/
task5_utils.py
137 lines (110 loc) · 4.59 KB
/
task5_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
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
import random
import numpy as np
from scipy.ndimage.interpolation import rotate
import matplotlib.pyplot as plt
from skimage.util import crop
from scipy.misc import imread
from skimage.transform import resize
import os
def rotate_image(image, angles=range(360)):
rotation_angle = random.choice(angles)
rotated_image = rotate(image, rotation_angle, reshape=False)
return rotated_image, rotation_angle
def rotate_images(images, num_rotated_per_image, angles=range(360)):
rotated_images = []
rotation_angles = []
for image in (images):
for n in range(num_rotated_per_image):
image = np.squeeze(image)
rotated_image, rotation_angle = rotate_image(image, angles)
rotated_image = rotated_image.reshape(1, rotated_image.shape[0], rotated_image.shape[1])
rotated_images.append(rotated_image)
rotation_angles.append(rotation_angle)
return np.asarray(rotated_images), np.asarray(rotation_angles)
def angle_to_class_label(y, angles=range(360)):
class_names = {}
for i, angle in enumerate(angles):
class_names[angle] = i
class_labels = [class_names[angle] for angle in y]
return np.asarray(class_labels)
def class_label_to_angle(y, angles=range(360)):
class_names = {}
for i, angle in enumerate(angles):
class_names[i] = angle
class_labels = [class_names[i] for i in y]
return np.asarray(class_labels)
def plot_examples(X, y, y_angles, y_predicted_angles,
num_test_images=5, number=None, angle=None, only_errors=False):
if only_errors:
mask = np.where(y_predicted_angles != y_angles)[0]
X = X[mask]
y = y[mask]
y_predicted_angles = y_predicted_angles[mask]
y_angles = y_angles[mask]
if angle is not None and number is not None:
indices = np.intersect1d(np.where(y_angles == angle)[0], np.where(y == number)[0])
elif angle is not None:
indices = np.where(y_angles == angle)[0]
elif number is not None:
indices = np.where(y == number)[0]
else:
indices = len(y_angles)
mask = np.random.choice(indices, num_test_images)
true_angles = y_angles[mask]
predicted_angles = y_predicted_angles[mask]
plt.rcParams['figure.figsize'] = (10.0, 2 * num_test_images)
fig_number = 0
for i in range(num_test_images):
rotated_image = X[mask[i]][0]
original_image = rotate(rotated_image, -true_angles[i])
corrected_image = rotate(rotated_image, -predicted_angles[i])
fig_number += 1
plt.subplot(num_test_images, 3, fig_number)
plt.imshow(original_image)
fig_number += 1
plt.subplot(num_test_images, 3, fig_number)
plt.title('Angle: {0}'.format(true_angles[i]))
plt.imshow(rotated_image)
fig_number += 1
plt.subplot(num_test_images, 3, fig_number)
reconstructed_angle = angle_difference(predicted_angles[i], true_angles[i])
plt.title('Angle: {0}'.format(reconstructed_angle))
plt.imshow(corrected_image)
plt.tight_layout(pad=0.4, w_pad=0.5, h_pad=1.0)
def plot_error_hist(y, y_angles, y_predicted_angles, error_type='angle'):
error_per_class = []
classes = range(np.max(y) + 1)
for i in classes:
mask = np.where(y == i)
y_test_angle_i = y_angles[mask]
y_predicted_angles_i = y_predicted_angles[mask]
if error_type == 'classification':
error_per_class.append((float(np.sum(y_test_angle_i != y_predicted_angles_i)) /
len(y_predicted_angles_i)) * 100)
elif error_type == 'angle':
error_per_class.append(float(np.mean(abs(angle_difference(y_test_angle_i, y_predicted_angles_i)))))
plt.rcParams['figure.figsize'] = (10.0, 10.0)
ax = plt.subplot(111)
ax.bar(classes, error_per_class, align='center')
ax.set_xticks(classes)
def angle_difference(x, y):
return 180 - abs(abs(x - y - 180))
def open_and_resize(im, size):
h, w, c = im.shape
if h < w:
before_n = (w - h) / 2
after_n = w - (h + before_n)
im_cropped = crop(im, ((0, 0), (before_n, after_n), (0, 0)))
else:
before_n = (h - w) / 2
after_n = h - (w + before_n)
im_cropped = crop(im, ((before_n, after_n), (0, 0), (0, 0)))
im_cropped = resize(im_cropped, (size, size, c))
return im_cropped
def load_and_resize_images(folder, size):
images = []
image_paths = os.listdir(folder)
for image_path in image_paths:
im = imread(os.path.join(folder, image_path))
images.append(open_and_resize(im, size))
return np.asarray(images)