-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataloader.py
168 lines (135 loc) · 6.32 KB
/
dataloader.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
import os
import matplotlib.pyplot as plt
import numpy as np
from random import randint
from transform import build_transform
from torch.utils.data import DataLoader, TensorDataset, Dataset
from torchvision import transforms as T
from torchvision.utils import make_grid
from torchvision import datasets
import torch
DATA_DIR = './tiny-imagenet-200' # Original images come in shapes of [3,64,64]
# Functions to display single or a batch of sample images
def imshow(img):
npimg = img.numpy()
plt.imshow(np.transpose(npimg, (1, 2, 0)))
plt.show()
def show_batch(dataloader):
for batch in dataloader:
images, labels = batch
imshow(make_grid(images)) # Using Torchvision.utils make_grid function
break
def show_image(dataloader):
dataiter = iter(dataloader)
images, labels = dataiter.next()
random_num = randint(0, len(images)-1)
imshow(images[random_num])
label = labels[random_num]
print(f'Label: {label}, Shape: {images[random_num].shape}')
# Setup function to create dataloaders for image datasets
def generate_dataloader(data, name, batch_size, transform=None, use_cuda=True):
if data is None:
return None
# Read image files to pytorch dataset using ImageFolder, a generic data
# loader where images are in format root/label/filename
# See https://pytorch.org/vision/stable/datasets.html
if transform is None:
dataset = datasets.ImageFolder(data, transform=T.ToTensor())
else:
dataset = datasets.ImageFolder(data, transform=transform)
# Set options for device
if use_cuda:
kwargs = {"pin_memory": True, "num_workers": 1}
else:
kwargs = {}
g = torch.Generator()
g.manual_seed(42)
# Wrap image dataset (defined above) in dataloader
dataloader = DataLoader(dataset, batch_size=batch_size,
shuffle=(name=="train"), generator=g,
**kwargs)
return dataloader
def get_pretrain_transform(imgsz):
return T.Compose([
# T.Resize(256), # Resize images to 256 x 256
T.RandomResizedCrop(size=(imgsz, imgsz), scale=(0.2, 1.0), interpolation=3),
T.RandomHorizontalFlip(),
T.ToTensor(), # Converting cropped images to tensors
T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
# DEIT-3 transforms
def get_pretrain_transform_deit(imgsz):
primary_tfl = [
T.Resize(imgsz, interpolation=3),
T.RandomCrop(imgsz, padding=4,padding_mode='reflect'),
T.RandomHorizontalFlip()
]
secondary_tfl = [T.RandomChoice([T.Grayscale(num_output_channels=3),
T.RandomSolarize(0,p=1.0),
T.GaussianBlur(3,sigma=(0.1*(1/3.5),2*(1/3.5)))])]
return T.Compose(primary_tfl+secondary_tfl+[
T.ToTensor(), # Converting cropped images to tensors
T.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
def get_finetune_transform(is_train, imgsz, args):
return build_transform(is_train, imgsz, args)
def get_finetune_dataloaders(datadir, batch_size, imgsz, args, use_cuda=True):
# Define training and validation data paths
train_dir = os.path.join(datadir, 'train')
valid_dir = os.path.join(datadir, 'val')
train_transform = get_finetune_transform(is_train=True, imgsz=imgsz, args=args)
valid_transfrom = get_finetune_transform(is_train=False, imgsz=imgsz, args=args)
fp = open(os.path.join(valid_dir, 'val_annotations.txt'), 'r')
data = fp.readlines()
# Create dictionary to store img filename (word 0) and corresponding
# label (word 1) for every line in the txt file (as key value pair)
val_img_dict = {}
for line in data:
words = line.split('\t')
val_img_dict[words[0]] = words[1]
fp.close()
val_img_dir = os.path.join(valid_dir, 'images')
for img, folder in val_img_dict.items():
newpath = (os.path.join(val_img_dir, folder))
if not os.path.exists(newpath):
os.makedirs(newpath)
if os.path.exists(os.path.join(val_img_dir, img)):
os.rename(os.path.join(val_img_dir, img), os.path.join(newpath, img))
# Create DataLoaders for pre-trained models (normalized based on specific requirements)
train_loader_finetune = generate_dataloader(train_dir, "train", batch_size=batch_size,
transform=train_transform, use_cuda=use_cuda)
val_loader_finetune = generate_dataloader(val_img_dir, "val", batch_size=batch_size,
transform=valid_transfrom, use_cuda=use_cuda)
return train_loader_finetune, val_loader_finetune
def get_pretrain_dataloaders(datadir, batch_size, imgsz=64, use_cuda=True, deit=False):
# Define training and validation data paths
train_dir = os.path.join(datadir, 'train')
valid_dir = os.path.join(datadir, 'val')
if deit:
transform = get_pretrain_transform_deit(imgsz)
else:
transform = get_pretrain_transform(imgsz)
fp = open(os.path.join(valid_dir, 'val_annotations.txt'), 'r')
data = fp.readlines()
# Create dictionary to store img filename (word 0) and corresponding
# label (word 1) for every line in the txt file (as key value pair)
val_img_dict = {}
for line in data:
words = line.split('\t')
val_img_dict[words[0]] = words[1]
fp.close()
val_img_dir = os.path.join(valid_dir, 'images')
for img, folder in val_img_dict.items():
newpath = (os.path.join(val_img_dir, folder))
if not os.path.exists(newpath):
os.makedirs(newpath)
if os.path.exists(os.path.join(val_img_dir, img)):
os.rename(os.path.join(val_img_dir, img), os.path.join(newpath, img))
# Create DataLoaders for pre-trained models (normalized based on specific requirements)
train_loader_pretrain = generate_dataloader(train_dir, "train", batch_size=batch_size,
transform=transform, use_cuda=use_cuda)
val_loader_pretrain = generate_dataloader(val_img_dir, "val", batch_size=batch_size,
transform=transform, use_cuda=use_cuda)
return train_loader_pretrain, val_loader_pretrain