-
Notifications
You must be signed in to change notification settings - Fork 2
/
data_helper.py
63 lines (50 loc) · 1.97 KB
/
data_helper.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
import scipy.io
import re
import glob
import numpy as np
from scipy import misc
root_directory = '/Users/vihanggodbole/Developer/restaurant-menu-ocr/English/Fnt/'
numbers = re.compile(r'(\d+)')
def numericalSort(value):
parts = numbers.split(value)
parts[1::2] = map(int, parts[1::2])
return parts
class Data(object):
def __init__(self):
self.training_input = np.zeros((50000, 16384))
self.training_labels = np.zeros((50000, 62))
self.test_input = np.zeros((12992, 16384))
self.test_labels = np.zeros((12992, 62))
def load_data(self):
# get paths of images
paths = sorted(glob.glob(root_directory + '**/*.png'),
key=numericalSort)
r_state = np.random.get_state()
np.random.shuffle(paths) # shuffle paths
# get images
for index, path in enumerate(paths):
# get image in grayscale. shape(28,28) and reshape it
grayscale_image = misc.imread(path, mode='L').reshape(1, 16384)
if index < 50000:
self.training_input[index] = grayscale_image
else:
self.test_input[index - 50000] = grayscale_image
# get classification labels
mat = scipy.io.loadmat('lists_var_size.mat')
training_labels = mat['list'][0, 0][
'ALLlabels'] # shape = (62992, 1)
# shuffle training data
np.random.set_state(r_state)
# shuffle the vector in the same way as images
np.random.shuffle(training_labels)
for i in range(len(paths)):
# since training_lables[i] returns an ndarray. of 1x1
j = training_labels[i][0] - 1
# eg: if the sample is 'z' then training_labels[i] = 62. Thus,
# temp[0] = [0,0...0,1]
t = np.zeros((1, 62), dtype=np.int)
t[0][j] = 1
if i < 50000:
self.training_labels[i] = t
else:
self.test_labels[i - 50000] = t