-
Notifications
You must be signed in to change notification settings - Fork 2
/
data_generator.py
136 lines (125 loc) · 6.84 KB
/
data_generator.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
""" Code for loading data. """
import numpy as np
import os
import random
import tensorflow as tf
import pdb
from tensorflow.python.platform import flags
from utils import get_images
from autoaugment import CIFAR10Policy
FLAGS = flags.FLAGS
class DataGenerator(object):
def __init__(self, batch_size):
self.batch_size = batch_size
self.img_size_cropped = 32
self.img_size = (self.img_size_cropped,self.img_size_cropped)
self.dim_input = np.prod(self.img_size)*3
if FLAGS.dataset == 'cifar10':
self.num_classes = 10
elif FLAGS.dataset == 'cinic10':
self.num_classes = 10
elif FLAGS.dataset == 'cifar100':
self.num_classes = 100
elif FLAGS.dataset == 'tiny':
self.num_classes = 200
elif FLAGS.dataset == 'mnist+svhn' or FLAGS.dataset == 'mnistm' or FLAGS.dataset == 'mnist+svhn+usps+mnistm':
self.num_classes = 10
elif FLAGS.dataset == 'pacs':
self.num_classes = 7
self.dim_output = self.num_classes
if FLAGS.dataset == 'cifar10':
self.metatrain_folder = os.path.join(FLAGS.data_dir, 'cifar', 'train_images')
if FLAGS.test_set:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'cifar', 'test_images')
else:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'cifar', 'val_images')
elif FLAGS.dataset == 'cifar100':
self.metatrain_folder = os.path.join(FLAGS.data_dir, 'cifar100', 'train_images')
if FLAGS.test_set:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'cifar100', 'test_images')
else:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'cifar100', 'val_images')
elif FLAGS.dataset == 'cinic10':
self.metatrain_folder = os.path.join(FLAGS.data_dir, 'cinic10', 'train_images')
if FLAGS.test_set:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'cinic10', 'test_images')
else:
self.metaval_folder = os.path.join(FLAGS.data_dir,'cinic10', 'val_images')
elif FLAGS.dataset == 'tiny':
self.metatrain_folder = os.path.join(FLAGS.data_dir, 'tiny-imagenet-200', 'train_images')
self.metaval_folder = os.path.join(FLAGS.data_dir, 'tiny-imagenet-200', 'val_images')
elif FLAGS.dataset == 'pacs':
self.metatrain_folder = os.path.join(FLAGS.data_dir, 'PACS', 'train')
if FLAGS.test_set:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'PACS', 'test')
else:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'PACS', 'val')
elif FLAGS.dataset == 'mnistm':
self.metatrain_folder = os.path.join(FLAGS.data_dir, 'mnistm', 'train')
if FLAGS.test_set:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'mnistm', 'test')
else:
self.metaval_folder = os.path.join(FLAGS.data_dir, 'mnistm', 'val')
def make_data_tensor(self, train=True):
def read_images_from_disk(input_queue):
"""Consumes a single filename and label as a ' '-delimited string.
Args:
filename_and_label_tensor: A scalar string tensor.
Returns:
Two tensors: the decoded image, and the string label.
"""
label = input_queue[1]
file_contents = tf.read_file(input_queue[0])
example = tf.image.decode_png(file_contents, channels=3)
#example = tf.cast(example, tf.float32) / 255.0
return example, label
def auto_aug(tensor):
policy = CIFAR10Policy()
return tf.py_func(policy, [tensor], tf.uint8)
def cut_out(array):
cut_size = 16
x = np.random.randint(self.img_size_cropped)
y = np.random.randint(self.img_size_cropped)
x1 = np.clip(x - cut_size // 2, 0, self.img_size_cropped)
x2 = np.clip(x + cut_size // 2, 0, self.img_size_cropped)
y1 = np.clip(y - cut_size // 2, 0, self.img_size_cropped)
y2 = np.clip(y + cut_size // 2, 0, self.img_size_cropped)
array[x1:x2,y1:y2, :] = 0
return array
def preprocess_image(image, train):
if train:
if FLAGS.dataset == 'tiny' or FLAGS.dataset == 'pacs' or FLAGS.dataset == 'mnistm' or FLAGS.dataset == 'mnist+svhn+usps+mnistm':
image = tf.image.resize_images(image, [self.img_size_cropped, self.img_size_cropped], method=0)
image = tf.image.resize_image_with_crop_or_pad(image, target_height=self.img_size_cropped+8, target_width=self.img_size_cropped+8)
image = tf.random_crop(image, size=[self.img_size_cropped, self.img_size_cropped, 3])
image = tf.image.random_flip_left_right(image)
image = auto_aug(image)
image.set_shape([self.img_size_cropped, self.img_size_cropped, 3])
image = tf.py_func(cut_out, [image], tf.uint8)
image.set_shape([self.img_size_cropped, self.img_size_cropped, 3])
image = tf.cast(image, tf.float32) / 255.0
else:
if FLAGS.dataset == 'tiny' or FLAGS.dataset == 'pacs' or FLAGS.dataset == 'mnistm' or FLAGS.dataset == 'mnist+svhn+usps+mnistm':
image = tf.image.resize_images(image, [self.img_size_cropped, self.img_size_cropped], method=0)
image = tf.image.resize_image_with_crop_or_pad(image,
target_height=self.img_size_cropped,
target_width=self.img_size_cropped)
image = tf.cast(image, tf.float32) / 255.0
return image
folders = self.metatrain_folder if train else self.metaval_folder
print("Generating filenames")
filenames_labels = [[os.path.join(folders, each),int(each.split('_')[0])] for each in os.listdir(folders)]
image_list = [l[0] for l in filenames_labels]
label_list = [l[1] for l in filenames_labels]
images = tf.convert_to_tensor(image_list, dtype=tf.string)
labels = tf.convert_to_tensor(label_list, dtype=tf.int32)
# Makes an input queue
input_queue = tf.train.slice_input_producer([images, labels],
shuffle=True)
image, label = read_images_from_disk(input_queue)
image = preprocess_image(image, train=train)
label = tf.one_hot(label, self.num_classes)
min_queue_examples = 256
parallel_threads = 1 if FLAGS.test_set else 16
image_batch, label_batch = tf.train.batch([image,label], batch_size=self.batch_size, num_threads=parallel_threads, capacity=min_queue_examples+3*self.batch_size)
return image_batch, label_batch