-
Notifications
You must be signed in to change notification settings - Fork 5
/
preprocess.py
139 lines (121 loc) · 4.53 KB
/
preprocess.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
import h5py
import numpy as np
import os
from config import config_dict
import math
import utils
clip_min_max = [
#### s1
(-0.7, 0.7), (-0.5, 0.5),
(-2.0, 2.0), (-2.0, 2.0),
(0.0, 0.25), (0.0, 1.0),
(-1.0, 1.0), (-0.2, 0.2),
#### s2
(0.0, 0.5), (0.0, 0.5), (0.0, 0.5), ###BGR
(0.0, 0.5), (0.0, 0.5), (0.0, 0.5), (0.0, 0.5), (0.0, 0.5), (0.0, 0.5), (0.0, 0.5), ###这些波段有双峰现象
]
def clip_by_min_max(data_18, eas_scale=1):
"""
:param data_18: N *32*32*18 data
:param eas_scale: 放宽系数
:return:
"""
print('clip data by channel specific min max val')
assert len(data_18.shape) == 4
assert data_18.shape[3]==18
for i in range(18):
min, max = clip_min_max[i]
min, max = eas_scale*min, eas_scale*max
data_18[:, :, :, i] = np.clip(data_18[:, :, :, i], min, max)
return data_18
def simple_clip(data, min=-10,max=10):
data = np.clip(data,min,max)
return data
def data_norm_0_1(data_18):
print('norm data by min max val')
assert len(data_18.shape) == 4
assert data_18.shape[3] == 18
for i in range(18):
min, max = clip_min_max[i]
data_18[:, :, :, i] = (data_18[:, :, :, i]- min)/ (max-min)
return data_18
def load_raw_data(path,load_label=True):
fid = h5py.File(path,'r')
s1 = fid['sen1']
s2 = fid['sen2']
if load_label:
label = fid['label']
return s1,s2,label
return s1,s2,None
def gen_data(s1,s2, label, data_save_path,
label_save_path,clip_func=simple_clip, norm_func=data_norm_0_1):
## check
assert len(s1) == len(s2)
if label is not None:
assert len(s1) == len(label)
s1 = np.array(s1,np.float32)
s2 = np.array(s2,np.float32)
s = np.concatenate((s1,s2),3)
print('data shape: ',s.shape)
s = clip_func(s)
### 不统一做归一化, 统一归一化影响效果
###s = norm_func(s)
np.save(data_save_path,s)
print('Saved in %s' % data_save_path)
if label is not None:
label = np.array(label, np.float32)
np.save(label_save_path,label)
print('Saved in %s' % label_save_path)
def calc_trainset_mean_std(s1,s2):
print('calc trainset mean and std...')
s1 = np.array(s1, np.float32)
s2 = np.array(s2, np.float32)
s = np.concatenate((s1, s2), 3)
s = clip_by_min_max(s)
print('data shape: ', s.shape)
means = np.zeros((18, 1), np.float32)
stds = np.zeros((18, 1), np.float32)
for i in range(18):
mean = s[:, :, :, i].mean()
means[i] = mean
std = s[:, :, :, i].std(ddof=1)
stds[i] = std
print('channel %d, mean: %3f, std: %3f' % (i,mean,std) )
data_dir_root = config_dict['data_root_dir']
save_path = os.path.join(data_dir_root, 'preprocess_dir', 'means.npy')
np.save(save_path, means)
save_path = os.path.join(data_dir_root, 'preprocess_dir', 'stds.npy')
np.save(save_path, stds)
print(means)
print(stds)
def preprocess():
data_dir_root = config_dict['data_root_dir']
train_h5_file = os.path.join(data_dir_root, 'training.h5')
val_h5_file = os.path.join(data_dir_root, 'validation.h5')
#test_h5_file = os.path.join(data_dir_root, 'round1_test_a_20181109.h5')
#test_h5_file = os.path.join(data_dir_root, 'round1_test_b_20190104.h5')
test_h5_file=os.path.join(data_dir_root, 'round2_test_b_20190211.h5')
s1_train, s2_train, label_train = load_raw_data(train_h5_file)
s1_val, s2_val, label_val = load_raw_data(val_h5_file)
s1_test, s2_test, _ = load_raw_data(test_h5_file, load_label=False)
clip_func = clip_by_min_max
norm_func = data_norm_0_1#data_norm_mean_std
# calc_trainset_mean_std(s1_train,s2_train)
print('preprocess train data...')
gen_data(s1_train, s2_train, label_train,
os.path.join(config_dict['preprocess_dir'], 'stage1_train_data.npy'),
os.path.join(config_dict['preprocess_dir'], 'stage1_train_label.npy'),
clip_func, norm_func)
print('preprocess val data...')
gen_data(s1_val, s2_val, label_val,
os.path.join(config_dict['preprocess_dir'], 'stage1_val_data.npy'),
os.path.join(config_dict['preprocess_dir'], 'stage1_val_label.npy'),
clip_func, norm_func)
print('preprocess test data...')
gen_data(s1_test, s2_test, None,
os.path.join(config_dict['preprocess_dir'], 'stage2_test_data_b.npy'),
None,
clip_func, norm_func)
print('finished!')
if __name__=='__main__':
preprocess()