-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodels.py
170 lines (131 loc) · 5.53 KB
/
models.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
#! /usr/bin/env python
# -*- coding: utf-8 -*-
# vim:fenc=utf-8
#
# Copyright © 2019 theo <theo@not-arch-linux>
#
# Distributed under terms of the MIT license.
"""
CNN and MLP branches
"""
import tensorflow as tf
from keras import regularizers
from keras.models import Model
from img_utils import crop_and_pad
from keras.layers import BatchNormalization, Dropout, Activation
from keras.layers import Input, Dense, Conv2D, MaxPooling2D, Flatten, Lambda, ReLU
class GatePoseEstimator:
@staticmethod
def build_rotation_branch(inputs, fine_tune):
x = Conv2D(16, (3,3), padding="same", use_bias=False, trainable=(not fine_tune))(inputs)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(32, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(64, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(128, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Flatten()(x)
# x = Dropout(0.5)(x)
# x = Dense(64, use_bias=False)(x)
# x = BatchNormalization()(x)
# x = Activation('linear')(x)
# x = Dropout(0.5)(x)
x = Dense(32, use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
# x = Dropout(0.5)(x)
x = Dense(16, use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
# x = Dropout(0.5)(x)
x = Dense(1, use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('tanh', name='rotation_output')(x)
return x
@staticmethod
def buld_distance_branch(inputs, fine_tune):
x = Conv2D(16, (3,3), padding="same", use_bias=False, trainable=(not fine_tune))(inputs)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(32, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(64, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(128, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Flatten()(x)
x = Dropout(0.5)(x)
x = Dense(32, use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
# x = Dropout(0.5)(x)
x = Dense(16, use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
# x = Dropout(0.5)(x)
x = Dense(1, use_bias=False)(x)
x = BatchNormalization()(x)
x = Activation('relu', name='distance_output')(x)
return x
@staticmethod
def build_rotation_dist_branch(inputs, fine_tune):
x = Conv2D(16, (3,3), padding="same", use_bias=False, trainable=(not fine_tune))(inputs)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(32, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(3, 3))(x)
x = Conv2D(64, (3,3), padding='same', use_bias=False, trainable=(not fine_tune))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = Flatten()(x)
x = Dropout(0.5)(x)
rot = Dense(16, use_bias=False)(x)
rot = BatchNormalization()(rot)
rot = Activation('relu')(rot)
rot = Dropout(0.5)(rot)
rot = Dense(1, use_bias=False)(rot)
rot = BatchNormalization()(rot)
rot = Activation('relu', name='rotation_output')(rot)
dist = Dense(16, use_bias=False)(x)
dist = BatchNormalization()(x)
dist = Activation('relu')(x)
dist = Dropout(0.5)(dist)
dist = Dense(1, use_bias=False)(dist)
dist = BatchNormalization()(dist)
dist = Activation('relu', name='distance_output')(dist)
return [rot, dist]
@staticmethod
def build(model, shape, fine_tune=False):
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
tf.keras.backend.set_session(tf.Session(config=config))
if model not in ['distance', 'rotation', 'combined']:
raise ValueError('Model must be either "distance" or "rotation" or "combined"')
branch_input = Input(shape=shape, name='img_input')
if model == 'distance':
branch_output = GatePoseEstimator.buld_distance_branch(branch_input, fine_tune)
elif model == 'combined':
branch_output = GatePoseEstimator.build_rotation_dist_branch(branch_input, fine_tune)
else:
branch_output = GatePoseEstimator.build_rotation_branch(branch_input, fine_tune)
model = Model(inputs=branch_input, outputs=branch_output,
name='GatePoseEstimator')
print(model.summary())
return model