-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdatasets.py
212 lines (175 loc) · 10 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
from torchvision.datasets import CIFAR10, CIFAR100, MNIST
from torchvision import transforms
from torchvision.datasets import ImageFolder
from torch.utils.data import DataLoader, RandomSampler
from torch.utils.data import random_split
import copy
import numpy as np
import random
def get_loaders(config, full_package):
print('==> Preparing data..')
train_transform, test_transform = get_transform(config)
# Continue: 6/20/2023 Afternoon
trainset, testset, valset = get_dataset(config,train_transform,test_transform)
# Imbalance loader. config['train']['imbalance_factor'] = 1 means no imbalance
full_package = get_imbalance_loader(config, full_package,trainset)
trainloader = DataLoader(trainset, batch_size=config['train']['batch_size'], shuffle=True, num_workers=config['train']['num_workers'])
valloader = DataLoader(valset, batch_size=config['train']['batch_size'], shuffle=True, num_workers=config['test']['num_workers'])
testloader = DataLoader(testset, batch_size=config['test']['batch_size'], shuffle=False, num_workers=config['test']['num_workers'])
full_package['trainloader'],full_package['valloader'],full_package['testloader'] = trainloader,valloader,testloader
if config['train']['loop_type'] == 'peer':
peer_sampler_x,peer_sampler_y = RandomSampler(trainset, replacement=True),RandomSampler(trainset, replacement=True)
full_package['peerloader_x'] = DataLoader(trainset, batch_size=config['train']['batch_size'], \
shuffle=False, sampler=peer_sampler_x, num_workers=config['train']['num_workers'])
full_package['peerloader_y'] = DataLoader(trainset, batch_size=config['train']['batch_size'], \
shuffle=False, sampler=peer_sampler_y, num_workers=config['train']['num_workers'])
return full_package
def split_train_val(config, trainset):
# Split valset from trainset. 10% of the trainset or 5000 samples
split_train_size = config['train']['train_size']*len(trainset) if config['train']['train_size'] < 1 else config['train']['train_size']
val_size = len(trainset) - split_train_size
split_a, split_b = random_split(range(len(trainset)), [int(split_train_size), int(val_size)])
split_train_idx, val_idx = split_a.indices, split_b.indices
split_trainset, valset = copy.deepcopy(trainset), copy.deepcopy(trainset)
split_trainset.data = split_trainset.data[split_train_idx]
split_trainset.targets = np.array(split_trainset.targets)[split_train_idx]
valset.data = valset.data[val_idx]
valset.targets = np.array(valset.targets)[val_idx]
if not config['train']['train_on_full']: trainset = split_trainset
if val_size > 20000: print('Warning: val_size is larger than 20000')
return trainset, valset
# data_set is subset of trainset
def get_imbalance_loader(config, full_package,data_set):
y_clean = np.array(data_set.targets)
samples_per_cls = get_cls_num(y_clean, config['train']['imbalance_factor'],
cls_num=config['data']['num_classes'])
selected_idx = get_cls_idx(num_per_cls=samples_per_cls, y_train=y_clean,
is_upsampling=config['train']['imbalance_is_upsampling'])
data_set.data = data_set.data[selected_idx] # Update data / select only the selected indices
data_set.targets = np.array(data_set.targets)[selected_idx] # Update targets / select only the selected indices
full_package['imbalance_info'] = {'samples_per_cls': samples_per_cls, 'selected_idx': selected_idx}
return full_package
# def get_dataset(name, root, train=True, transform=None, download=True):
def get_dataset(config, train_transform, test_transform, download=True):
name = config['data']['dataset_name']
train_root, test_root = config['data']['train_dir'], config['data']['test_dir']
if name.lower() == 'cifar10':
trainset = CIFAR10(root=train_root, train=True, transform=train_transform, download=download)
testset = CIFAR10(root=test_root, train=False, transform=test_transform, download=download)
elif name.lower() == 'cifar100':
trainset = CIFAR100(root=train_root, train=True, transform=train_transform, download=download)
testset = CIFAR100(root=test_root, train=False, transform=test_transform, download=download)
elif name.lower() == 'mnist':
trainset = MNIST(root=train_root, train=True, transform=train_transform, download=download)
testset = MNIST(root=test_root, train=False, transform=test_transform, download=download)
elif name.lower() == 'clothing1mpp':
from loaders.clothing1mpp import Clothing1mPP
path_dict ={
'root': config['data']['root'], # Root dir of the dataset that includes folders of images: Jacket, Shirt, etc.
'full_label_json_path': config['data']['full_label_json_path'], # Json file that includes all the labels
'split_path':config['data']['split_path'] # Dir includes to test_ids.pt, train_ids.pt, val_ids.pt
}
image_size = config['data']['image_size']
label_types = config['data']['label_types']
debug = config['train']['debug']
# Load datasets
trainset = Clothing1mPP(path_dict, image_size,debug=debug,split='train',label_types=label_types,transform=train_transform)
pre_load = trainset.pre_load
valset = Clothing1mPP(path_dict, image_size,debug=debug,split='val',label_types=label_types,pre_load=pre_load,transform=test_transform)
testset = Clothing1mPP(path_dict, image_size,debug=debug,split='test',label_types=label_types,pre_load=pre_load,transform=test_transform)
return trainset, testset, valset
elif name.lower() == 'custom':
# Assuming your images are in a folder structure like:
# root/dog/xxx.png
# root/cat/yyy.png
# ...
# Where 'root' is specified in your yaml file.
return ImageFolder(root=root, transform=transform)
else:
raise ValueError(f'Dataset {name} not recognized. Choose from ["cifar10", "cifar100", "mnist"]')
trainset, valset = split_train_val(config, trainset)
return trainset, testset, valset
def get_transform(config):
name = config['data']['dataset_name']
image_size = config['data']['image_size']
if name.lower() == 'cifar10':
if image_size!=32: print('Warning: cifar10 image_size is not 32')
train_transform = transforms.Compose([
transforms.RandomCrop(image_size, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
test_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
elif name.lower() == 'cifar100':
if image_size!=32: print('Warning: cifar10 image_size is not 32')
train_transform = transforms.Compose([
transforms.RandomCrop(image_size, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5071, 0.4867, 0.4408), (0.2675, 0.2565, 0.2761)),
])
test_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.5071, 0.4867, 0.4408), (0.2675, 0.2565, 0.2761)),
])
else:
train_transform = transforms.Compose([
transforms.Resize((image_size, image_size)), # Resize the image to the desired crop size
transforms.RandomCrop(image_size, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.6959, 0.6537, 0.6371), (0.3113, 0.3192, 0.3214))
])
test_transform = transforms.Compose([
transforms.Resize((image_size, image_size)), # Resize the image to the desired crop size
transforms.ToTensor(),
transforms.Normalize((0.6959, 0.6537, 0.6371), (0.3113, 0.3192, 0.3214))
])
return train_transform, test_transform
def get_cls_num(y_train, imb_factor=None, cls_num=10):
"""Get a list of image numbers for each class.
Given cifar version Num of imgs follows emponential distribution
img max: 5000 / 500 * e^(-imb * 0);
img min: 5000 / 500 * e^(-imb * int(dataset_name - 1))
exp(-imb * (int(dataset_name) - 1)) = img_max / img_min
Args:
dataset_name: str, 'cifar10', 'cifar100
y_train: the training label
imb_factor: float, imbalance factor: img_min/img_max, None if geting
default cifar data number
Returns:
img_num_per_cls: a list of number of images per class
"""
img_max = len(list(np.where(np.array(y_train) == 0)[0]))
if imb_factor is None:
return [img_max] * cls_num
img_num_per_cls = []
for cls_idx in range(cls_num):
idx_this_class = list(np.where(np.array(y_train) == cls_idx)[0])
num = img_max * (imb_factor**(cls_idx / (cls_num - 1.0)))
img_num_per_cls.append(min(int(num), len(idx_this_class)))
return img_num_per_cls
def get_cls_idx(num_per_cls, y_train, is_upsampling=False):
"""Get the seleted index for the training images.
Given number of selected images per class, return the image indexes adopted.
Args:
num_per_cls: a list of number of images per class
y_train: the clean label for class selection
is_upsampling: whether up sampling to strick a balance
Returns:
selected_idx: a list of selected images for training use
"""
cls_num = len(num_per_cls)
selected_idx = []
for cls_idx in range(cls_num):
idx_this_class = list(np.where(np.array(y_train) == cls_idx)[0])
indices = random.sample(idx_this_class, num_per_cls[cls_idx])
if is_upsampling:
up_samling = random.choices(indices, k=int(50000/cls_num)-len(indices))
selected_idx.extend(up_samling)
selected_idx.extend(indices)
return selected_idx