-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils_data.py
188 lines (148 loc) · 6.07 KB
/
utils_data.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
import numpy as np
import json
import os
from PIL import Image
import matplotlib.pyplot as plt
import math
def crop_images(images, upper, lower):
cropped_image = []
for image, top_cord, bottom_cord in zip(images, upper, lower):
cropped_image.append(image[top_cord[1]:bottom_cord[1], top_cord[0]:bottom_cord[0], :])
return np.array(cropped_image, dtype=object)
def read_extended_dataset(root_folder='./images/', extended_gt_json='./images/gt_reduced.json', w=60, h=80):
"""
reads the extended ground truth, returns:
images: the images in color (80x60x3)
shape labels: array of strings
color labels: array of arrays of strings
upper_left_coord: (x, y) coordinates of the window top left
lower_right_coord: (x, y) coordinates of the window bottom right
background: array of booleans indicating if the defined window contains background or not
"""
ground_truth_extended = json.load(open(extended_gt_json, 'r'))
img_names, class_labels, color_labels, upper, lower, background = [], [], [], [], [], []
for k, v in ground_truth_extended.items():
img_names.append(os.path.join(root_folder, 'train', k))
class_labels.append(v[0])
color_labels.append(v[1])
upper.append(v[2])
lower.append(v[3])
background.append(True if v[4] == 1 else False)
imgs = load_imgs(img_names, w, h, True)
idxs = np.arange(imgs.shape[0])
np.random.seed(42)
np.random.shuffle(idxs)
imgs = imgs[idxs]
class_labels = np.array(class_labels)[idxs]
color_labels = np.array(color_labels, dtype=object)[idxs]
upper = np.array(upper)[idxs]
lower = np.array(lower)[idxs]
background = np.array(background)[idxs]
return imgs, class_labels, color_labels, upper, lower, background
def read_dataset(root_folder='./images/', gt_json='./test/gt.json', w=60, h=80, with_color=True):
"""
reads the dataset (train and test), returns the images and labels (class and colors) for both sets
"""
np.random.seed(123)
ground_truth = json.load(open(gt_json, 'r'))
train_img_names, train_class_labels, train_color_labels = [], [], []
test_img_names, test_class_labels, test_color_labels = [], [], []
for k, v in ground_truth['train'].items():
train_img_names.append(os.path.join(root_folder, 'train', k))
train_class_labels.append(v[0])
train_color_labels.append(v[1])
for k, v in ground_truth['test'].items():
test_img_names.append(os.path.join(root_folder, 'test', k))
test_class_labels.append(v[0])
test_color_labels.append(v[1])
train_imgs = load_imgs(train_img_names, w, h, with_color)
test_imgs = load_imgs(test_img_names, w, h, with_color)
np.random.seed(42)
idxs = np.arange(train_imgs.shape[0])
np.random.shuffle(idxs)
train_imgs = train_imgs[idxs]
train_class_labels = np.array(train_class_labels)[idxs]
train_color_labels = np.array(train_color_labels, dtype=object)[idxs]
idxs = np.arange(test_imgs.shape[0])
np.random.shuffle(idxs)
test_imgs = test_imgs[idxs]
test_class_labels = np.array(test_class_labels)[idxs]
test_color_labels = np.array(test_color_labels, dtype=object)[idxs]
return train_imgs, train_class_labels, train_color_labels, test_imgs, test_class_labels, test_color_labels
def load_imgs(img_names, w, h, with_color):
imgs = []
for tr in img_names:
imgs.append(read_one_img(tr + '.jpg', w, h, with_color))
return np.array(imgs)
def read_one_img(img_name, w, h, with_color):
img = Image.open(img_name)
if with_color:
img = img.convert("RGB")
else:
img = img.convert("L")
if img.size != (w, h):
img = img.resize((w, h))
return np.array(img)
def visualize_retrieval(imgs, topN, info=None, ok=None, title='', query=None):
def add_border(color):
return np.stack(
[np.pad(imgs[i, :, :, c], 3, mode='constant', constant_values=color[c]) for c in range(3)], axis=2
)
columns = 4
rows = math.ceil(topN/columns)
if query is not None:
fig = plt.figure(figsize=(10, 8*6/8))
columns += 1
fig.add_subplot(rows, columns, 1+columns)
plt.imshow(query)
plt.axis('off')
plt.title(f'query', fontsize=8)
else:
fig = plt.figure(figsize=(8, 8*6/8))
for i in range(min(topN, len(imgs))):
sp = i+1
if query is not None:
sp = (sp - 1) // (columns-1) + 1 + sp
fig.add_subplot(rows, columns, sp)
if ok is not None:
im = add_border([0, 255, 0] if ok[i] else [255, 0, 0])
else:
im = imgs[i]
plt.imshow(im)
plt.axis('off')
if info is not None:
plt.title(f'{info[i]}', fontsize=8)
plt.gcf().suptitle(title)
plt.show()
# Visualize k-mean with 3D plot
def Plot3DCloud(km, rows=1, cols=1, spl_id=1):
ax = plt.gcf().add_subplot(rows, cols, spl_id, projection='3d')
for k in range(km.K):
Xl = km.X[km.labels == k, :]
ax.scatter(
Xl[:, 0], Xl[:, 1], Xl[:, 2], marker='.', c=km.centroids[np.ones((Xl.shape[0]), dtype='int') * k, :] / 255
)
plt.xlabel('dim 1')
plt.ylabel('dim 2')
ax.set_zlabel('dim 3')
return ax
def visualize_k_means(kmeans, img_shape):
def prepare_img(x, img_shape):
x = np.clip(x.astype('uint8'), 0, 255)
x = x.reshape(img_shape)
return x
fig = plt.figure(figsize=(8, 8))
X_compressed = kmeans.centroids[kmeans.labels]
X_compressed = prepare_img(X_compressed, img_shape)
org_img = prepare_img(kmeans.X, img_shape)
fig.add_subplot(131)
plt.imshow(org_img)
plt.title('original')
plt.axis('off')
fig.add_subplot(132)
plt.imshow(X_compressed)
plt.axis('off')
plt.title('kmeans')
Plot3DCloud(kmeans, 1, 3, 3)
plt.title('núvol de punts')
plt.show()