-
Notifications
You must be signed in to change notification settings - Fork 6
/
models_c.py
165 lines (135 loc) · 6.14 KB
/
models_c.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
# -*- coding: utf-8 -*-
from __future__ import print_function, division
import tensorflow as tf
import keras as K
import keras.layers as L
import numpy as np
import os
import time
import h5py
import argparse
import matplotlib.pyplot as plt
from attention import *
from data_F import *
from keras.callbacks import ModelCheckpoint
from keras.callbacks import EarlyStopping
from keras.models import Model
from keras.layers import Input, Activation, Conv2D, Dropout
from keras.layers import MaxPooling2D, BatchNormalization
from keras.layers import UpSampling2D
from keras.layers import concatenate
from keras.layers import add
from attention import PAM, DPAM
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense,Activation,Layer
import keras.backend as K1
from cross import *
from keras.optimizers import RMSprop
# ===================cascade net=============
def attention_block_3(inputs,feature_cnt,dim):
a = Flatten()(inputs)
a = Dense(feature_cnt*dim,activation='softmax')(a)
a = L.Reshape((feature_cnt,dim,))(a)
a = L.Lambda(lambda x: K1.sum(x, axis=2), name='attention')(a)
a = L.RepeatVector(dim)(a)
a_probs = L.Permute((2, 1), name='attention_vec')(a)
attention_out = L.Multiply()([inputs, a_probs])
return attention_out
def small_cnn_branch(input_tensor, small_mode=True):
filters=[32,64,100,200,256]
conv0_spat=L.Conv2D(filters[2],(3,3),padding='same')(input_tensor)
conv0_spat=L.BatchNormalization(axis=-1)(conv0_spat)
conv0_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv0_spat)
conv1_spat=L.Conv2D(filters[2],(3,3),padding='same')(conv0_spat)
conv1_spat=L.BatchNormalization(axis=-1)(conv1_spat)
conv1_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv1_spat)
conv2_spat=L.Conv2D(filters[3],(1,1),padding='same')(conv1_spat)
conv2_spat=L.BatchNormalization(axis=-1)(conv2_spat)
conv2_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv2_spat)
conv3_spat=L.Conv2D(filters[3],(1,1),padding='same')(conv2_spat)
conv3_spat=L.BatchNormalization(axis=-1)(conv3_spat)
conv3_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv3_spat)
conv3_spat = PAM()(conv3_spat)
pool3=L.MaxPool2D(pool_size=(2,2),padding='same')(conv3_spat)
Dense1=L.Dense(1024)(pool3)
Dense1=L.Activation('relu')(Dense1)
Dense1=L.Dropout(0.5)(Dense1)
Dense2=L.Dense(512)(Dense1)
Dense2=L.Activation('relu')(Dense2)
Dense2=L.Dropout(0.5)(Dense2)
conv7_spat=L.Flatten()(Dense2)
return conv7_spat
def small_cnn_branch_front(input_tensor):
filters=[32,64,100,200,256]
conv0_spat=L.Conv2D(filters[2],(3,3),padding='same')(input_tensor)
conv0_spat=L.BatchNormalization(axis=-1)(conv0_spat)
conv0_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv0_spat)
conv1_spat=L.Conv2D(filters[2],(3,3),padding='same')(conv0_spat)
conv1_spat=L.BatchNormalization(axis=-1)(conv1_spat)
conv1_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv1_spat)
conv2_spat=L.Conv2D(filters[3],(1,1),padding='same')(conv1_spat)
conv2_spat=L.BatchNormalization(axis=-1)(conv2_spat)
conv2_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv2_spat)
conv3_spat=L.Conv2D(filters[3],(1,1),padding='same')(conv2_spat)
conv3_spat=L.BatchNormalization(axis=-1)(conv3_spat)
conv3_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv3_spat)
conv3_spat=L.Conv2D(filters[3],(1,1),padding='same')(conv3_spat)
conv3_spat=L.BatchNormalization(axis=-1)(conv3_spat)
conv3_spat=L.advanced_activations.LeakyReLU(alpha=0.2)(conv3_spat)
return conv3_spat
def small_cnn_branch_latter(input_tensor):
pool1=L.MaxPool2D(pool_size=(2,2),padding='same')(input_tensor)
Dense1=L.Dense(1024)(pool1)
Dense1=L.Activation('relu')(Dense1)
Dense1=L.Dropout(0.4)(Dense1)
Dense2=L.Dense(512)(Dense1)
Dense2=L.Activation('relu')(Dense2)
Dense2=L.Dropout(0.4)(Dense2)
conv7_spat=L.Flatten()(Dense2)
return conv7_spat
def pixel_branch(input_tensor):
filters = [8, 16, 32, 64, 96, 128]
conv0 = L.Conv1D(filters[3], 11, padding='valid')(input_tensor)
conv0_a = attention_block_3(conv0,170,64)
conv0 = L.concatenate([conv0,conv0_a])
conv0 = L.BatchNormalization(axis=-1)(conv0)
conv0 = L.advanced_activations.LeakyReLU(alpha=0.2)(conv0)
conv3 = L.Conv1D(filters[5], 3, padding='valid')(conv0)
conv3 = L.advanced_activations.LeakyReLU(alpha=0.2)(conv3)
conv3 = L.MaxPool1D(pool_size=2, padding='valid')(conv3)
conv3 = L.Conv1D(filters[5], 3, padding='valid')(conv3)
conv3 = L.advanced_activations.LeakyReLU(alpha=0.2)(conv3)
conv3 = L.MaxPool1D(pool_size=2, padding='valid')(conv3)
conv3 = L.Conv1D(filters[5], 3, padding='valid')(conv3)
conv3 = L.advanced_activations.LeakyReLU(alpha=0.2)(conv3)
conv3 = L.MaxPool1D(pool_size=2, padding='valid')(conv3)
conv3 = L.Flatten()(conv3)
return conv3
def merge_branch():
ksize = 2 * r + 1
filters = [64, 128, 256, 512]
hsi_in = L.Input((ksize, ksize, hchn))
hsi_pxin = L.Input((hchn, 1))
lidar_in = L.Input((ksize, ksize,lchn))
h_simple = small_cnn_branch(hsi_in, small_mode=False)
px_out = pixel_branch(hsi_pxin)
ha_simple = small_cnn_branch_front(hsi_in)
la_simple = small_cnn_branch_front(lidar_in)
ha_simple_c = L.Lambda(cross)([la_simple,ha_simple])
la_simple_c = L.Lambda(cross)([ha_simple,la_simple])
ha_simple = DPAM()(ha_simple_c)
la_simple = DPAM()(la_simple_c)
ha_simple = small_cnn_branch_latter(ha_simple)
la_simple = small_cnn_branch_latter(la_simple)
merge1=L.concatenate([h_simple,px_out], axis=-1)
merge1=L.Dropout(0.5)(merge1)
merge2=L.concatenate([ha_simple,la_simple], axis=-1)
merge2=L.Dropout(0.5)(merge2)
merge=L.concatenate([merge1,merge2], axis=-1)
merge=L.Dropout(0.5)(merge)
logits = L.Dense(NUM_CLASS, activation='softmax',name='logits_out')(merge)
model = K.models.Model([hsi_in,hsi_pxin,lidar_in], logits)
adam = K.optimizers.Adam(lr=0.0001,beta_1=0.9,beta_2=0.999,epsilon=1e-8)
optm = K.optimizers.SGD(lr=0.00005,momentum=1e-6,nesterov=True)
model.compile(optimizer=adam,
loss='categorical_crossentropy', metrics=['acc'])
return model