-
Notifications
You must be signed in to change notification settings - Fork 11
/
datasets.py
242 lines (223 loc) · 12.1 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
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
import os
import cv2
import json
import torch
import scipy
import scipy.io as sio
from skimage import io
from torchvision import datasets, transforms
from torchvision.datasets.folder import ImageFolder, default_loader
from timm.data.constants import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
from timm.data import create_transform
class general_dataset(ImageFolder):
def __init__(self, root, train=True, transform=None, target_transform=None, mode=None,is_individual_prompt=False,**kwargs):
self.dataset_root = root
self.loader = default_loader
self.target_transform = None
self.transform = transform
if mode == 'super' and is_individual_prompt==False:
train_list_path = os.path.join(self.dataset_root, 'train800.txt')
elif mode == 'search' and is_individual_prompt==False:
train_list_path = os.path.join(self.dataset_root, 'val200.txt')
else:
train_list_path = os.path.join(self.dataset_root, 'train800val200.txt')
if mode != 'search':
test_list_path = os.path.join(self.dataset_root, 'test.txt')
else:
test_list_path = os.path.join(self.dataset_root, 'val200.txt')
self.samples = []
if train:
with open(train_list_path, 'r') as f:
for line in f:
img_name = line.split(' ')[0]
label = int(line.split(' ')[1])
self.samples.append((os.path.join(root,img_name), label))
else:
with open(test_list_path, 'r') as f:
for line in f:
img_name = line.split(' ')[0]
label = int(line.split(' ')[1])
self.samples.append((os.path.join(root,img_name), label))
class general_dataset_few_shot(ImageFolder):
def __init__(self, root, dataset,train=True, transform=None, target_transform=None, mode=None,is_individual_prompt=False,shot=2,seed=0,**kwargs):
self.dataset_root = root
self.dataset = dataset.replace('-FS','')
self.loader = default_loader
self.target_transform = None
self.transform = transform
if mode == 'super' and is_individual_prompt==False:
train_list_path = os.path.join(self.dataset_root, 'annotations/train_meta.list.num_shot_'+str(shot)+'.seed_'+str(seed))
elif mode == 'search' and is_individual_prompt==False:
if 'imagenet' in root:
train_list_path = os.path.join(self.dataset_root, 'annotations/unofficial_val_list_4_shot16seed0')
else:
train_list_path = os.path.join(self.dataset_root, 'annotations/val_meta.list')
else:
if 'imagenet' in root and self.dataset != 'imagenet':
train_list_path = os.path.join(self.dataset_root, 'annotations/val_meta.list')
else:
train_list_path = os.path.join(self.dataset_root, 'annotations/train_meta.list.num_shot_'+str(shot)+'.seed_'+str(seed))
if mode == 'search':
test_list_path = os.path.join(self.dataset_root, 'annotations/train_meta.list.num_shot_1.seed_0')
elif 'imagenet' in root:
test_list_path = os.path.join(self.dataset_root, 'annotations/val_meta.list')
else:
test_list_path = os.path.join(self.dataset_root, 'annotations/test_meta.list')
self.samples = []
if train:
with open(train_list_path, 'r') as f:
for line in f:
img_name = line.rsplit(' ',1)[0]
label = int(line.rsplit(' ',1)[1])
if 'stanford_cars' in root or ('imagenet' in root and 'imagenet' != self.dataset):
self.samples.append((os.path.join(root,img_name), label))
elif 'imagenet' == self.dataset:
self.samples.append((os.path.join(root+'/train',img_name), label))
else:
self.samples.append((os.path.join(root+'/images',img_name), label))
else:
with open(test_list_path, 'r') as f:
for line in f:
img_name = line.rsplit(' ',1)[0]
label = int(line.rsplit(' ',1)[1])
if 'stanford_cars' in root or ('imagenet' in root and 'imagenet' != self.dataset):
self.samples.append((os.path.join(root,img_name), label))
elif 'imagenet' == self.dataset:
if mode == 'search':
self.samples.append((os.path.join(root+'/train',img_name), label))
else:
self.samples.append((os.path.join(root+'/val',img_name), label))
else:
self.samples.append((os.path.join(root+'/images',img_name), label))
def build_dataset(is_train, args, folder_name=None,is_individual_prompt=False):
transform = build_transform(is_train, args)
if args.data_set == 'CIFAR10':
dataset = datasets.CIFAR10(args.data_path, train=is_train, transform=transform, download=True)
nb_classes = 10
elif args.data_set == 'CIFAR100':
dataset = datasets.CIFAR100(args.data_path, train=is_train, transform=transform, download=True)
nb_classes = 100
elif args.data_set == 'CARS':
dataset = Cars196(args.data_path, train=is_train, transform=transform)
nb_classes = 196
elif args.data_set == 'PETS':
dataset = Pets(args.data_path, train=is_train, transform=transform)
nb_classes = 37
elif args.data_set == 'FLOWERS':
dataset = Flowers(args.data_path, train=is_train, transform=transform)
nb_classes = 102
elif args.data_set == 'IMNET':
dataset = ImageNet(args.data_path, train=is_train, transform=transform)
nb_classes = 1000
elif args.data_set == 'EVO_IMNET':
root = os.path.join(args.data_path, folder_name)
dataset = datasets.ImageFolder(root, transform=transform)
nb_classes = 1000
elif args.data_set == 'INAT':
dataset = INatDataset(args.data_path, train=is_train, year=2018,
category=args.inat_category, transform=transform)
nb_classes = dataset.nb_classes
elif args.data_set == 'INAT19':
dataset = INatDataset(args.data_path, train=is_train, year=2019,
category=args.inat_category, transform=transform)
nb_classes = dataset.nb_classes
elif args.data_set == 'clevr_count':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 8
elif args.data_set == 'diabetic_retinopathy':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 5
elif args.data_set == 'dsprites_loc':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 16
elif args.data_set == 'dtd':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 47
elif args.data_set == 'kitti':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 4
elif args.data_set == 'oxford_pet':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 37
elif args.data_set == 'resisc45':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 45
elif args.data_set == 'smallnorb_ele':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 9
elif args.data_set == 'svhn':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 10
elif args.data_set == 'cifar100':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 100
elif args.data_set == 'clevr_dist':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 6
elif args.data_set == 'caltech101':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 102
elif args.data_set == 'dmlab':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 6
elif args.data_set == 'dsprites_ori':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 16
elif args.data_set == 'eurosat':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 10
elif args.data_set == 'oxford_flowers102':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 102
elif args.data_set == 'patch_camelyon':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 2
elif args.data_set == 'smallnorb_azi':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 18
elif args.data_set == 'sun397':
dataset = general_dataset(args.data_path, train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt)
nb_classes = 397
elif '-FS' in args.data_set:
dataset = general_dataset_few_shot(args.data_path, args.data_set,train=is_train, transform=transform,mode=args.mode,is_individual_prompt=is_individual_prompt,shot=args.few_shot_shot,seed=args.few_shot_seed)
if 'stanford_cars' in args.data_set:
nb_classes = 196
elif 'oxford_flowers' in args.data_set:
nb_classes = 102
elif 'food-101' in args.data_set:
nb_classes = 101
elif 'oxford_pets'in args.data_set:
nb_classes = 37
elif 'fgvc_aircraft' in args.data_set:
nb_classes = 100
elif 'imagenet' in args.data_set:
nb_classes = 1000
return dataset, nb_classes
def build_transform(is_train, args):
if not args.no_aug and is_train and args.mode != 'search':
transform = create_transform(
input_size=args.input_size,
is_training=True,
color_jitter=args.color_jitter,
auto_augment=args.aa,
interpolation=args.train_interpolation,
re_prob=args.reprob,
re_mode=args.remode,
re_count=args.recount,
)
return transform
t = []
if args.direct_resize:
size = args.input_size
else:
size = int((256 / 224) * args.input_size)
t.append(
transforms.Resize((size,size), interpolation=3) # to maintain same ratio w.r.t. 224 images
)
t.append(transforms.CenterCrop(args.input_size))
t.append(transforms.ToTensor())
if args.inception:
t.append(transforms.Normalize(IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD))
else:
t.append(transforms.Normalize(IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD))
return transforms.Compose(t)