-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathdatasets.py
169 lines (142 loc) · 7.1 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
import torch
from torch.utils.data import DataLoader, Dataset
from torchvision import transforms
from torchvision.datasets import MNIST, EMNIST, FashionMNIST, CIFAR10, KMNIST
import numpy as np
import os
from PIL import Image
class ImageNet(Dataset):
def __init__(self, path, transform=None):
"""
Args:
csv_file (string): Path to the csv file with annotations.
root_dir (string): Directory with all the images.
transform (callable, optional): Optional transform to be applied
on a sample.
"""
self.path = path
self.data = os.listdir(path)
self.transform = transform
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
if torch.is_tensor(idx):
idx = idx.tolist()
sample = np.load(f'{self.path}/{self.data[idx]}')
image, label = sample['image'], sample['label']
if self.transform:
image = self.transform(image)
return image, label
class CelebAHQ(Dataset):
def __init__(self, path, transform=None):
"""
Args:
csv_file (string): Path to the csv file with annotations.
root_dir (string): Directory with all the images.
transform (callable, optional): Optional transform to be applied
on a sample.
"""
self.path = path
self.data = os.listdir(path)
self.transform = transform
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
if torch.is_tensor(idx):
idx = idx.tolist()
image = np.load(f'{self.path}/{self.data[idx]}')['image']
image = Image.fromarray(image)
if self.transform:
image = self.transform(image)
return image, torch.zeros(1)
def get_transform(dataset: str = 'mnist'):
if dataset == 'cifar10' or dataset == 'celebahq':
img_transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.ToTensor()
])
else:
img_transform = transforms.Compose([
transforms.ToTensor()
])
return img_transform
def get_list_of_datasets():
lst = ['mnist', 'kmnist', 'emnist', 'fashionmnist', 'cifar10', 'celebahq', 'celebahq64']
return lst
def get_train_val_dataloaders(dataset: str = 'mnist', batch_size: int = 128, p_validation: float = 0.1, seed: int = 3,
return_img_dim: bool = True, return_alpha: bool = True, data_dir=""):
'''
Returns a train dataloader and a validation dataloader if p_validation > 0.
:param dataset: dataset to retrieve. Either "mnist", "emnist" or "fashionmnist".
:param batch_size: batch size for each dataloader.
:param p_validation: percentage of original training set to take for validation
:param seed: manual seed for reproducibility
:param return_img_dim: whether to return the img dimensions
:param return_alpha: whether to return the alpha corresponding to a dataset (for conversion to logit space)
:return: train_dataloader, val_dataloader (if p_validation > 0), img_dim (if return_img_dim), alpha (if return_alpha)
'''
assert (p_validation >= 0) and (p_validation <= 1)
img_transform = get_transform(dataset)
torch_rng = torch.Generator().manual_seed(seed)
if dataset.lower() == 'mnist':
train_dataset = MNIST(root='./data/MNIST', download=True, train=True, transform=img_transform)
alpha = 1e-6
elif dataset.lower() == 'kmnist':
train_dataset = KMNIST(root='./data/KMNIST', download=True, train=True, transform=img_transform)
alpha = 1e-6
elif dataset.lower() == 'emnist':
train_dataset = EMNIST(root='./data/EMNIST', split='letters', download=True, train=True, transform=img_transform)
alpha = 1e-6
elif dataset.lower() == 'fashionmnist':
train_dataset = FashionMNIST(root='./data/FashionMNIST', download=True, train=True, transform=img_transform)
alpha = 1e-6
elif dataset.lower() == 'cifar10' or dataset.lower() == 'cifar':
train_dataset = CIFAR10(root='./data/CIFAR', download=True, train=True, transform=img_transform)
alpha = .05
elif dataset.lower() == 'imagenet':
train_dataset=ImageNet(f'{data_dir}/train', transform=img_transform)
alpha = .05
elif dataset.lower() == 'celebahq' or dataset.lower() == 'celebahq64':
train_dataset = CelebAHQ(f'{data_dir}/train', transform=img_transform)
valid_dataset = CelebAHQ(f'{data_dir}/valid', transform=img_transform)
alpha = .05
if dataset.lower() == 'celebahq' or dataset.lower() == 'celebahq64':
train_dataloader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
validation_dataloader = DataLoader(valid_dataset, batch_size=batch_size, shuffle=False)
return_tuple = [train_dataloader, validation_dataloader]
elif p_validation > 0:
size_validation = round(p_validation * len(train_dataset))
size_train = len(train_dataset) - size_validation
train_subset, val_subset = torch.utils.data.random_split(train_dataset, [size_train, size_validation],
generator=torch_rng)
train_dataloader = DataLoader(train_subset, batch_size=batch_size, shuffle=True, generator=torch_rng)
validation_dataloader = DataLoader(val_subset, batch_size=batch_size, shuffle=False, generator=torch_rng)
return_tuple = [train_dataloader, validation_dataloader]
else:
return_tuple = [DataLoader(train_dataset, batch_size=batch_size, shuffle=True, generator=torch_rng)]
if return_img_dim:
img_dim = list(train_dataset[0][0].shape)
return_tuple.append(img_dim)
if return_alpha:
return_tuple.append(alpha)
return return_tuple
def get_test_dataloader(dataset: str = 'mnist', batch_size: int = 128, shuffle=False, data_dir=""):
img_transform = transforms.Compose([
transforms.ToTensor()
])
if dataset.lower() == 'mnist':
test_dataset = MNIST(root='./data/MNIST', download=True, train=False, transform=img_transform)
elif dataset.lower() == 'kmnist':
test_dataset = KMNIST(root='./data/KMNIST', download=True, train=False, transform=img_transform)
elif dataset.lower() == 'emnist':
test_dataset = EMNIST(root='./data/EMNIST', split='letters', download=True, train=False, transform=img_transform)
elif dataset.lower() == 'fashionmnist':
test_dataset = FashionMNIST(root='./data/FashionMNIST', download=True, train=False, transform=img_transform)
elif dataset.lower() == 'cifar10' or dataset.lower() == 'cifar':
test_dataset = CIFAR10(root='./data/CIFAR', download=True, train=False, transform=img_transform)
elif dataset.lower() == 'imagenet':
test_dataset = ImageNet(f'{data_dir}/test', transform=img_transform)
elif dataset.lower() == 'celebahq' or dataset.lower() == 'celebahq64':
test_dataset = CelebAHQ(f'{data_dir}/test', transform=img_transform)
test_dataloader = DataLoader(test_dataset, batch_size=batch_size, shuffle=shuffle)
return test_dataloader