-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmodel.py
197 lines (160 loc) · 6.99 KB
/
model.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
import cv2
import numpy as np
import glob
import random
from keras.models import Model, Sequential
from keras.layers import Input, Activation, Add, Reshape, Permute
from keras.layers import Conv2D, BatchNormalization, MaxPooling2D, UpSampling2D
from keras import backend as K
from keras.callbacks import ModelCheckpoint
from prepareData import Data_Generator, getImageArray
K.set_image_dim_ordering("th") # channel first
def displayImage(real,predicted):
#cv2.imshow("Real",real)
cv2.imshow("Predicted",predicted)
cv2.waitKey(0)
cv2.destroyAllWindows()
def visualize(filename):
test_image = getImageArray(filename)
test_image = np.expand_dims(test_image, axis=0)
predicted_image = model.predict(test_image)[0]
predicted_image = predicted_image.reshape((img_height, img_width, 3)).argmax(axis=2)
seg_img = np.zeros((img_height, img_width, 3))
colors = [(0, 0, 0),(0, 255, 0),(0, 0, 255)]
for c in range(3):
seg_img[:,:,0] += ((predicted_image[:,: ] == c)*(colors[c][0])).astype('uint8')
seg_img[:,:,1] += ((predicted_image[:,: ] == c)*(colors[c][1])).astype('uint8')
seg_img[:,:,2] += ((predicted_image[:,: ] == c)*(colors[c][2])).astype('uint8')
seg_img = cv2.resize(seg_img, (img_width, img_height))
displayImage(test_image,seg_img)
path = "results\\"+filename.split('/')[-1]
cv2.imwrite(path,seg_img)
def residual_block(x):
m = Conv2D(8,(1,1),activation='relu',padding="same")(x)
m = Conv2D(8,(5,1),activation='relu',padding="same")(m)
m = Conv2D(8,(1,5),activation='relu',padding="same")(m)
m = Conv2D(16,(1,1),activation='relu',padding="same")(m)
return Add()([m, x])
img_height = 384
img_width = 512
batch_size = 5
# Load the training data
filenames = glob.glob("rgb_resized/*.png")
filenames.sort() # make sure that the filenames have a fixed order before shuffling
random.seed(230)
random.shuffle(filenames)
split_1 = int(0.8 * len(filenames))
split_2 = int(0.9 * len(filenames))
train_filenames = filenames[:split_1]
validation_filenames = filenames[split_1:split_2]
test_filenames = filenames[split_2:]
gt_training = ["annotation\\"+filename.split('\\')[-1] for filename in train_filenames]
gt_validation = ["annotation\\"+filename.split('\\')[-1] for filename in validation_filenames]
gt_testing = ["groundtruth_resized\\"+filename.split('\\')[-1] for filename in test_filenames]
training_batch_generator = Data_Generator(train_filenames, gt_training, batch_size)
validation_batch_generator = Data_Generator(validation_filenames, gt_validation, batch_size)
# Model
input_shape = (14,img_height,img_width)
i = Input(shape=input_shape)
# Encoder
conv_1 = Conv2D(16,(5,5),padding="same")(i)
conv_1 = BatchNormalization()(conv_1)
conv_1 = Activation('relu')(conv_1)
rb_1 = residual_block(conv_1)
rb_2 = residual_block(rb_1)
rb_3 = residual_block(rb_2)
pool_1 = MaxPooling2D(pool_size=(2, 2))(rb_3)
rb_4 = residual_block(pool_1)
rb_5 = residual_block(rb_4)
rb_6 = residual_block(rb_5)
pool_2 = MaxPooling2D(pool_size=(2, 2))(rb_6)
rb_7 = residual_block(pool_2)
rb_8 = residual_block(rb_7)
rb_9 = residual_block(rb_8)
pool_3 = MaxPooling2D(pool_size=(2, 2))(rb_9)
rb_10 = residual_block(pool_3)
rb_11 = residual_block(rb_10)
rb_12 = residual_block(rb_11)
pool_4 = MaxPooling2D(pool_size=(2, 2))(rb_12)
# Decoder
upsamp_1 = UpSampling2D(size=(2,2))(pool_4)
rb_13 = residual_block(upsamp_1)
rb_14 = residual_block(rb_13)
rb_15 = residual_block(rb_14)
upsamp_2 = UpSampling2D(size=(2,2))(rb_15)
rb_16 = residual_block(upsamp_2)
rb_17 = residual_block(rb_16)
rb_18 = residual_block(rb_17)
upsamp_3 = UpSampling2D(size=(2,2))(rb_18)
rb_19 = residual_block(upsamp_3)
rb_20 = residual_block(rb_19)
rb_21 = residual_block(rb_20)
upsamp_4 = UpSampling2D(size=(2,2))(rb_21)
rb_22 = residual_block(upsamp_4)
rb_23 = residual_block(rb_22)
rb_24 = residual_block(rb_23)
conv_2 = Conv2D(3,(1,1),padding="same")(rb_24)
conv_2 = Reshape((3,img_height*img_width), input_shape=(3,img_width,img_height))(conv_2)
conv_2 = Permute((2, 1))(conv_2)
j = Activation('softmax')(conv_2)
model = Model(inputs=i, outputs=j)
model.compile(optimizer='sgd',loss='categorical_crossentropy',metrics=['accuracy'])
checkpoint = ModelCheckpoint('model.h5',verbose = 1,monitor = 'acc',save_best_only = True,mode = "auto")
model.fit_generator(generator=training_batch_generator,
steps_per_epoch=(len(train_filenames) // batch_size),
epochs=1,
verbose=1,
validation_data=validation_batch_generator,
validation_steps=(len(validation_filenames) // batch_size),callbacks = [checkpoint])
#model.save_weights('model.h5')
#model.load_weights('model.h5')
# Evaluating
x_test, y_test = next(load_data("test/" , "test_masks/" , 21, 3 , img_height , img_width , img_height , img_width))
score = model.evaluate(x_test[:10],y_test[:10])
visualize(getImageArray(test_filenames[15]))
# Testing Data
filenames = glob.glob("test_annotations/*.png")
filenames.sort()
test_annotations = [cv2.imread(img) for img in filenames]
test_annotations = [cv2.resize(img,(img_width,img_height),interpolation = cv2.INTER_NEAREST) for img in test_annotations]
# Segnet Model
'''
kernel = 3
filter_size = 64
pad = 1
pool_size = 2
model = Sequential()
# encoder
model.add(Conv2D(filter_size, (kernel, kernel), padding = 'same', input_shape=(14, img_height , img_width)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(pool_size, pool_size)))
model.add(Conv2D(128, (kernel, kernel), padding = 'same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(pool_size, pool_size)))
model.add(Conv2D(256, (kernel, kernel), padding = 'same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(pool_size, pool_size)))
model.add(Conv2D(512, (kernel, kernel), padding = 'same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
# decoder
model.add(Conv2D(512, (kernel, kernel), padding = 'same'))
model.add( BatchNormalization())
model.add(UpSampling2D(size=(pool_size,pool_size)))
model.add(Conv2D(256, (kernel, kernel), padding = 'same'))
model.add(BatchNormalization())
model.add(UpSampling2D(size=(pool_size,pool_size)))
model.add(Conv2D(128, (kernel, kernel), padding = 'same'))
model.add(BatchNormalization())
model.add(UpSampling2D(size=(pool_size,pool_size)))
model.add(Conv2D(filter_size, (kernel, kernel), padding = 'same'))
model.add(BatchNormalization())
model.add(Conv2D(3, (1, 1), padding = 'same'))
model.add(Reshape((3, model.output_shape[-2]*model.output_shape[-1]), input_shape=(3, model.output_shape[-2], model.output_shape[-1])))
model.add(Permute((2, 1)))
model.add(Activation('softmax'))
model.compile(loss="categorical_crossentropy", optimizer= 'sgd' , metrics=['accuracy'])
'''