-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathNETWORK.py
155 lines (117 loc) · 6.57 KB
/
NETWORK.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
# -*- coding: utf-8 -*-
from keras.models import Model, Sequential, model_from_json
from keras import backend as K
from keras.layers import Input, Lambda
from keras.layers.core import Dense, Dropout, Flatten, Activation
from keras.layers import Conv2D, MaxPooling2D, Reshape
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.merge import concatenate
from keras.optimizers import SGD
from keras.layers.recurrent import LSTM
from keras.optimizers import Adam, Nadam
from keras.layers.wrappers import TimeDistributed
from keras.callbacks import ModelCheckpoint
import numpy as np
import os, sys, cv2, glob
import pandas as pd
from sklearn.model_selection import train_test_split
def CNN(img_rows, img_cols):
# 左目のネットワーク
left_inputs = Input(shape=(img_rows,img_cols,1))
conv1_l = Conv2D(20, (3, 3), padding='same', activation='linear')(left_inputs)
relu1_l = LeakyReLU(alpha=0.01)(conv1_l)
conv2_l = Conv2D(20, (3, 3), padding='same', activation='linear')(relu1_l)
relu2_l = LeakyReLU(alpha=0.01)(conv2_l)
pool1_l = MaxPooling2D(pool_size=(2 ,2))(conv2_l)
drop1_l = Dropout(0.5)(pool1_l)
conv3_l = Conv2D(50, (3, 3), padding='same', activation='linear')(drop1_l)
relu3_l = LeakyReLU(alpha=0.01)(conv3_l)
conv4_l = Conv2D(50, (3, 3), padding='same', activation='linear')(relu3_l)
relu4_l = LeakyReLU(alpha=0.01)(conv4_l)
pool2_l = MaxPooling2D(pool_size=(2 ,2))(conv4_l)
drop2_l = Dropout(0.5)(pool2_l)
flatten_l = Flatten()(drop2_l)
dense_l = Dense(512, activation='linear')(flatten_l)
# 右目のネットワーク
right_inputs = Input(shape=(img_rows,img_cols,1))
conv1_r = Conv2D(20, (3, 3), padding='same', activation='linear')(right_inputs)
relu1_r = LeakyReLU(alpha=0.01)(conv1_r)
conv2_r = Conv2D(20, (3, 3), padding='same', activation='linear')(relu1_r)
relu2_r = LeakyReLU(alpha=0.01)(conv2_r)
pool1_r = MaxPooling2D(pool_size=(2 ,2))(conv2_r)
drop1_r = Dropout(0.5)(pool1_r)
conv3_r = Conv2D(50, (3, 3), padding='same', activation='linear')(drop1_r)
relu3_r = LeakyReLU(alpha=0.01)(conv3_r)
conv4_r = Conv2D(50, (3, 3), padding='same', activation='linear')(relu3_r)
relu4_r = LeakyReLU(alpha=0.01)(conv4_r)
pool2_r = MaxPooling2D(pool_size=(2 ,2))(conv4_r)
drop2_r = Dropout(0.5)(pool2_r)
flatten_r = Flatten()(drop2_r)
dense_r = Dense(512, activation='linear')(flatten_r)
# ネットワークをマージ
concat = concatenate([dense_l,dense_r])
relu1 = LeakyReLU(alpha=0.01)(concat)
drop1 = Dropout(0.5)(relu1)
dense2 = Dense(1024, activation='linear', name='dense2_layer')(drop1)
relu2 = LeakyReLU(alpha=0.01)(dense2)
drop2 = Dropout(0.5)(relu2)
dense3 = Dense(1024, activation='linear', name='dense3_layer')(drop2)
relu3 = LeakyReLU(alpha=0.01)(dense3)
drop3 = Dropout(0.5)(relu3)
output = Dense(2,activation='softmax')(drop3)
model = Model(inputs=[left_inputs,right_inputs],outputs=[output])
adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
sgd = SGD(lr=5e-4, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(optimizer=sgd, loss='categorical_crossentropy', metrics=["accuracy"])
return model
def CNN_Time(time_step,img_rows, img_cols):
#まずCNN
# 左目のネットワーク
left_inputs = Input(shape=(time_step, img_rows,img_cols,1))
conv1_l = TimeDistributed(Conv2D(20, (3, 3), padding='same', activation='linear'))(left_inputs)
relu1_l = TimeDistributed(LeakyReLU(alpha=0.01))(conv1_l)
conv2_l = TimeDistributed(Conv2D(20, (3, 3), padding='same', activation='linear'))(relu1_l)
relu2_l = TimeDistributed(LeakyReLU(alpha=0.01))(conv2_l)
pool1_l = TimeDistributed(MaxPooling2D(pool_size=(2 ,2)))(conv2_l)
drop1_l = TimeDistributed(Dropout(0.5))(pool1_l)
conv3_l = TimeDistributed(Conv2D(50, (3, 3), padding='same', activation='linear'))(drop1_l)
relu3_l = TimeDistributed(LeakyReLU(alpha=0.01))(conv3_l)
conv4_l = TimeDistributed(Conv2D(50, (3, 3), padding='same', activation='linear'))(relu3_l)
relu4_l = TimeDistributed(LeakyReLU(alpha=0.01))(conv4_l)
pool2_l = TimeDistributed(MaxPooling2D(pool_size=(2 ,2)))(conv4_l)
drop2_l = TimeDistributed(Dropout(0.5))(pool2_l)
flatten_l = TimeDistributed(Flatten())(drop2_l)
dense_l = TimeDistributed(Dense(512, activation='linear'), name='dense_l_layer')(flatten_l)
# 右目のネットワーク
right_inputs = Input(shape=(time_step, img_rows ,img_cols ,1))
conv1_r = TimeDistributed(Conv2D(20, (3, 3), padding='same', activation='linear'))(right_inputs)
relu1_r = TimeDistributed(LeakyReLU(alpha=0.01))(conv1_r)
conv2_r = TimeDistributed(Conv2D(20, (3, 3), padding='same', activation='linear'))(relu1_r)
relu2_r = TimeDistributed(LeakyReLU(alpha=0.01))(conv2_r)
pool1_r = TimeDistributed(MaxPooling2D(pool_size=(2 ,2)))(conv2_r)
drop1_r = TimeDistributed(Dropout(0.5))(pool1_r)
conv3_r = TimeDistributed(Conv2D(50, (3, 3), padding='same', activation='linear'))(drop1_r)
relu3_r = TimeDistributed(LeakyReLU(alpha=0.01))(conv3_r)
conv4_r = TimeDistributed(Conv2D(50, (3, 3), padding='same', activation='linear'))(relu3_r)
relu4_r = TimeDistributed(LeakyReLU(alpha=0.01))(conv4_r)
pool2_r = TimeDistributed(MaxPooling2D(pool_size=(2 ,2)))(conv4_r)
drop2_r = TimeDistributed(Dropout(0.5))(pool2_r)
flatten_r = TimeDistributed(Flatten())(drop2_r)
dense_r = TimeDistributed(Dense(512, activation='linear'), name='dense_r_layer')(flatten_r)
# ネットワークをマージ
concat = concatenate([dense_l,dense_r])
relu1 = TimeDistributed(LeakyReLU(alpha=0.01), name='relu1_layer')(concat)
drop1 = TimeDistributed(Dropout(0.5))(relu1)
dense2 = TimeDistributed(Dense(1024, activation='linear'), name='dense2_layer')(drop1)
relu2 = TimeDistributed(LeakyReLU(alpha=0.01))(dense2)
drop2 = TimeDistributed(Dropout(0.5))(relu2)
dense3 = TimeDistributed(Dense(1024, activation='linear'), name='dense3_layer')(drop2)
relu3 = TimeDistributed(LeakyReLU(alpha=0.01))(dense3)
drop3 = TimeDistributed(Dropout(0.5))(relu3)
output = TimeDistributed(Dense(2,activation='softmax'), name='softmax_layer')(drop3)
cnn_model = Model(inputs=[left_inputs,right_inputs],outputs=[output])
adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
sgd = SGD(lr=5e-4, decay=1e-6, momentum=0.9, nesterov=True)
nadam = Nadam(lr=0.002, beta_1=0.9, beta_2=0.999, schedule_decay=0.004)
cnn_model.compile(optimizer=nadam, loss='categorical_crossentropy', metrics=["accuracy"])
return cnn_model