-
Notifications
You must be signed in to change notification settings - Fork 2
/
dataloader_red.py
171 lines (155 loc) · 7.34 KB
/
dataloader_red.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
from torch.utils.data import Dataset, DataLoader
import torchvision.transforms as transforms
import random
import numpy as np
import pandas as pd
from PIL import Image
import json
import os
import torch
import tools
from torchnet.meter import AUCMeter
def unpickle(file):
import _pickle as cPickle
with open(file, 'rb') as fo:
dict = cPickle.load(fo, encoding='latin1')
return dict
class red_dataset(Dataset):
def __init__(self, root_dir, transform, r,color = 'red', pred = [], probability = [], mode = 'train'):
self.root = root_dir + 'mini-imagenet/'
self.transform = transform
self.mode = mode
pred = pred
num_class = 100
noise_rate = r
self.probability = probability
if self.mode=='test':
with open(self.root+'split/clean_validation') as f:
lines=f.readlines()
self.val_imgs = []
self.val_labels = {}
for line in lines:
img, target = line.split()
target = int(target)
img_path = 'validation/'+str(target) + '/'+img
self.val_imgs.append(img_path)
self.val_labels[img_path]=target
else:
noise_file = '{}_noise_nl_{}'.format(color,noise_rate)
with open(self.root+'split/'+noise_file) as f:
lines=f.readlines()
train_imgs = []
self.train_labels = {}
for line in lines:
img, target = line.split()
target = int(target)
train_path = 'all_images/'
train_imgs.append(train_path + img)
self.train_labels[train_path + img]=target
if (self.mode == 'all') or (self.mode == 'neighbor') or (self.mode=='pretext'):
self.train_imgs = train_imgs
else:
if self.mode == "labeled":
pred_idx = pred.nonzero()[0]
self.train_imgs = [train_imgs[i] for i in pred_idx]
self.probability = [self.probability[i] for i in pred_idx]
print("%s data has a size of %d"%(self.mode,len(self.train_imgs)))
elif self.mode == "unlabeled":
pred_idx = (1-pred).nonzero()[0]
self.train_imgs = [train_imgs[i] for i in pred_idx]
print("%s data has a size of %d"%(self.mode,len(self.train_imgs)))
def __getitem__(self, index):
if self.mode=='labeled':
img_path = self.train_imgs[index]
target = self.train_labels[img_path]
prob = self.probability[index]
image = Image.open(self.root+img_path).convert('RGB')
img1 = self.transform(image)
img2 = self.transform(image)
return img1, img2, target, prob
elif self.mode=='unlabeled':
img_path = self.train_imgs[index]
image = Image.open(self.root+img_path).convert('RGB')
img1 = self.transform(image)
img2 = self.transform(image)
return img1, img2
elif self.mode=='all':
img_path = self.train_imgs[index]
target = self.train_labels[img_path]
img = Image.open(self.root+img_path).convert('RGB')
if self.transform is not None:
img = self.transform(img)
return img, target, index
elif self.mode=='test':
img_path = self.val_imgs[index]
target = self.val_labels[img_path]
img = Image.open(self.root+img_path).convert('RGB')
if self.transform is not None:
img = self.transform(img)
return img, target
def __len__(self):
if self.mode!='test':
return len(self.train_imgs)
else:
return len(self.val_imgs)
class red_dataloader():
def __init__(self, dataset, r, noise_mode, batch_size, num_workers, root_dir, log, noise_file=''):
self.dataset = dataset
self.r = r
self.noise_mode = noise_mode
self.batch_size = batch_size
self.num_workers = num_workers
self.root_dir = root_dir
self.log = log
self.noise_file = noise_file
self.transform_train = transforms.Compose([
transforms.Resize((32,32)),
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
#transforms.Normalize((0.4914, 0.4822, 0.4465),(0.2023, 0.1994, 0.2010)),
])
self.transform_test = transforms.Compose([
transforms.Resize((32,32)),
transforms.ToTensor(),
#transforms.Normalize((0.4914, 0.4822, 0.4465),(0.2023, 0.1994, 0.2010)),
])
def run(self,mode,pred=[],prob=[]):
if mode=='warmup':
all_dataset = red_dataset(r=self.r, root_dir=self.root_dir, transform=self.transform_train, mode="all")
trainloader = DataLoader(
dataset=all_dataset,
batch_size=self.batch_size*2,
shuffle=True,
num_workers=self.num_workers, drop_last = True)
return trainloader
elif mode=='train':
labeled_dataset = red_dataset(r=self.r, root_dir=self.root_dir, transform=self.transform_train, mode="labeled", pred=pred, probability=prob)
labeled_trainloader = DataLoader(
dataset=labeled_dataset,
batch_size=self.batch_size,
shuffle=True,
num_workers=self.num_workers, drop_last = True)
unlabeled_dataset = red_dataset(r=self.r, root_dir=self.root_dir, transform=self.transform_train, mode="unlabeled", pred=pred)
unlabeled_trainloader = DataLoader(
dataset=unlabeled_dataset,
batch_size=self.batch_size,
shuffle=True,
num_workers=self.num_workers, drop_last = True)
return labeled_trainloader, unlabeled_trainloader
elif mode=='test':
test_dataset = red_dataset(r=self.r, root_dir=self.root_dir, transform=self.transform_test, mode='test')
test_loader = DataLoader(
dataset=test_dataset,
batch_size=self.batch_size,
shuffle=False,
num_workers=self.num_workers, drop_last = True)
return test_loader
elif mode=='eval_train':
eval_dataset = red_dataset(r=self.r, root_dir=self.root_dir, transform=self.transform_test, mode='all')
eval_loader = DataLoader(
dataset=eval_dataset,
batch_size=self.batch_size,
shuffle=False,
num_workers=self.num_workers, drop_last = True)
return eval_loader