgs)
show(batch_rec[i], normalize, 2 * i + 2, 'reconstr', num_imgs)
if path is None:
plt.show()
else:
plt.savefig(path)
plt.close()
def load_data(normalize=True):
# LOAD DATA
svhn_train = loadmat('./data/svhn/train_32x32.mat')
svhn_test = loadmat('./data/svhn/test_32x32.mat')
train_x = np.rollaxis(svhn_train['X'], 3).astype(np.float32)
test_x = np.rollaxis(svhn_test['X'], 3).astype(np.float32)
train_y = svhn_train['y'].flatten() - 1
test_y = svhn_test['y'].flatten() - 1
idx = np.random.permutation(train_x.shape[0])
train_x = train_x[idx]
train_y = train_y[idx]
if normalize:
train_x = (train_x - data_mean) / data_std
test_x = (test_x - data_mean) / data_std
else:
train_x = train_x / 255.
test_x = test_x / 255.
val_idx = math.ceil(train_x.shape[0] * 0.1)
val_x = train_x[:val_idx]
val_y = train_y[:val_idx]
train_x = train_x[val_idx:]
train_y = train_y[val_idx:]
return train_x, train_y, val_x, val_y, test_x, test_y
def input(img, batch_size):
return np.tile(img, (batch_size, 1, 1, 1)).reshape(batch_size,
img_dim, img_dim, 3)
def dist(imgs, img, batch_size):
if imgs.shape == (batch_size, 32, 32, 3):
imgs = imgs.reshape(batch_size, 3072)
assert imgs.shape == (batch_size, 3072)
img = img.reshape((3072,))
imgs_pixels = imgs * data_std + data_mean
img_pixels = img * data_std + data_mean
diff = np.linalg.norm(imgs_pixels - img_pixels, axis=1)
return np.mean(diff), np.std(diff)