-
Notifications
You must be signed in to change notification settings - Fork 10
/
dataloader.py
194 lines (161 loc) · 8.4 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
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
from absl import logging
import os
import numpy as np
import jax.numpy as jnp
import tensorflow as tf
import tensorflow_datasets as tfds
from .imagewoof import ImagewoofV2
from .imagenette import ImagenetteV2
from .tinyimagenet import TinyImagenetV2
from .ops import compute_zca_mean_cov_ds, get_whitening_transform, get_preprocess_op_np, load_data, process2tfrecord
# Precomputed mean and std
data_stats = {
'mnist': ([0.1307], [0.3081]),
'fashion_mnist': ([0.2861], [0.3530]),
'cifar10': ([0.4914, 0.4822, 0.4465], [0.2470, 0.2435, 0.2616]),
'cifar100': ([0.5071, 0.4866, 0.4409], [0.2673, 0.2564, 0.2762]),
'tiny_imagenet': ([0.4759, 0.4481, 0.3926], [0.2763, 0.2687, 0.2813]),
'imagenette': ([0.4626, 0.4588, 0.4251], [0.2790, 0.2745, 0.2973]),
'imagewoof': ([0.4917, 0.4613, 0.3931], [0.2513, 0.2442, 0.2530]),
'imagenet_resized/32x32': ([0.4811, 0.4575, 0.4079], [0.2604, 0.2532, 0.2682]),
'imagenet_resized/64x64': ([0.4815, 0.4578, 0.4082], [0.2686, 0.2613, 0.2758]),
'caltech_birds2011': ([0.4810, 0.4964, 0.4245], [0.2129, 0.2084, 0.2468])
}
def get_ds_builder(dataset_name, data_dir):
if dataset_name == 'imagewoof':
ds_builder = ImagewoofV2(data_dir=data_dir)
elif dataset_name == 'imagenette':
ds_builder = ImagenetteV2(data_dir=data_dir)
elif dataset_name == 'tiny_imagenet':
ds_builder = TinyImagenetV2(data_dir=data_dir)
else:
ds_builder = tfds.builder(dataset_name, data_dir=data_dir)
ds_builder.download_and_prepare()
return ds_builder
def configure_dataloader(ds, batch_size, x_transform=None, y_transform=None, train=False, shuffle=False, seed=0):
if y_transform is None:
y_transform = lambda x: x
else:
y_transform = y_transform
ds = ds.cache()
if train:
ds = ds.repeat()
if shuffle:
ds = ds.shuffle(16 * batch_size, seed=seed)
if x_transform:
ds = ds.map(lambda x, y: (x_transform(x), y_transform(y)), tf.data.AUTOTUNE)
else:
ds = ds.map(lambda x, y: (x, y_transform(y)), tf.data.AUTOTUNE)
ds = ds.batch(batch_size=batch_size)
ds = ds.prefetch(buffer_size=tf.data.AUTOTUNE)
return ds
def get_dataset(config, return_raw=False):
dataset_name = config.name
data_path = config.data_path
zca_path = config.zca_path
zca_reg = config.zca_reg
if dataset_name in ['imagenet_resized/64x64', 'imagenette', 'imagewoof']:
split = ['train', 'validation']
else:
split = ['train', 'test']
if dataset_name in ['mnist', 'fashion_mnist']:
preprocess_type = 'standard'
else:
preprocess_type = 'normalize_zca'
if dataset_name in ['imagenette', 'imagewoof']:
use_checkboard = True
use_mean_block = True
block_size = 64
resolution = 128
elif dataset_name in ['imagenet_resized/64x64', 'tiny_imagenet']:
use_checkboard = False
use_mean_block = False
block_size = None
resolution = 64
else:
use_checkboard = False
use_mean_block = False
block_size = None
resolution = 32
ds_builder = get_ds_builder(dataset_name, data_path)
img_shape = ds_builder.info.features['image'].shape
num_train, num_test = ds_builder.info.splits[split[0]].num_examples, ds_builder.info.splits[split[1]].num_examples
num_classes, class_names = ds_builder.info.features['label'].num_classes, ds_builder.info.features['label'].names
mean, std = data_stats[dataset_name]
mean, std = np.array(mean), np.array(std)
if preprocess_type == 'standard':
zca_mean, whitening_transform, rev_whitening_transform = None, None, None
elif preprocess_type == 'normalize_zca':
if not os.path.exists(zca_path):
os.makedirs(zca_path)
if '/' in dataset_name:
name = dataset_name.split('/')[0]
if not os.path.exists('{}/{}'.format(zca_path, name)):
os.makedirs('{}/{}'.format(zca_path, name))
if block_size is None:
path = os.path.join(zca_path, '{}_{}.npz'.format(dataset_name, preprocess_type))
else:
if use_checkboard:
path = os.path.join(zca_path,
'{}_{}_res{}_block{}_mean{}_cb.npz'.format(dataset_name, preprocess_type,
resolution, block_size, use_mean_block))
else:
path = os.path.join(zca_path,
'{}_{}_res{}_block{}_mean{}.npz'.format(dataset_name, preprocess_type,
resolution, block_size, use_mean_block))
if not os.path.exists(path):
logging.info('Compute block zca with block_size {} and save to {}!'.format(block_size, path))
ds_train = ds_builder.as_dataset(split='train', as_supervised=True)
zca_mean, cov = compute_zca_mean_cov_ds(ds_train, img_shape, mean=mean, std=std, resolution=resolution,
block_size=block_size, batch_size=10000,
use_checkboard=use_checkboard)
whitening_transform, rev_whitening_transform = get_whitening_transform(cov, num_train, zca_reg=zca_reg,
use_mean_block=use_mean_block)
np.savez(path, whitening_transform=whitening_transform, rev_whitening_transform=rev_whitening_transform,
zca_mean=zca_mean)
else:
logging.info('Load from {}!'.format(path))
npzfile = np.load(path)
whitening_transform, rev_whitening_transform, zca_mean = npzfile['whitening_transform'], npzfile[
'rev_whitening_transform'], npzfile['zca_mean']
else:
raise ValueError('Unknown PreprocessType {}!'.format(preprocess_type))
preprocess_op, rev_preprocess_op = get_preprocess_op_np(mean=mean, std=std, zca_mean=zca_mean,
whitening_transform=whitening_transform,
rev_whitening_transform=rev_whitening_transform,
block_size=block_size, use_mean_block=use_mean_block,
use_checkboard=use_checkboard)
ds_train, ds_test = ds_builder.as_dataset(split=split, as_supervised=True)
if dataset_name in ['imagenet_resized/64x64', 'caltech_birds2011']:
data_dir = os.path.join(zca_path,
'{}_{}_res{}_block{}_mean{}'.format(dataset_name, preprocess_type, resolution,
block_size, use_mean_block))
if not os.path.exists(data_dir):
os.mkdir(data_dir)
if '/' in dataset_name:
name = dataset_name.split('/')[0]
else:
name = dataset_name
process2tfrecord(ds_train, ds_test, data_dir, name, img_shape, num_classes, preprocess_op,
resolution, batch_size=10000, num_per_shard=10000)
builder = tfds.core.builder_from_directory(data_dir)
ds_train, ds_test = builder.as_dataset(split=['train', 'test'], as_supervised=True, shuffle_files=True)
x_train, y_train, x_test, y_test = None, None, None, None
else:
x_train, y_train = load_data(ds_train, img_shape, preprocess_op, resolution, batch_size=5000)
x_test, y_test = load_data(ds_test, img_shape, preprocess_op, resolution, batch_size=5000)
ds_train = tf.data.Dataset.from_tensor_slices((x_train, y_train))
ds_test = tf.data.Dataset.from_tensor_slices((x_test, y_test))
proto_scale = {'x_proto': jnp.sqrt(3 * (resolution ** 2))}
logging.info('Resolution: {}'.format(resolution))
logging.info('Proto Scale: {}'.format(proto_scale))
with config.unlocked():
config.img_shape = (resolution, resolution, 3) if None in img_shape else img_shape
config.num_classes = num_classes
config.class_names = class_names
config.train_size = num_train
config.test_size = num_test
if return_raw:
return (x_train, y_train, x_test, y_test), preprocess_op, rev_preprocess_op, proto_scale
else:
return (ds_train, ds_test), preprocess_op, rev_preprocess_op, proto_scale