/
generators.py
232 lines (193 loc) · 8.03 KB
/
generators.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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Dec 22 13:16:12 2021
@author: c
"""
#%%
import os
import tensorflow as tf
from tensorflow.compat.v1.keras.backend import set_session
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
config.log_device_placement = True
sess = tf.compat.v1.Session(config=config)
set_session(sess)
from tensorflow.keras.initializers import RandomNormal
from tensorflow.keras.models import Model
from tensorflow.keras.layers import *
import tensorflow_addons as tfa
import tensorflow.keras.backend as K
from matplotlib.image import imread
from tensorflow_addons.layers import InstanceNormalization, GroupNormalization
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import time
#%%
#helper functions
def resnet_block(f,input_layer,red = False, norm= True):
shortcut = input_layer
if red:
g = Conv2D(f, (3,3), padding='same', strides =(2,2), kernel_initializer='he_normal')(input_layer)
shortcut = Conv2D(f, (1,1), padding='same', strides=(2,2), kernel_initializer='he_normal')(shortcut)
if norm:
shortcut = GroupNormalization(groups=min(f,norm))(shortcut)
else:
g = Conv2D(f, (3,3), padding='same', kernel_initializer='he_normal')(input_layer)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2D(f, (3,3), padding='same', kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Add()([g, shortcut])
g = Activation('relu')(g)
return g
def up_block(f, input_layer, skip, norm = True):
g = UpSampling2D(size=(2,2),interpolation='nearest')(input_layer)
g = Conv2D(f, (3,3), strides=(1,1), padding='same')(g)
if norm:
g = GroupNormalization(groups=f)(g)
g = Activation('elu')(g)
g = Concatenate()([g,skip])
g = Conv2D(f, (3,3), padding='same',strides=(1,1))(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('elu')(g)
return g
def unet_enc(f, input_layer, red=True, norm=True):
g = Conv2D(f, (3,3), padding='same', kernel_initializer='he_normal')(input_layer)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2D(f, (3,3), padding='same', kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=f)(g)
g = Activation('relu')(g)
shortcut = g
if red:
g = Conv2D(f, (3,3), padding='same', strides=(2,2), kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
return(g,shortcut)
def unet_dec(f, input_layer, skip, short=True, norm = True):
g = Conv2DTranspose(f, (3,3), strides=(2,2), padding='same', kernel_initializer='he_normal')(input_layer)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
if short:
g = Concatenate()([g,skip])
g = Conv2D(f, (3,3), padding='same', kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2D(f, (3,3), padding='same', kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
return(g)
def residual_block(f, input_layer,norm=True):
g = Conv2D(f, (3,3), padding='same',kernel_initializer='he_normal')(input_layer)
#if norm:
#g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2D(f, (3,3), padding='same',kernel_initializer='he_normal')(g)
#if norm:
#g = GroupNormalization(groups=min(f,norm))(g)
g = Add()([g, input_layer])
return g
#%%
def define_styletransfer(inp_dim,out_dim,f,norm, out_act = 'tanh'):
f = int(f); norm = int(norm)
in_image = Input(shape=inp_dim)
g = Conv2D(f, (7,7), padding='same', strides=(1,1), kernel_initializer='he_normal')(in_image)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2D(2*f, (3,3), strides=(2,2), padding='same',kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2D(4*f, (3,3), strides=(2,2), padding='same',kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
for kkk in range(9):
g = residual_block(4*f, g,norm=norm)
g = Conv2DTranspose(2*f, (3,3), strides=(2,2), padding='same',kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2DTranspose(f, (3,3), strides=(2,2), padding='same',kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g = Conv2D(out_dim[-1], (7,7), padding='same', strides=(1,1), kernel_initializer='he_normal')(g)
out_image = Activation(out_act)(g)
model = Model(in_image, out_image)
return model
#%%
def define_unet(inp_dim,out_dim,f,norm, out_act = 'tanh'):
f=int(f); norm=int(norm)
in_image = Input(shape=inp_dim)
g, g_0 = unet_enc(f, in_image, red= True, norm=norm)
g, g_1 = unet_enc(2*f, g, red= True, norm=norm)
g, g_2 = unet_enc(4*f, g, red= True, norm=norm)
g, g_3 = unet_enc(8*f, g, red= True, norm=norm)
g, foo = unet_enc(16*f, g, red= False, norm=norm)
g = unet_dec(8*f, g, g_3, norm=norm)
g = unet_dec(4*f, g, g_2, norm=norm)
#disp1 = Conv2D(out_dim[-1], (3,3), padding='same',strides=(1,1),
#activation = out_act)(g)
#disp1 = UpSampling2D(size=(4,4),interpolation='nearest')(disp1)
g = unet_dec(2*f, g, g_1, norm=norm)
#disp2 = Conv2D(out_dim[-1], (3,3), padding='same',strides=(1,1),
#activation = out_act)(g)
#disp2 = UpSampling2D(size=(2,2),interpolation='nearest')(disp2)
g = unet_dec(1*f, g, g_0, norm=norm)
out_image = Conv2D(out_dim[-1], (3,3), padding='same',strides=(1,1),
activation = out_act, kernel_initializer='he_normal')(g)
model = Model(in_image, out_image)
return model
#%%
def define_resnet18(inp_dim,out_dim,f, norm,out_act = 'tanh'):
f=int(f);norm=int(norm)
# taken from Goddard - Digging Into Self-Supervised Monocular Depth Estimation
# ResNet18 + Decoder with multiscale depth
in_image = Input(shape=inp_dim)
g = Conv2D(f,(7,7),padding='same',strides=(2,2), kernel_initializer='he_normal')(in_image)
if norm:
g = GroupNormalization(groups=min(f,norm))(g)
g = Activation('relu')(g)
g_0=g
g = MaxPooling2D(pool_size=(3,3), strides=(2,2),padding='same')(g)
g = resnet_block(f, g, norm = norm)
g = resnet_block(f, g, norm = norm)
g_1 = g
g = resnet_block(2*f, g, red = True, norm = norm)
g = resnet_block(2*f, g, norm = norm)
g_2 = g
g = resnet_block(4*f, g, red = True, norm = norm)
g = resnet_block(4*f, g, norm = norm)
g_4 = g
g = resnet_block(8*f, g, red = True, norm = norm)
g = resnet_block(8*f, g, norm = norm)
g = up_block(8*f, g, g_4, norm = norm)
g = up_block(4*f, g, g_2, norm = norm)
g = up_block(2*f, g, g_1, norm = norm)
g = up_block(f, g, g_0, norm = norm)
g = UpSampling2D(size=(2,2),interpolation='nearest')(g)
g = Conv2D(f//2, (3,3), strides=(1,1), padding='same', kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups = min(f//2,norm))(g)
g = Activation('elu')(g)
g = Conv2D(f//2, (3,3), strides=(1,1), padding='same', kernel_initializer='he_normal')(g)
if norm:
g = GroupNormalization(groups = min(f//2,norm))(g)
g = Activation('elu')(g)
out_image = Conv2D(out_dim[-1], (3,3), padding='same',strides=(1,1),
activation = out_act, kernel_initializer='he_normal')(g)
model = Model(inputs = in_image, outputs= out_image)
return model