-
Notifications
You must be signed in to change notification settings - Fork 10
/
preprocess.py
125 lines (108 loc) · 4.43 KB
/
preprocess.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
'''
Vectorize batik dataset
Author: [email protected]
'''
import os
import sys
import cv2
import numpy as np
import json
import progressbar
import tables
import argparse
import imutils
from random import sample
from helper import normalize, resize, zoomin
# config
EXPECTED_MAX = 100.0
EXPECTED_MIN = -1 * EXPECTED_MAX
FILTER_THRESHOLD = -90.0
DATASET_PATH = 'dataset.h5'
DATASET_INDEX_PATH = 'dataset.index.json'
# global vars
EXPECTED_SIZE = 224
EXPECTED_CHANNELS = 3
# EXPECTED_DIM = (EXPECTED_CHANNELS, EXPECTED_SIZE, EXPECTED_SIZE) # Theano
EXPECTED_DIM = (EXPECTED_SIZE, EXPECTED_SIZE, EXPECTED_CHANNELS) # TensorFlow
EXPECTED_CLASS = 5
MAX_VALUE = 255
MEDIAN_VALUE = MAX_VALUE / 2.0
def append_data_and_label(m, c, dataset, labels):
# m = np.transpose(m, (2, 0, 1)) # Theano
assert m.shape == EXPECTED_DIM
dataset.append(np.array([m]))
# one-hot encoding
label = np.zeros(EXPECTED_CLASS)
label[c] = 1.0
assert label.shape == (EXPECTED_CLASS,)
labels.append(np.array([label]))
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Preprocess and vectorize images dataset', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('dataset_path', help="Path to raw dataset directory")
parser.add_argument('--classes_file', '-c', default=DATASET_INDEX_PATH, help="Output path for dataset classes index")
parser.add_argument('--vector_file', '-v', default=DATASET_PATH, help="Output path for preprocessed and vectorized dataset")
parser.add_argument('--grayscale', '-g', action='store_true', help="Convert images to grayscale")
parser.add_argument('--rotate', '-r', metavar='ANGLE', type=int, default=0, help="Rotate images to certain angle (integer)")
parser.add_argument('--zoomin', '-z', metavar='SCALE', type=float, default=1.0, help="Zoom in images to certain Scale (float)")
args = parser.parse_args()
mypath = args.dataset_path
index_file = args.classes_file
dataset_file = args.vector_file
grayscale = args.grayscale
rotate = args.rotate
scale = args.zoomin
print(args)
# iterate dir content
stat = {}
label_indexes = {}
count = 0
i = 0
# pytables file
datafile = tables.open_file(dataset_file, mode='w')
data = datafile.create_earray(datafile.root, 'data', tables.Float32Atom(shape=EXPECTED_DIM), (0,), 'batik')
labels = datafile.create_earray(datafile.root, 'labels', tables.UInt8Atom(shape=(EXPECTED_CLASS)), (0,), 'batik')
# iterate subfolders
num_dir = len([name for name in os.listdir(mypath)])
bar = progressbar.ProgressBar(maxval=num_dir).start()
for f in os.listdir(mypath):
path = os.path.join(mypath, f)
# exclude Mix motif
if os.path.isdir(path) and f != 'Mix motif':
label_indexes[i] = f
for f_sub in os.listdir(path):
path_sub = os.path.join(path, f_sub)
if os.path.isfile(path_sub):
try:
img = cv2.imread(path_sub)
# rotate
img = imutils.rotate(img, rotate) if rotate is not 0 else img
# scale
img = zoomin(img, scale) if scale > 1 else img
# grayscale with 3 channels
img = cv2.cvtColor(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), cv2.COLOR_GRAY2RGB) if grayscale else img
# normalize and filter
img = normalize(img, EXPECTED_MAX, MEDIAN_VALUE)
# gather stat
stat[img.shape] = stat[img.shape] + 1 if img.shape in stat else 1
r = resize(img, EXPECTED_SIZE)
append_data_and_label(r, i, data, labels)
except Exception as err:
print(err)
print(path_sub)
sys.exit(0)
i += 1
count += 1
bar.update(count)
bar.finish()
print('{} records saved'.format(data.nrows))
# write label index as json file
with open(index_file, 'w') as f:
json.dump(label_indexes, f)
print((data.nrows,) + data[0].shape)
print((labels.nrows,) + labels[0].shape)
print(label_indexes)
# print(stat)
assert data[0].shape == EXPECTED_DIM
assert labels[0].shape == (EXPECTED_CLASS,)
# close file
datafile.close()