resnet/data.py at master · ilango100/resnet · GitHub

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
import math
import tensorflow as tf
import tensorflow_datasets as tfds


def split_trainval(dataset, l, frac=0.8):
    trsteps = math.ceil(frac*l)

    val = dataset.skip(trsteps)
    train = dataset.take(trsteps)
    return train, val, trsteps


def std(x, y):
    return x/255, y


def cifar10(bsize):
    # Download and prepare dataset
    cifar = tfds.builder("cifar10")
    cifar.download_and_prepare()

    # Get epoch steps
    totsteps = cifar.info.splits[tfds.Split.TRAIN].num_examples
    trsteps = math.ceil(totsteps * 0.8)
    valsteps = totsteps - trsteps
    testeps = cifar.info.splits[tfds.Split.TEST].num_examples

    # Get data
    # train, val = tfds.Split.TRAIN.subsplit(weighted=[4, 1])
    splits = ("train[:80%]", "train[80%:]", "test")
    train, val, test = [cifar.as_dataset(split=x,
                                         shuffle_files=True,
                                         batch_size=bsize,
                                         as_supervised=True)
                        for x in splits]

    def augment(x, y):
        x = tf.pad(x, [[0, 0], [4, 4], [4, 4], [0, 0]])
        x = tf.image.random_crop(x, [bsize, 32, 32, 3])
        x = tf.image.random_flip_left_right(x)
        return x, y

    # Apply operations
    train = train \
        .map(augment, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
        .map(std, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
        .prefetch(tf.data.experimental.AUTOTUNE).repeat()
    val = val \
        .map(std, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
        .prefetch(tf.data.experimental.AUTOTUNE).repeat()
    test = test \
        .map(std, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
        .prefetch(tf.data.experimental.AUTOTUNE).repeat()

    # Epoch steps
    trsteps = math.ceil(trsteps/bsize)
    valsteps = math.ceil(valsteps/bsize)
    testeps = math.ceil(testeps/bsize)

    return train, val, test, trsteps, valsteps, testeps, 10


def cats_vs_dogs(bsize):
    cvd = tfds.builder("cats_vs_dogs")
    cvd.download_and_prepare()

    totsteps = cvd.info.splits[tfds.Split.TRAIN].num_examples
    trsteps = math.ceil(totsteps*0.7)
    valsteps = math.ceil(totsteps*0.15)
    testeps = totsteps - trsteps - valsteps

    train, val, test = "train[:70%]", "train[70%:85%]", "train[85%:]"
    train, val, test = [cvd.as_dataset(split=x,
                                       shuffle_files=True,
                                       batch_size=bsize,
                                       as_supervised=True)
                        for x in [train, val, test]]

    # Per epoch steps
    trsteps = math.ceil(trsteps/bsize)
    valsteps = math.ceil(valsteps/bsize)
    testeps = math.ceil(testeps/bsize)

    return train, val, test, trsteps, valsteps, testeps, 2


def imagenette(bsize):
    imt = tfds.builder("imagenette/160px")
    imt.download_and_prepare()

    totsteps = imt.info.splits[tfds.Split.TRAIN].num_examples
    trsteps = math.ceil(totsteps * 0.8)
    valsteps = totsteps - trsteps
    testeps = imt.info.splits[tfds.Split.VALIDATION].num_examples

    train, val, test = "train[:80%]", "train[80%:]", "validation"
    train, val, test = [imt.as_dataset(split=x,
                                       shuffle_files=True,
                                       batch_size=bsize,
                                       as_supervised=True)
                        for x in [train, val, test]]

    def augment(x, y):
        x = tf.pad(x, [[0, 0], [20, 20], [20, 20], [0, 0]])
        x = tf.image.random_crop(x, (160, 160))
        x = tf.image.random_flip_left_right(x)
        x = tf.image.random_brightness(x, 0.2)
        return x, y

    train = train \
        .map(augment, num_parallel_calls=tf.data.experimental.AUTOTUNE) \
        .map(std, num_parallel_calls=tf.data.experimental.AUTOTUNE)
    val = val.map(std, num_parallel_calls=tf.data.experimental.AUTOTUNE)
    test = test.map(std, num_parallel_calls=tf.data.experimental.AUTOTUNE)

    trsteps, valsteps, testeps = [
        math.ceil(x/bsize) for x in [trsteps, valsteps, testeps]]

    return train, val, test, trsteps, valsteps, testeps, 10


# tiny_imagenet not implemented yet
# def tiny_imagenet(bsize):
#     # Download and prepare dataset
#     cifar = tfds.builder("tiny_imagenet")
#     cifar.download_and_prepare()

#     # Get epoch steps
#     totsteps = cifar.info.splits[tfds.Split.TRAIN].num_examples
#     trsteps = math.ceil(totsteps * 0.8)
#     valsteps = totsteps - trsteps
#     testeps = cifar.info.splits[tfds.Split.TEST].num_examples

#     # Get data
#     train = cifar.as_dataset(split="train[:80%]", as_supervised=True)
#     val = cifar.as_dataset(split="train[80%:]", as_supervised=True)
#     test = cifar.as_dataset(split="validation", as_supervised=True)

#     # Apply operations
#     train = train.repeat().batch(bsize).prefetch(tf.data.experimental.AUTOTUNE)
#     val = val.repeat().batch(bsize).prefetch(tf.data.experimental.AUTOTUNE)
#     test = test.repeat().batch(bsize).prefetch(tf.data.experimental.AUTOTUNE)

#     # Epoch steps
#     trsteps = math.ceil(trsteps/bsize)
#     valsteps = math.ceil(valsteps/bsize)
#     testeps = math.ceil(testeps/bsize)

#     return train, val, test, trsteps, valsteps, testeps, 200


__all__ = ["cifar10", "cats_vs_dogs", "imagenette"]