-
Notifications
You must be signed in to change notification settings - Fork 0
/
datatsets.py
127 lines (111 loc) · 5.48 KB
/
datatsets.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
import logging
import torch
from torch.utils.data import DataLoader, random_split
from torchvision import transforms
from torchvision.datasets import CIFAR10, MNIST, FashionMNIST, CelebA
def data_augmentation():
trans = [transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip()]
return trans
def get_dataset(args):
logging.info('Loading {} dataset'.format(args.dataset))
data_aug_trans = data_augmentation() if args.data_augmentation else []
if args.dataset == "CIFAR10":
train = CIFAR10(root="Datasets/",
download=True,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, *data_aug_trans,
transforms.ToTensor()]))
test = CIFAR10(root="Datasets/",
download=True,
train=False,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, *data_aug_trans,
transforms.ToTensor()]))
elif args.dataset == "MNIST":
if args.custom_mnist_download:
new_mirror = 'https://ossci-datasets.s3.amazonaws.com/mnist'
MNIST.resources = [
('/'.join([new_mirror, url.split('/')[-1]]), md5)
for url, md5 in MNIST.resources
]
train = MNIST(root="Datasets/MNIST",
download=True,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, *data_aug_trans, transforms.ToTensor(),
transforms.Normalize([0.5], [0.5])]),
)
test = MNIST(root="Datasets/",
download=True,
train=False,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, *data_aug_trans, transforms.ToTensor(),
transforms.Normalize([0.5], [0.5])]),
)
elif args.dataset == "MNIST_128":
if args.custom_mnist_download:
new_mirror = 'https://ossci-datasets.s3.amazonaws.com/mnist'
MNIST.resources = [
('/'.join([new_mirror, url.split('/')[-1]]), md5)
for url, md5 in MNIST.resources
]
train = MNIST(root="Datasets/",
download=True,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, *data_aug_trans, transforms.ToTensor(),
transforms.Normalize([0.5], [0.5])])
)
train.data = train.data[:128]
train.targets = train.targets[:128]
test = MNIST(root="Datasets/MNIST",
download=True,
train=False,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, *data_aug_trans, transforms.Grayscale(3),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5])])
)
elif args.dataset == "FashionMNIST":
train = FashionMNIST(root="Datasets/",
download=True,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, transforms.ToTensor()]))
test = FashionMNIST(root="Datasets/",
download=True,
train=False,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, transforms.ToTensor()]))
elif args.dataset == "CelebA":
train = CelebA(root="Datasets/",
download=True,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, transforms.ToTensor()]))
test = CelebA(root="Datasets/",
download=True,
train=False,
transform=transforms.Compose(
[transforms.Resize(args.image_size), *data_aug_trans, transforms.ToTensor()]))
else:
raise NotImplementedError('Unknown dataset')
img, _ = train[1] # take second image
img_shape = img.size()
data_size = len(train)
train_size = int(args.train_valid_split * data_size)
train, validation = random_split(train, [train_size, data_size - train_size],
generator=torch.Generator().manual_seed(41))
train_loader = DataLoader(train,
batch_size=args.batch_size,
shuffle=args.shuffle,
num_workers=args.num_workers
)
valid_loader = DataLoader(validation,
batch_size=args.batch_size,
shuffle=args.shuffle,
num_workers=args.num_workers
)
test_loader = DataLoader(test,
batch_size=args.batch_size,
shuffle=args.shuffle,
num_workers=args.num_workers
)
return train_loader, valid_loader, test_loader, tuple(img_shape)