This repository was archived by the owner on Apr 16, 2023. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdataset.py
332 lines (290 loc) · 15.4 KB
/
dataset.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
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
import numpy as np
import random
import tensorflow as tf
from ops.ops import load_json, patchs_full_gen
import os
import math as m
def train_data_gen(list_files, model_props):
with np.load('statistics.npz', mmap_mode='r') as statistics:
opt_mean = statistics['opt_mean']
opt_std = statistics['opt_std']
sar_mean = statistics['sar_mean']
sar_std = statistics['sar_std']
cmap_mean = statistics['cmap_mean']
cmap_std = statistics['cmap_std']
patches_path = os.path.join('img', 'patches')
model_input = model_props['input']
model_output = model_props['output']
def func():
while True:
#random.shuffle(list_files)
for fin in list_files:
with np.load(os.path.join(patches_path, f'{fin}.npz'), mmap_mode='r') as data:
#print(fin)
patches_idx = data['patches_idx']
if model_input == 'opt':
input = data['opt']
if model_input == 'sar':
input = data['sar']
if model_input in {'opt+sar', 'opt_sar'}:
input_opt = data['opt']
input_sar = data['sar']
if model_input in {'opt_sar_cmap', 'opt+sar_sar_cmap', 'opt_sar_sar_cmap'}:
input_opt = data['opt']
input_sar = data['sar']
input_cmap = data['cmap']
if model_output == 'pred_fus':
label = data['label']
for patch in patches_idx:
if model_output == 'pred_fus':
if model_input in {'opt', 'sar'}:
yield input[patch].astype(np.float32), label[patch].astype(np.uint8)
if model_input in {'opt+sar', 'opt_sar'}:
yield input_opt[patch].astype(np.float32), input_sar[patch].astype(np.float32), label[patch].astype(np.uint8)
if model_input in {'opt_sar_cmap', 'opt+sar_sar_cmap', 'opt_sar_sar_cmap'}:
yield input_opt[patch].astype(np.float32), input_sar[patch].astype(np.float32), input_cmap[patch].astype(np.float32), label[patch].astype(np.uint8)
return func
class PredictDataGen(tf.keras.utils.Sequence):
def __init__(self, fin, model_props, batch_size):
patches_path = os.path.join('img', 'patches')
self.batch_size = batch_size
self.model_input = model_props['input']
self.model_output = model_props['output']
self.counter = 0
conf = load_json(os.path.join('conf', 'conf.json'))
patch_size = conf['patch_size']
test_patch_step = conf['test_patch_step']
with np.load('statistics.npz', mmap_mode='r') as statistics:
opt_mean = statistics['opt_mean']
opt_std = statistics['opt_std']
sar_mean = statistics['sar_mean']
sar_std = statistics['sar_std']
cmap_mean = statistics['cmap_mean']
cmap_std = statistics['cmap_std']
with np.load(os.path.join(patches_path, f'{fin}.npz'), mmap_mode='r') as data:
self.img_shape = data['shape']
idx_matrix = np.arange(self.img_shape[0]*self.img_shape[1]).reshape(self.img_shape)
self.patches_idx = patchs_full_gen(idx_matrix, patch_size, test_patch_step)
self.shape_pred_patches = self.patches_idx.shape[:2]
self.patches_idx = self.patches_idx.reshape((-1, patch_size, patch_size))
if self.model_input == 'opt':
self.data = (data['opt'] - opt_mean) / opt_std
if self.model_input == 'sar':
self.data = (data['sar'] - sar_mean) / sar_std
if self.model_input == 'opt+sar':
x_opt = (data['opt'] - opt_mean) / opt_std
x_sar = (data['sar'] - sar_mean) / sar_std
self.data = np.concatenate([x_opt, x_sar], axis=-1)
if self.model_input == 'opt_sar':
self.x_opt = (data['opt'] - opt_mean) / opt_std
self.x_sar = (data['sar'] - sar_mean) / sar_std
if self.model_input in {'opt_sar_cmap', 'opt+sar_sar_cmap'}:
x_opt = (data['opt'] - opt_mean) / opt_std
self.x_sar = (data['sar'] - sar_mean) / sar_std
self.x_fus1 = np.concatenate([x_opt, self.x_sar], axis=-1)
self.x_cmap = (data['cmap'] - cmap_mean) / cmap_std
if self.model_input in {'opt_sar_sar_cmap'}:
self.x_opt = (data['opt'] - opt_mean) / opt_std
self.x_sar = (data['sar'] - sar_mean) / sar_std
self.x_cmap = (data['cmap'] - cmap_mean) / cmap_std
def __len__(self):
return m.ceil((self.shape_pred_patches[0]*self.shape_pred_patches[1])/self.batch_size)
def __getitem__(self, index):
batch_patches_idx = self.patches_idx[index*self.batch_size: (index+1)*self.batch_size]
if self.model_input in {'opt', 'sar', 'opt+sar'}:
return tf.convert_to_tensor(self.data[batch_patches_idx])
if self.model_input == 'opt_sar':
return tf.convert_to_tensor(self.x_opt[batch_patches_idx]), tf.convert_to_tensor(self.x_sar[batch_patches_idx])
if self.model_input == 'opt_sar_cmap':
return tf.convert_to_tensor(self.x_opt[batch_patches_idx]), tf.convert_to_tensor(self.x_sar[batch_patches_idx]), tf.convert_to_tensor(self.x_cmap[batch_patches_idx])
if self.model_input == 'opt+sar_sar_cmap':
return tf.convert_to_tensor(self.x_fus1[batch_patches_idx]), tf.convert_to_tensor(self.x_sar[batch_patches_idx]), tf.convert_to_tensor(self.x_cmap[batch_patches_idx])
if self.model_input == 'opt_sar_sar_cmap':
return (
(tf.convert_to_tensor(self.x_opt[batch_patches_idx]), tf.convert_to_tensor(self.x_sar[batch_patches_idx])),
tf.convert_to_tensor(self.x_sar[batch_patches_idx]),
tf.convert_to_tensor(self.x_cmap[batch_patches_idx])
)
def __iter__(self):
return self
def __call__(self):
return self.next()
def __next__(self):
self.counter+=1
if self.counter > len(self):
raise StopIteration
else:
return self[self.counter-1]
def prep_data(model_props, normalize = True):
conf = load_json(os.path.join('conf', 'conf.json'))
n_opt_layers = conf['n_opt_layers']
n_sar_layers = conf['n_sar_layers']
n_cmap_layers = conf['n_cmap_layers']
n_classes = conf['n_classes']
model_input = model_props['input']
model_output = model_props['output']
if normalize:
with np.load('statistics.npz', mmap_mode='r') as statistics:
opt_mean = statistics['opt_mean']
opt_std = statistics['opt_std']
sar_mean = statistics['sar_mean']
sar_std = statistics['sar_std']
cmap_mean = statistics['cmap_mean']
cmap_std = statistics['cmap_std']
def prep_func(*data):
#prepare data
if model_output == 'pred_fus':
y = tf.one_hot(tf.squeeze(data[-1]), n_classes)
if model_input == 'opt':
x = (data[0] - opt_mean) / opt_std if normalize else data[0]
return x, y
if model_input == 'sar':
x = (data[0] - sar_mean) / sar_std if normalize else data[0]
return x, y
if model_input == 'opt+sar':
x_opt = (data[0] - opt_mean) / opt_std if normalize else data[0]
x_sar = (data[1] - sar_mean) / sar_std if normalize else data[1]
x = tf.concat([x_opt, x_sar], axis=-1)
return x, y
if model_input == 'opt_sar':
x_opt = (data[0] - opt_mean) / opt_std if normalize else data[0]
x_sar = (data[1] - sar_mean) / sar_std if normalize else data[1]
return (x_opt, x_sar), y
if model_input == 'opt_sar_cmap':
x_opt = (data[0] - opt_mean) / opt_std if normalize else data[0]
x_sar = (data[1] - sar_mean) / sar_std if normalize else data[1]
x_cmap = (data[2] - cmap_mean) / cmap_std if normalize else data[2]
return (x_opt, x_sar, x_cmap), y
if model_input == 'opt+sar_sar_cmap':
x_opt = (data[0] - opt_mean) / opt_std if normalize else data[0]
x_sar = (data[1] - sar_mean) / sar_std if normalize else data[1]
x_cmap = (data[2] - cmap_mean) / cmap_std if normalize else data[2]
return (tf.concat([x_opt, x_sar], axis=-1), x_sar, x_cmap), y
if model_input == 'opt_sar_sar_cmap':
x_opt = (data[0] - opt_mean) / opt_std if normalize else data[0]
x_sar = (data[1] - sar_mean) / sar_std if normalize else data[1]
x_cmap = (data[2] - cmap_mean) / cmap_std if normalize else data[2]
return ((x_opt, x_sar), x_sar, x_cmap), y
return prep_func
def data_augmentation(model_props):
model_input = model_props['input']
model_output = model_props['output']
def da_fn(*data):
#print('data aug')
#print(data)
if model_output == 'pred_fus':
y = data[-1]
if model_input in {'opt', 'sar', 'opt+sar'}:
x = data[0]
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x = tf.image.flip_left_right(x)
y = tf.image.flip_left_right(y)
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x = tf.image.flip_up_down(x)
y = tf.image.flip_up_down(y)
k = tf.random.uniform(shape=[], minval=0, maxval=3, dtype=tf.int32)
x = tf.image.rot90(x, k)
y = tf.image.rot90(y, k)
elif model_input in {'opt_sar'}:
x_opt = data[0][0]
x_sar = data[0][1]
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x_opt = tf.image.flip_left_right(x_opt)
x_sar = tf.image.flip_left_right(x_sar)
y = tf.image.flip_left_right(y)
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x_opt = tf.image.flip_up_down(x_opt)
x_sar = tf.image.flip_up_down(x_sar)
y = tf.image.flip_up_down(y)
k = tf.random.uniform(shape=[], minval=0, maxval=3, dtype=tf.int32)
x_opt = tf.image.rot90(x_opt, k)
x_sar = tf.image.rot90(x_sar, k)
y = tf.image.rot90(y, k)
x = (x_opt, x_sar)
elif model_input in {'opt_sar_cmap', 'opt+sar_sar_cmap'}:
x_opt = data[0][0]
x_sar = data[0][1]
x_cmap = data[0][2]
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x_opt = tf.image.flip_left_right(x_opt)
x_sar = tf.image.flip_left_right(x_sar)
x_cmap = tf.image.flip_left_right(x_cmap)
y = tf.image.flip_left_right(y)
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x_opt = tf.image.flip_up_down(x_opt)
x_sar = tf.image.flip_up_down(x_sar)
x_cmap = tf.image.flip_up_down(x_cmap)
y = tf.image.flip_up_down(y)
k = tf.random.uniform(shape=[], minval=0, maxval=3, dtype=tf.int32)
x_opt = tf.image.rot90(x_opt, k)
x_sar = tf.image.rot90(x_sar, k)
x_cmap = tf.image.rot90(x_cmap, k)
y = tf.image.rot90(y, k)
x = (x_opt, x_sar, x_cmap)
elif model_input == 'opt_sar_sar_cmap':
x_opt = data[0][0][0]
x_sar = data[0][0][1]
x_cmap = data[0][2]
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x_opt = tf.image.flip_left_right(x_opt)
x_sar = tf.image.flip_left_right(x_sar)
x_cmap = tf.image.flip_left_right(x_cmap)
y = tf.image.flip_left_right(y)
if tf.math.greater(tf.random.uniform(shape=[], minval=0, maxval=1), tf.constant(0.5)):
x_opt = tf.image.flip_up_down(x_opt)
x_sar = tf.image.flip_up_down(x_sar)
x_cmap = tf.image.flip_up_down(x_cmap)
y = tf.image.flip_up_down(y)
k = tf.random.uniform(shape=[], minval=0, maxval=3, dtype=tf.int32)
x_opt = tf.image.rot90(x_opt, k)
x_sar = tf.image.rot90(x_sar, k)
x_cmap = tf.image.rot90(x_cmap, k)
y = tf.image.rot90(y, k)
x = ((x_opt, x_sar), x_sar, x_cmap)
return x, y
return da_fn
def get_dataset(list_files, model_props):
conf = load_json(os.path.join('conf', 'conf.json'))
patch_size = conf['patch_size']
batch_size = conf['batch_size']
n_opt_layers = conf['n_opt_layers']
n_sar_layers = conf['n_sar_layers']
n_cmap_layers = conf['n_cmap_layers']
model_input = model_props['input']
model_output = model_props['output']
if model_output == 'pred_fus':
if model_input == 'opt':
output_signature = (
tf.TensorSpec(shape=(patch_size , patch_size , n_opt_layers), dtype=tf.float32),
tf.TensorSpec(shape=(patch_size , patch_size , 1), dtype=tf.uint8),
)
if model_input == 'sar':
output_signature = (
tf.TensorSpec(shape=(patch_size , patch_size , n_sar_layers), dtype=tf.float32),
tf.TensorSpec(shape=(patch_size , patch_size , 1), dtype=tf.uint8),
)
if model_input in {'opt+sar', 'opt_sar'}:
output_signature = (
tf.TensorSpec(shape=(patch_size , patch_size , n_opt_layers), dtype=tf.float32),
tf.TensorSpec(shape=(patch_size , patch_size , n_sar_layers), dtype=tf.float32),
tf.TensorSpec(shape=(patch_size , patch_size , 1), dtype=tf.uint8),
)
if model_input in {'opt_sar_cmap', 'opt+sar_sar_cmap', 'opt_sar_sar_cmap'}:
output_signature = (
tf.TensorSpec(shape=(patch_size , patch_size , n_opt_layers), dtype=tf.float32),
tf.TensorSpec(shape=(patch_size , patch_size , n_sar_layers), dtype=tf.float32),
tf.TensorSpec(shape=(patch_size , patch_size , n_cmap_layers), dtype=tf.float32),
tf.TensorSpec(shape=(patch_size , patch_size , 1), dtype=tf.uint8),
)
ds = tf.data.Dataset.from_generator(
generator = train_data_gen(list_files, model_props),
output_signature = output_signature
)
return ds
def get_n_steps(list_files):
n = 0
patches_path = os.path.join('img', 'patches')
for fin in list_files:
data = np.load(os.path.join(patches_path, f'{fin}.npz'), mmap_mode='r')
n += data['patches_idx'].shape[0]
return n