DenseNet-tf2/data_loader.py at master · vmelan/DenseNet-tf2 · 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
import pickle
import numpy as np
import tensorflow as tf

import pdb

class DataLoader():
    """ Load CIFAR dataset """
    def __init__(self, config):
    	# Load config file
        self.config = config

    def unpickle(self, file):
    	with open(file, 'rb') as fo:
    		dict = pickle.load(fo, encoding='bytes')
    	return dict

    def load_cifar10(self, data_path):
        # from https://luckydanny.blogspot.com/2016/07/load-cifar-10-dataset-in-python3.html
        train_data = None
        train_labels = []
        test_data = None
        test_labels = None

        for i in range(1, 6):
        	data_dict = self.unpickle(data_path + "data_batch_" + str(i))
        	if (i == 1):
        		train_data = data_dict[b'data']
        	else:
        		train_data = np.vstack((train_data, data_dict[b'data']))
        	train_labels += data_dict[b'labels']

        test_data_dict = self.unpickle(data_path + "test_batch")
        test_data = test_data_dict[b'data']
        test_labels = test_data_dict[b'labels']

        train_data = train_data.reshape((50000, 3, 32, 32))
        train_data = np.rollaxis(train_data, 1, 4)
        train_labels = np.array(train_labels)

        test_data = test_data.reshape((10000, 3, 32, 32))
        test_data = np.rollaxis(test_data, 1, 4)
        test_labels = np.array(test_labels)

        return train_data, train_labels, test_data, test_labels

    def prepare_data(self):
        """ Prepare data by using TensorFlow data API"""
        X_train, y_train, X_test, y_test = self.load_cifar10(self.config["input"]["data_path"])
        self.len_data = X_train.shape[0]

        train_dataset = tf.data.Dataset.from_tensor_slices((X_train, y_train))
        test_dataset = tf.data.Dataset.from_tensor_slices((X_test, y_test))

        train_dataset = train_dataset.map(self.preprocess)
        test_dataset = test_dataset.map(self.preprocess)

        batch_size = self.config["trainer"]["batch_size"]
        buffer_size = self.config["trainer"]["buffer_size"]
        train_loader = train_dataset.shuffle(buffer_size).batch(batch_size)
        test_loader = test_dataset.shuffle(buffer_size).batch(batch_size)

        return train_loader, test_loader


    def preprocess(self, image, label):
        # cast image to float32 type
        image = tf.cast(image, tf.float32)

        # resize images
        image = tf.image.resize(image, (self.config["input"]["size"][0], self.config["input"]["size"][1]))
        # normalize according to training data stats
        image = (image - self.config["input"]["mean"]) / self.config["input"]["std"]
        # data augmentation
        image = self.augment(image)
        # convert label to one hot
        label = tf.one_hot(label, 10)

        return image, label

    def augment(self, image):
        """ Image augmentation """
        image = self._random_flip(image)
        return image

    def _random_flip(self, image):
        """ Flip augmentation randomly"""
        image = tf.image.random_flip_left_right(image)
        image = tf.image.random_flip_up_down(image)

        return image

    def __len__(self):
        """ Length of data """
        return self.len_data