-
Notifications
You must be signed in to change notification settings - Fork 8
/
imagelist2lmdb.py
200 lines (166 loc) · 6.11 KB
/
imagelist2lmdb.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
189
190
191
192
193
194
195
196
197
198
199
200
import os
from os import path
import lmdb
import msgpack
import numpy as np
import pandas as pd
import pyarrow as pa
from PIL import Image
from torch.utils.data import DataLoader
from torchvision.datasets import ImageFolder
from tqdm import tqdm
from image_utils import blur_image_faster
class ImageListRaw(ImageFolder):
ATTRIBUTES = {"race": 1, "gender": 2, "age": 3, "recognition": 1}
MAX_CLASS = {"race": 4, "gender": 1, "age": 100, "recognition": float("Inf")}
def __init__(self, attribute_name, source, mask_source, image_list, train=True):
self.max_blur = None
attribute = self.ATTRIBUTES[attribute_name]
max_class = self.MAX_CLASS[attribute_name]
image_names = pd.read_csv(image_list, delimiter=" ", header=None)
image_names = np.asarray(image_names)
# remove images that have labels outside of desired range
image_names = image_names[image_names[:, attribute].astype("int") >= 0]
image_names = image_names[image_names[:, attribute].astype("int") <= max_class]
if source is not None:
self.samples = [
path.join(source, image_name) for image_name in image_names[:, 0]
]
if mask_source is not None:
self.masks = [
path.join(mask_source, image_name)
for image_name in image_names[:, 0]
]
self.masks = [
f"{image_name[:-4]}_mask.png" for image_name in self.masks
]
else:
self.masks = None
else:
self.samples = image_names[:, 0]
self.targets = image_names[:, attribute].astype("int")
if attribute_name == "age":
self.classnum = np.max(self.targets)
else:
self.classnum = np.max(self.targets) + 1
def __len__(self):
return len(self.samples)
def __getitem__(self, index):
if self.masks is not None:
img = Image.open(self.samples[index]).convert("RGB")
mask = Image.open(self.masks[index])
img = blur_image_faster(
np.asarray(img).copy(), np.asarray(mask).copy(), max_blur=self.max_blur
)
img = Image.fromarray(img)
temp_file = f"/hd3/{index}{path.splitext(self.samples[index])[1]}"
img.save(temp_file)
with open(temp_file, "rb") as f:
img = f.read()
os.remove(temp_file)
else:
with open(self.samples[index], "rb") as f:
img = f.read()
return img, self.targets[index]
class CustomRawLoader(DataLoader):
def __init__(self, attribute, workers, source, mask_source, image_list):
self._dataset = ImageListRaw(attribute, source, mask_source, image_list)
super(CustomRawLoader, self).__init__(
self._dataset, num_workers=workers, collate_fn=lambda x: x
)
def list2lmdb(
attribute,
source,
mask_source,
image_list,
dest,
num_workers=16,
write_frequency=5000,
):
print("Loading dataset from %s" % image_list)
data_loader = CustomRawLoader(
attribute, num_workers, source, mask_source, image_list
)
name = f"{path.split(image_list)[1][:-4]}.lmdb"
lmdb_path = path.join(dest, name)
isdir = path.isdir(lmdb_path)
print(f"Generate LMDB to {lmdb_path}")
# sigmas = np.linspace(2, 16, 8).astype(int)
sigmas = [10]
image_size = 224
size = len(data_loader.dataset) * image_size * image_size * 3
if mask_source is not None:
size *= len(sigmas)
print(f"LMDB max size: {size}")
db = lmdb.open(
lmdb_path,
subdir=isdir,
map_size=size * 2,
readonly=False,
meminit=False,
map_async=True,
)
if mask_source is None:
print(len(data_loader.dataset))
txn = db.begin(write=True)
for idx, data in tqdm(enumerate(data_loader)):
image, label = data[0]
txn.put(
"{}".format(idx).encode("ascii"), msgpack.dumps((image, int(label)))
)
if idx % write_frequency == 0:
print("[%d/%d]" % (idx, len(data_loader)))
txn.commit()
txn = db.begin(write=True)
idx += 1
else:
print(len(data_loader.dataset) * len(sigmas))
txn = db.begin(write=True)
idx = 0
for sigma in sigmas:
data_loader.dataset.max_blur = sigma
for _, data in tqdm(enumerate(data_loader)):
image, label = data[0]
txn.put(
"{}".format(idx).encode("ascii"), msgpack.dumps((image, int(label)))
)
if idx % write_frequency == 0:
print("[%d/%d]" % (idx, len(data_loader)))
txn.commit()
txn = db.begin(write=True)
idx += 1
# finish iterating through dataset
txn.commit()
keys = ["{}".format(k).encode("ascii") for k in range(idx)]
with db.begin(write=True) as txn:
txn.put(b"__keys__", msgpack.dumps(keys))
txn.put(b"__len__", msgpack.dumps(len(keys)))
txn.put(b"__classnum__", msgpack.dumps(int(data_loader.dataset.classnum)))
print("Flushing database ...")
db.sync()
db.close()
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("--image_list", "-l", help="List of images.")
parser.add_argument("--source", "-s", help="Path to the images.")
parser.add_argument(
"--mask_source", "-ms", help="Path to the image masks [optional]."
)
parser.add_argument("--workers", "-w", help="Workers number.", default=16, type=int)
parser.add_argument(
"--attribute",
"-a",
help="Which attribute to load [race, gender, age, recognition].",
type=str,
)
parser.add_argument("--dest", "-d", help="Path to save the lmdb file.")
args = parser.parse_args()
list2lmdb(
args.attribute,
args.source,
args.mask_source,
args.image_list,
args.dest,
args.workers,
)