/
dataset.py
121 lines (101 loc) · 4.74 KB
/
dataset.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
import os
import pickle
from skimage.transform import resize
import numpy as np
import random
from torch.utils.data.dataloader import default_collate
from torchvision import datasets, transforms
class MNIST_Dataset():
def __init__(self, cfg):
self.dataset_dir = cfg.dataset_dir
self.prepro_dir = cfg.prepro_dir
self.num_instances = cfg.num_instances
self.image_height = cfg.image_height
self.image_width = cfg.image_width
self.image_channel_size = cfg.image_channel_size
self.transform = transforms.Compose([transforms.ToTensor(),
transforms.ToPILImage(),
transforms.Grayscale(num_output_channels=1),
transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,)),
])
self.prepro_train_file = os.path.join(self.prepro_dir, str(self.num_instances), 'mnist_train.pickle')
self.prepro_test_file = os.path.join(self.prepro_dir, str(self.num_instances), 'mnist_test.pickle')
if not os.path.exists(self.dataset_dir):
os.makedirs(self.dataset_dir)
self.load_dataset()
if not os.path.exists(self.prepro_dir):
os.makedirs(self.prepro_dir)
if not os.path.exists(os.path.join(self.prepro_dir, str(self.num_instances))):
os.makedirs(os.path.join(self.prepro_dir, str(self.num_instances)))
if not os.path.exists(self.prepro_train_file):
self.preprocess_dataset(train=True)
print('Load train dataset -->', self.prepro_train_file)
with open(self.prepro_train_file, 'rb') as f:
self.train_dataset = pickle.load(f)
else:
print('Load train dataset -->', self.prepro_train_file)
with open(self.prepro_train_file, 'rb') as f:
self.train_dataset = pickle.load(f)
if not os.path.exists(self.prepro_test_file):
self.preprocess_dataset(train=False)
print('Load test dataset -->', self.prepro_test_file)
with open(self.prepro_test_file, 'rb') as f:
self.test_dataset = pickle.load(f)
else:
print('Load test dataset -->', self.prepro_test_file)
with open(self.prepro_test_file, 'rb') as f:
self.test_dataset = pickle.load(f)
def load_dataset(self):
self.raw_train_dataset = datasets.MNIST(root=self.dataset_dir,
train=True,
download=True,
transform=None)
self.raw_test_dataset = datasets.MNIST(root=self.dataset_dir,
train=False,
download=True,
transform=None)
def preprocess_dataset(self, train=True):
if train:
print()
print('Preprocess train dataset')
print()
images = self.raw_train_dataset.data.numpy()
labels = self.raw_train_dataset.targets.numpy()
else:
print()
print('Preprocess test dataset')
print()
images = self.raw_test_dataset.data.numpy()
labels = self.raw_test_dataset.targets.numpy()
_dataset = []
for i, (img, label) in enumerate(zip(images, labels)):
_dataset.append((img, label))
if train:
random.shuffle(_dataset)
dataset = []
instance_idx = 0
for i, (img, label) in enumerate(_dataset):
img = np.expand_dims(img, axis=2)
img = resize(img, (self.image_height, self.image_width), anti_aliasing=True)
img = img.astype(np.float32)
dataset.append((self.transform(img), label, instance_idx))
instance_idx += 1
if self.num_instances <= instance_idx and train:
break
self.max_num_instances = len(dataset)
print('The number of instances: %s' % self.max_num_instances)
if train:
with open(self.prepro_train_file, 'wb') as f:
pickle.dump(dataset, f)
else:
with open(self.prepro_test_file, 'wb') as f:
pickle.dump(dataset, f)
class BatchCollator():
def __init__(self, image_height, image_width, image_channel_size):
self.image_height = image_height
self.image_width = image_width
self.image_channel_size = image_channel_size
def __call__(self, batch):
batch_padded = [b for b in batch]
return default_collate(batch_padded)