forked from martinkersner/train-CRF-RNN
/
data2lmdb.py
178 lines (140 loc) · 4.72 KB
/
data2lmdb.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
#!/usr/bin/env python
# Martin Kersner, m.kersner@gmail.com
# 2016/01/18
from __future__ import print_function
import os
import sys
import lmdb
from random import shuffle
from skimage.io import imread
from scipy.misc import imresize
import numpy as np
from PIL import Image
import caffe
from utils import get_id_classes
def main():
##
preprocess_mode = 'pad'
im_sz = 500
#file_src_images = 'train.txt'
class_names = ['bird', 'bottle', 'chair']
test_ratio = 0.1
##
ext = '.png'
class_ids = get_id_classes(class_names)
train_imgs, test_imgs = split_train_test_imgs(class_names, ext, test_ratio)
## Train
# Images
print('Train images')
path_src = 'images/'
path_dst = 'train_images_3_lmdb'
#train_imgs = get_src_imgs(file_src_images, ext)
convert2lmdb(path_src, train_imgs, path_dst, class_ids, preprocess_mode, im_sz, 'image')
# Labels
print('Train labels')
path_src = 'labels/'
path_dst = 'train_labels_3_lmdb'
#train_imgs = get_src_imgs(file_src_images, ext)
convert2lmdb(path_src, train_imgs, path_dst, class_ids, preprocess_mode, im_sz, 'label')
## Test
# Images
print('Test images')
path_src = 'images/'
path_dst = 'test_images_3_lmdb'
convert2lmdb(path_src, test_imgs, path_dst, class_ids, preprocess_mode, im_sz, 'image')
# Labels
print('Test labels')
path_src = 'labels/'
path_dst = 'test_labels_3_lmdb'
convert2lmdb(path_src, test_imgs, path_dst, class_ids, preprocess_mode, im_sz, 'label')
def split_train_test_imgs(class_names, ext, test_ratio):
train_imgs = []
test_imgs = []
for i in class_names:
file_name = i + '.txt'
num_lines = get_num_lines(file_name)
num_test_imgs = test_ratio * num_lines
current_line = 1
with open(file_name, 'rb') as f:
for line in f:
if current_line < num_test_imgs:
test_imgs.append(line.strip() + ext)
else:
train_imgs.append(line.strip() + ext)
current_line += 1
shuffle(train_imgs)
shuffle(test_imgs)
print(str(len(train_imgs)) + ' train images')
print(str(len(test_imgs)) + ' test images')
return train_imgs, test_imgs
def get_num_lines(file_name):
num_lines = 0
with open(file_name, 'rb') as f:
for line in f:
num_lines += 1
return num_lines
def get_src_imgs(file_name, ext):
src_imgs = []
with open(file_name, 'rb') as f:
for img_file in f:
img_file = img_file.strip()
src_imgs.append(img_file + ext)
return src_imgs
def convert2lmdb(path_src, src_imgs, path_dst, class_ids, preprocess_mode, im_sz, data_mode):
if os.path.isdir(path_dst):
print('DB ' + path_dst + ' already exists.\n'
'Skip creating ' + path_dst + '.', file=sys.stderr)
return None
if data_mode == 'label':
lut = create_lut(class_ids)
db = lmdb.open(path_dst, map_size=int(1e12))
with db.begin(write=True) as in_txn:
for idx, img_name in enumerate(src_imgs):
#img = imread(os.path.join(path_src + img_name))
img = np.array(Image.open(os.path.join(path_src + img_name)))
img = img.astype(np.uint8)
if data_mode == 'label':
img = preprocess_label(img, lut, preprocess_mode, im_sz)
elif data_mode == 'image':
img = preprocess_image(img, preprocess_mode, im_sz)
img_dat = caffe.io.array_to_datum(img)
in_txn.put('{:0>10d}'.format(idx), img_dat.SerializeToString())
def preprocess_image(img, mode, im_sz):
img = preprocess_data(img, mode, im_sz, 'image')
img = img[:,:,::-1] # RGB to BGR
img = img.transpose((2,0,1))
return img
def preprocess_label(img, lut, mode, im_sz):
img = preprocess_data(img, mode, im_sz, 'label')
img = lut[img]
img = np.expand_dims(img, axis=0)
#img = _2D_to_ND(img, len(np.unique(lut)))
#img = img.transpose((2,0,1))
return img
def create_lut(class_ids, max_id=256):
lut = np.zeros(max_id, dtype=np.uint8)
new_index = 1
for i in class_ids:
lut[i] = new_index
new_index += 1
return lut
def _2D_to_ND(label, n_levels):
nd_label = np.zeros((label.shape[0], label.shape[1], n_levels)).astype(np.uint8)
for l in range(n_levels):
nd_label[:,:,l] = (label==l) * 1
return nd_label
def preprocess_data(img, preprocess_mode, im_sz, data_mode):
if preprocess_mode == 'pad':
if data_mode == 'image':
img = np.pad(img, ((0, im_sz-img.shape[0]), (0, im_sz-img.shape[1]), (0,0)), 'constant', constant_values=(0))
elif data_mode == 'label':
img = np.pad(img, ((0, im_sz-img.shape[0]), (0, im_sz-img.shape[1])), 'constant', constant_values=(0))
else:
print('Invalid data mode.', file=sys.stderr)
elif preprocess_mode == 'res':
img = imresize(img, (im_sz, im_sz), interp='bilinear')
else:
print('Invalid preprocess mode.', file=sys.stderr)
return img
if __name__ == '__main__':
main()