-
Notifications
You must be signed in to change notification settings - Fork 0
/
datasets.py
204 lines (185 loc) · 7.47 KB
/
datasets.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
201
202
203
204
# =============================================================================
# Import required libraries
# =============================================================================
import os
import json
import numpy as np
from PIL import Image
import torch
from torch.utils.data import DataLoader
import torchvision.transforms as transforms
from sklearn.utils import shuffle
# =============================================================================
# Create annotation dataset
# =============================================================================
class AnnotationDataset(torch.utils.data.Dataset):
'''
image dim: (batch-size, 3, image-size, image-size)
binary-annotations dim: (batch-size, (number of classes))
annotations-X dim: (batch-size, max-seq-len + 1)
label_lengths dim: (batch-size)
'''
def __init__(self,
root,
annotation_path,
max_seq_len,
sorted_labels=None,
transforms=None):
self.root = root
# maximum number of annotated labels + START and STOP
self.max_seq_len = max_seq_len + 2
self.transforms = transforms
#
with open(annotation_path) as fp:
json_data = json.load(fp)
samples = json_data['samples']
samples = shuffle(samples, random_state=0)
self.classes = json_data['labels']
#
self.imgs = []
self.annotations = []
for sample in samples:
self.imgs.append(sample['image_name'])
self.annotations.append(sample['image_labels'])
# converting all labels of each image into a binary array
# of the class length
self.binary_annotations = []
for idx in range(len(self.annotations)):
item = self.annotations[idx]
vector = [cls in item for cls in self.classes]
self.binary_annotations.append(np.array(vector, dtype=float))
# sorting each label set according to rare-first ordering
# the rare-first order put the rarer label before
# the more frequent ones (based on label frequency in the dataset)
if sorted_labels is not None:
for idx in range(len(self.annotations)):
self.annotations[idx] = sorted(
self.annotations[idx], key=lambda x: sorted_labels[x])
#
classes_new = self.classes.copy()
classes_new.append('stop')
classes_new.append('start')
self.word_map = {cl: i for i, cl in enumerate(classes_new)}
def __getitem__(self, idx):
# image
img_path = os.path.join(self.root, self.imgs[idx])
image = Image.open(img_path).convert("RGB")
if self.transforms is not None:
image = self.transforms(image)
binary_annotations = torch.tensor(self.binary_annotations[idx])
# annotations
annotations_X = []
annotations_X.append(self.word_map['start'])
for c in self.annotations[idx]:
annotations_X.append(self.word_map[c])
for _ in range(self.max_seq_len - len(self.annotations[idx])):
annotations_X.append(self.word_map['stop'])
annotations_X = torch.tensor(annotations_X)
label_lengths = len(self.annotations[idx])
return image, binary_annotations, annotations_X, label_lengths
def __len__(self):
return len(self.imgs)
# =============================================================================
# Make data loader
# =============================================================================
def get_mean_std():
mean = [0.3928, 0.4079, 0.3531]
std = [0.2559, 0.2436, 0.2544]
return mean, std
def get_transforms(args):
mean, std = get_mean_std()
transform_train = transforms.Compose([
transforms.Resize((args.image_size, args.image_size)),
transforms.RandomVerticalFlip(),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(
mean=mean,
std=std,
)
])
transform_validation = transforms.Compose([
transforms.Resize((args.image_size, args.image_size)),
transforms.ToTensor(),
transforms.Normalize(
mean=mean,
std=std,
)
])
return transform_train, transform_validation
def sort_labels(annotation_path):
'''
sorting all labels in training data based on their frequency
'''
with open(annotation_path) as fp:
json_data = json.load(fp)
samples = json_data['samples']
#
annotations = []
for sample in samples:
annotations.append(sample['image_labels'])
all_annotations = [
item for sublist in annotations for item in sublist]
num_labels = {i: all_annotations.count(i) for i in all_annotations}
return {k: v for k, v in sorted(
num_labels.items(), key=lambda item: item[1])}
def make_data_loader(args):
root_dir = args.data_root_dir
transform_train, transform_validation = get_transforms(args)
if args.sort:
sorted_labels = sort_labels(os.path.join(root_dir, 'train.json'))
else:
sorted_labels = None
#
train_set = AnnotationDataset(root=os.path.join(root_dir, 'images'),
annotation_path=os.path.join(
root_dir, 'train.json'),
max_seq_len=args.max_seq_len,
sorted_labels=sorted_labels,
transforms=transform_train)
train_loader = DataLoader(train_set,
batch_size=args.batch_size,
num_workers=args.num_workers,
shuffle=True)
#
validation_set = AnnotationDataset(root=os.path.join(root_dir, 'images'),
annotation_path=os.path.join(
root_dir, 'test.json'),
max_seq_len=args.max_seq_len,
sorted_labels=sorted_labels,
transforms=transform_validation)
validation_loader = DataLoader(validation_set,
batch_size=args.batch_size,
num_workers=args.num_workers,
shuffle=False)
# all vocabulary except START and STOP
classes = train_set.classes
#
classes_new = classes.copy()
classes_new.append('stop')
classes_new.append('start')
word_map = {cl: i for i, cl in enumerate(classes_new)}
return train_loader, validation_loader, classes, word_map
# =============================================================================
# Word embedding
# =============================================================================
def word_embedding(glove_path, classes):
'''
embedding each class including START and STOP
'''
with open(glove_path, 'r', encoding='UTF-8') as f:
words = set()
word_to_vec_map = {}
for line in f:
line = line.strip().split()
curr_word = line[0]
words.add(curr_word)
word_to_vec_map[curr_word] = np.array(line[1:], dtype=np.float32)
#
emb = []
classes_plus = classes.copy()
classes_plus.append('stop')
classes_plus.append('start')
for word in classes_plus:
emb.append(word_to_vec_map[word])
return torch.from_numpy(np.array(emb))