forked from freelzy/Baidu_Dogs
-
Notifications
You must be signed in to change notification settings - Fork 0
/
ExtractFeatures.py
125 lines (116 loc) · 4.93 KB
/
ExtractFeatures.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
#Baseline参考博客 https://ypw.io/dogs-vs-cats-2/#more
from keras.applications import *
from keras.preprocessing.image import *
import h5py
import math
import gc
from keras.models import *
from keras.layers import *
import numpy as np
import pandas as pd
from sklearn.utils import shuffle
from keras.callbacks import *
from nets.DenseNet.densenet121 import DenseNet121
from nets.DenseNet.densenet161 import DenseNet161
from nets.DenseNet.densenet169 import DenseNet169
from nets.resnet152 import ResNet152
from nets.resnet101 import ResNet101
from nets import inception_resnet_v2
from nets.inception_resnet_v2 import InceptionResNetV2
from nets.InceptionV4.inception_v4 import InceptionV4
import warnings
warnings.filterwarnings('ignore')
import winsound
Freq = 2500
Dur = 1200
#VGG，ResNet图片预处理
def preprocess_input(x):
x[:, :, 0] = (x[:, :, 0] - 124) * 0.0167
x[:, :, 1] = (x[:, :, 1] - 117) * 0.0167
x[:, :, 2] = (x[:, :, 2] - 104) * 0.0167
return x
def r_preprocess_input(x):
# 'RGB'->'BGR'
x = x[:, :, ::-1]
# Zero-center by mean pixel
x[:, :, 0] -= 103.939
x[:, :, 1] -= 116.779
x[:, :, 2] -= 123.68
return x
def d_preprocess_input(x):
x = x[:, :, ::-1]
x[:, :, 0] = (x[:, :, 0] - 103.94) * 0.017
x[:, :, 1] = (x[:, :, 1] - 116.78) * 0.017
x[:, :, 2] = (x[:, :, 2] - 123.68) * 0.017
return x
# def addup(x):
# return 0.5*(x[0]+x[1])
h5path="D:\\PyCharm\\PyProjects\\NN\\model-final\\"
densenet_weights_root='nets\\DenseNet\\imagenet_models\\'
def write_gap(MODEL, image_size, lambda_func=None):
batch_size = 32
print(MODEL.__name__)
if MODEL.__name__ in ['ResNet50','InceptionV3','Xception','ResNet152','ResNet101','InceptionResNetV2']:
batch_size = 20
input_tensor = Input((image_size[0], image_size[1], 3))
x = input_tensor
base_model = MODEL(input_tensor=x, weights='imagenet', include_top=False)
out=GlobalAveragePooling2D()(base_model.output)
print(out.shape)
model = Model(input_tensor, out)
elif MODEL.__name__=='DenseNet121':
densenet_weights_path = densenet_weights_root +'densenet121_weights_tf.h5'
model = MODEL(reduction=0.5, weights_path=densenet_weights_path)
elif MODEL.__name__=='DenseNet161':
densenet_weights_path = densenet_weights_root +'densenet161_weights_tf.h5'
model = MODEL(reduction=0.5, weights_path=densenet_weights_path)
elif MODEL.__name__=='DenseNet169':
densenet_weights_path = densenet_weights_root + 'densenet169_weights_tf.h5'
model = MODEL(reduction=0.5, weights_path=densenet_weights_path)
elif MODEL.__name__=='InceptionV4':
base_model = MODEL(weights='imagenet', include_top=False)
out = GlobalAveragePooling2D()(base_model.output)
model = Model(base_model.input, out)
gen = ImageDataGenerator(
preprocessing_function=lambda_func
)
classes = list(range(97))
for i, c in zip(range(97), classes):
classes[i] = str(c)
train_generator = gen.flow_from_directory(
"bddogtrain\\train",
image_size,
shuffle=False,
batch_size=batch_size,
classes=classes,
)
test_generator = gen.flow_from_directory(
"bddogtrain\\test",
image_size,
shuffle=False,
batch_size=batch_size,
class_mode=None
)
y_train=np.array(train_generator.classes)
# y_t=pd.Series(train_generator.filenames).str.split('\\').apply(lambda x:x[0]).astype(int).values
# print((y_train-y_t).sum())
train = model.predict_generator(train_generator, math.ceil(len(train_generator.filenames) / batch_size),verbose=1)
with h5py.File(h5path+"gap_%s.h5" % MODEL.__name__) as h:
h.create_dataset("train", data=train)
h.create_dataset("label", data=y_train)
test = model.predict_generator(test_generator, math.ceil(len(test_generator.filenames)/batch_size),verbose=1)
with h5py.File(h5path+"gap_%s_test.h5" % MODEL.__name__) as h:
h.create_dataset("test", data=test)
# Xception 和 Inception V3 都需要将数据限定在 (-1, 1) 的范围内
write_gap(ResNet50, (224, 224),r_preprocess_input)
write_gap(ResNet101, (224, 224), r_preprocess_input)
write_gap(ResNet152, (224, 224), r_preprocess_input)
write_gap(InceptionV3, (299, 299), inception_v3.preprocess_input)
write_gap(InceptionV4, (299, 299), inception_v3.preprocess_input)
write_gap(Xception, (299, 299), xception.preprocess_input)
write_gap(InceptionResNetV2, (299, 299), inception_resnet_v2.preprocess_input)
write_gap(DenseNet121, (224, 224), d_preprocess_input)
write_gap(DenseNet161, (224, 224), d_preprocess_input)
write_gap(DenseNet169, (224, 224), d_preprocess_input)
winsound.Beep(Freq,Dur)
gc.collect()