-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdata.py
204 lines (168 loc) · 7.58 KB
/
data.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
195
196
197
198
199
200
201
202
203
204
import numpy as np
import tensorflow as tf
import multiprocessing
def batch_dataset(dataset,
batch_size,
drop_remainder=True,
n_prefetch_batch=1,
filter_fn=None,
map_fn=None,
n_map_threads=None,
filter_after_map=False,
shuffle=True,
shuffle_buffer_size=None,
repeat=None):
# set defaults
if n_map_threads is None:
n_map_threads = multiprocessing.cpu_count()
if shuffle and shuffle_buffer_size is None:
shuffle_buffer_size = max(batch_size * 128, 2048) # set the minimum buffer size as 2048
# [*] it is efficient to conduct `shuffle` before `map`/`filter` because `map`/`filter` is sometimes costly
if shuffle:
dataset = dataset.shuffle(shuffle_buffer_size)
if not filter_after_map:
if filter_fn:
dataset = dataset.filter(filter_fn)
if map_fn:
dataset = dataset.map(map_fn, num_parallel_calls=n_map_threads)
else: # [*] this is slower
if map_fn:
dataset = dataset.map(map_fn, num_parallel_calls=n_map_threads)
if filter_fn:
dataset = dataset.filter(filter_fn)
dataset = dataset.batch(batch_size, drop_remainder=drop_remainder)
dataset = dataset.repeat(repeat).prefetch(n_prefetch_batch)
return dataset
def memory_data_batch_dataset(memory_data,
batch_size,
drop_remainder=True,
n_prefetch_batch=1,
filter_fn=None,
map_fn=None,
n_map_threads=None,
filter_after_map=False,
shuffle=True,
shuffle_buffer_size=None,
repeat=None):
"""Batch dataset of memory data.
Parameters
----------
memory_data : nested structure of tensors/ndarrays/lists
"""
dataset = tf.data.Dataset.from_tensor_slices(memory_data)
dataset = batch_dataset(dataset,
batch_size,
drop_remainder=drop_remainder,
n_prefetch_batch=n_prefetch_batch,
filter_fn=filter_fn,
map_fn=map_fn,
n_map_threads=n_map_threads,
filter_after_map=filter_after_map,
shuffle=shuffle,
shuffle_buffer_size=shuffle_buffer_size,
repeat=repeat)
return dataset
def disk_image_batch_dataset(img_paths,
batch_size,
labels=None,
drop_remainder=True,
n_prefetch_batch=1,
filter_fn=None,
map_fn=None,
n_map_threads=None,
filter_after_map=False,
shuffle=True,
shuffle_buffer_size=None,
repeat=None):
"""Batch dataset of disk image for PNG and JPEG.
Parameters
----------
img_paths : 1d-tensor/ndarray/list of str
labels : nested structure of tensors/ndarrays/lists
"""
if labels is None:
memory_data = img_paths
else:
memory_data = (img_paths, labels)
def parse_fn(path, *label):
img = tf.io.read_file(path)
img = tf.image.decode_png(img, 3) # fix channels to 3
return (img,) + label
if map_fn: # fuse `map_fn` and `parse_fn`
def map_fn_(*args):
return map_fn(*parse_fn(*args))
else:
map_fn_ = parse_fn
dataset = memory_data_batch_dataset(memory_data,
batch_size,
drop_remainder=drop_remainder,
n_prefetch_batch=n_prefetch_batch,
filter_fn=filter_fn,
map_fn=map_fn_,
n_map_threads=n_map_threads,
filter_after_map=filter_after_map,
shuffle=shuffle,
shuffle_buffer_size=shuffle_buffer_size,
repeat=repeat)
return dataset
def make_dataset(img_paths, labels, batch_size, load_size, crop_size, training, drop_remainder=True, shuffle=True, repeat=1):
if training:
@tf.function
def _map_fn(img, *label): # preprocessing
img = tf.image.random_flip_left_right(img)
img = tf.image.resize(img, [load_size, load_size])
img = tf.image.random_crop(img, [crop_size, crop_size, tf.shape(img)[-1]])
img = tf.clip_by_value(img, 0, 255) / 255.0 # or img = tl.minmax_norm(img)
img = img * 2 - 1
return (img,) + label
else:
@tf.function
def _map_fn(img, *label): # preprocessing
img = tf.image.resize(img, [crop_size, crop_size]) # or img = tf.image.resize(img, [load_size, load_size]); img = tl.center_crop(img, crop_size)
img = tf.clip_by_value(img, 0, 255) / 255.0 # or img = tl.minmax_norm(img)
img = img * 2 - 1
return (img,) + label
return disk_image_batch_dataset(img_paths,
batch_size,
labels,
drop_remainder=drop_remainder,
map_fn=_map_fn,
shuffle=shuffle,
repeat=repeat)
def make_zip_dataset(A_img_paths, B_img_paths, A_label, B_label, batch_size, load_size, crop_size, training, shuffle=True, repeat=False):
# zip two datasets aligned by the longer one
if repeat:
A_repeat = B_repeat = None # cycle both
else:
if len(A_img_paths) >= len(B_img_paths):
A_repeat = 1
B_repeat = None # cycle the shorter one
else:
A_repeat = None # cycle the shorter one
B_repeat = 1
A_dataset = make_dataset(A_img_paths, A_label, batch_size, load_size, crop_size, training, drop_remainder=True, shuffle=shuffle, repeat=A_repeat)
B_dataset = make_dataset(B_img_paths, B_label, batch_size, load_size, crop_size, training, drop_remainder=True, shuffle=shuffle, repeat=B_repeat)
A_B_dataset = tf.data.Dataset.zip((A_dataset, B_dataset))
len_dataset = max(len(A_img_paths), len(B_img_paths)) // batch_size
return A_B_dataset, len_dataset
class ItemPool:
def __init__(self, pool_size=50):
self.pool_size = pool_size
self.items = []
def __call__(self, in_items):
# `in_items` should be a batch tensor
if self.pool_size == 0:
return in_items
out_items = []
for in_item in in_items:
if len(self.items) < self.pool_size:
self.items.append(in_item)
out_items.append(in_item)
else:
if np.random.rand() > 0.5:
idx = np.random.randint(0, len(self.items))
out_item, self.items[idx] = self.items[idx], in_item
out_items.append(out_item)
else:
out_items.append(in_item)
return tf.stack(out_items, axis=0)