-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathkeras_lstm.py
More file actions
140 lines (116 loc) · 5.91 KB
/
Copy pathkeras_lstm.py
File metadata and controls
140 lines (116 loc) · 5.91 KB
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
import os
import numpy as np
from definitions import *
from keras import backend
from keras.models import Sequential
from keras.models import Model, load_model
from keras.layers import Dense, Lambda
from keras.layers import Input
from keras.layers import ConvLSTM2D
from keras.layers import Conv2D
from keras.backend import squeeze
from keras.layers import Conv2DTranspose
from keras.layers import concatenate
from keras.layers.normalization import BatchNormalization
from keras.layers import Permute
from keras.layers import MaxPooling2D
from keras.layers import LSTM
from keras.layers.embeddings import Embedding
from keras.preprocessing import sequence
from keras.callbacks import TensorBoard
from keras.callbacks import ModelCheckpoint
# fix random seed for reproducibility
launch=os.getcwd()
os.environ["CUDA_VISIBLE_DEVICES"]="1"
NUM_CHANNELS=5
BATCH_SIZE=16
IMG_HEIGHT=256
IMG_WIDTH=256
FILTERS=16
KERNEL_SIZE=3
training=Dataset2D('/home/petmri/training/',NUM_CHANNELS)
validation=Dataset2D('/home/petmri/Testing/',NUM_CHANNELS)
os.chdir(launch)
#NOTE THAT 'CHANNELS' IS SOMEWHAT MISLEADING IN THIS CODE SINCE WE ARE TALKING ABOUT TIME FRAMES NOT CHANNELS IN THE CONVENTIONAL SENSE OF THE WORD
#timeframe,row,column,channel
inputs = Input((NUM_CHANNELS,1,IMG_HEIGHT, IMG_WIDTH))
print(inputs.shape)
#print(inputs.__dict__)
squeezed = Lambda (lambda x: backend.squeeze(x,2))(inputs)
print(squeezed.shape)
l1 = ConvLSTM2D(FILTERS,KERNEL_SIZE,data_format='channels_first',padding='same',return_sequences=False) (inputs)
print(l1.shape)
merged = concatenate([squeezed,l1],axis=1)
#concat input with l1 output
#c1 = Conv2D(16, (1,1),data_format='channels_first',padding='same') (l1)
#c2 = Conv2D(16, (1,1),data_format='channels_first', padding='same') (c1)
#c3 = Conv2D(16, (1,1),data_format='channels_first', padding='same') (c2)
#c4 = Conv2D(16, (1,1),data_format='channels_first', padding='same') (c3)
c1 = Conv2D(64, (3, 3), activation='relu', data_format='channels_first', padding='same') (l1)
c1 = BatchNormalization(axis=1) (c1)
c1 = Conv2D(64, (3, 3), activation='relu', data_format='channels_first', padding='same') (c1)
c1 = BatchNormalization(axis=1) (c1)
p1 = MaxPooling2D((2, 2),data_format='channels_first') (c1)
print(p1.shape)
c2 = Conv2D(128, (3, 3), activation='relu', data_format='channels_first', padding='same') (p1)
c2 = BatchNormalization(axis=1) (c2)
c2 = Conv2D(128, (3, 3), activation='relu', data_format='channels_first', padding='same') (c2)
c2 = BatchNormalization(axis=1) (c2)
p2 = MaxPooling2D((2, 2),data_format='channels_first') (c2)
print(p2.shape)
c3 = Conv2D(256, (3, 3), activation='relu', data_format='channels_first', padding='same') (p2)
c3 = BatchNormalization(axis=1) (c3)
c3 = Conv2D(256, (3, 3), activation='relu', data_format='channels_first', padding='same') (c3)
c3 = BatchNormalization(axis=1) (c3)
p3 = MaxPooling2D((2, 2),data_format='channels_first') (c3)
print(p3.shape)
c4 = Conv2D(512, (3, 3), activation='relu', data_format='channels_first', padding='same') (p3)
c4 = BatchNormalization(axis=1) (c4)
c4 = Conv2D(512, (3, 3), activation='relu', data_format='channels_first', padding='same') (c4)
c4 = BatchNormalization(axis=1) (c4)
p4 = MaxPooling2D(pool_size=(2, 2),data_format='channels_first') (c4)
print(p4.shape)
c5 = Conv2D(1024, (3, 3), activation='relu', data_format='channels_first', padding='same') (p4)
c5 = BatchNormalization(axis=1) (c5)
c5 = Conv2D(1024, (3, 3), activation='relu', data_format='channels_first', padding='same') (c5)
c5 = BatchNormalization(axis=1) (c5)
print(c5.shape)
u6 = Conv2DTranspose(512, (2, 2), strides=(2, 2), data_format='channels_first', padding='same') (c5)
u6 = concatenate([u6, c4], axis=1)
c6 = Conv2D(512, (3, 3), activation='relu', data_format='channels_first', padding='same') (u6)
c6 = BatchNormalization(axis=1) (c6)
c6 = Conv2D(512, (3, 3), activation='relu', data_format='channels_first', padding='same') (c6)
c6 = BatchNormalization(axis=1) (c6)
print(c6.shape)
u7 = Conv2DTranspose(256, (2, 2), strides=(2, 2), data_format='channels_first', padding='same') (c6)
u7 = concatenate([u7, c3],axis=1)
c7 = Conv2D(256, (3, 3), activation='relu', data_format='channels_first', padding='same') (u7)
c7 = BatchNormalization(axis=1) (c7)
c7 = Conv2D(256, (3, 3), activation='relu', data_format='channels_first', padding='same') (c7)
c7 = BatchNormalization(axis=1) (c7)
print(c7.shape)
u8 = Conv2DTranspose(128, (2, 2), strides=(2, 2), data_format='channels_first', padding='same') (c7)
u8 = concatenate([u8, c2],axis=1)
c8 = Conv2D(128, (3, 3), activation='relu', data_format='channels_first', padding='same') (u8)
c8 = BatchNormalization(axis=1) (c8)
c8 = Conv2D(128, (3, 3), activation='relu', data_format='channels_first', padding='same') (c8)
c8 = BatchNormalization(axis=1) (c8)
print(c8.shape)
u9 = Conv2DTranspose(64, (2, 2), strides=(2, 2), data_format='channels_first', padding='same') (c8)
u9 = concatenate([u9, c1],axis=1)
c9 = Conv2D(64, (3, 3), activation='relu', data_format='channels_first', padding='same') (u9)
c9 = BatchNormalization(axis=1) (c9)
c9 = Conv2D(64, (3, 3), activation='relu', data_format='channels_first', padding='same') (c9)
c9 = BatchNormalization(axis=1) (c9)
print(c9.shape)
outputs = Conv2D(1, (1, 1), data_format='channels_first', activation='sigmoid') (c9)
#merged = concatenate([inputs, outputs])
model = Model(inputs=[inputs], outputs=[outputs])
model.compile(optimizer='adam', loss='binary_crossentropy')
model.summary()
tensorboard = TensorBoard(log_dir='./logs', histogram_freq=0,
write_graph=True, write_images=False)
tensorboard.set_model(model)
checkpoint = ModelCheckpoint('logs/simple_lstm.h5', verbose=1, monitor='val_loss',save_best_only=True, mode='auto')
checkpoint.set_model(model)
results = model.fit_generator(generator=training.lstm_generator_keras(BATCH_SIZE), epochs=1000, validation_data=validation.lstm_generator_keras(BATCH_SIZE), use_multiprocessing=True, steps_per_epoch=10, validation_steps=5,callbacks=[tensorboard, checkpoint])