-
Notifications
You must be signed in to change notification settings - Fork 0
/
jigsaw_multitask.py
141 lines (126 loc) · 5.94 KB
/
jigsaw_multitask.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
import os
import numpy as np
import tensorflow as tf
from cv2 import imread
from sklearn.model_selection import train_test_split
def oneHot(num):
return [1 if int(num) == (i+1) else 0 for i in range(9)]
def readDir(path="./imgs", info=True):
features, labels = [], []
files = os.listdir(path)
if info:
print(f"Sao {len(files)} arquivos a serem lidos!")
i = 0
for img_name in files:
features.append(imread(os.path.join(path, img_name)))
img_name = os.path.splitext(img_name)[0]
labels.append(list(map(oneHot, img_name.split("=")[1].split("-"))))
if info:
if (i%100 == 0):
print(f"Estamos no arquivo numero {i}!")
i+=1
return np.array(features), np.array(labels)
def divideLabels(labels, info=True):
o1, o2, o3, o4, o5, o6, o7, o8, o9 = [], [], [], [], [], [], [], [], []
if info:
print(f"\n\n\nSao {len(labels)} labels a serem subdividas!")
i = 0
for label in labels: # Cada `label` eh um array contendo os one-hot encoded vectors
o1.append(label[0])
o2.append(label[1])
o3.append(label[2])
o4.append(label[3])
o5.append(label[4])
o6.append(label[5])
o7.append(label[6])
o8.append(label[7])
o9.append(label[8])
if info:
if (i%100 == 0):
print(f"Estamos no arquivo numero {i}!")
i+=1
return np.array(o1), np.array(o2), np.array(o3), np.array(o4), np.array(o5), np.array(o6), np.array(o7), np.array(o8), np.array(o9)
def create_branch(base_model, name):
return tf.keras.layers.Dense(9, activation='softmax', name=name)(base_model)
if __name__ == "__main__":
# Carregando o dataset
features, labels = readDir()
x_train, x_test, y_train, y_test = train_test_split(features, labels, test_size=0.33, random_state=23, shuffle=True)
y_train_branch1, y_train_branch2, y_train_branch3, y_train_branch4, y_train_branch5, y_train_branch6, y_train_branch7, y_train_branch8, y_train_branch9 = divideLabels(y_train)
y_test_branch1, y_test_branch2, y_test_branch3, y_test_branch4, y_test_branch5, y_test_branch6, y_test_branch7, y_test_branch8, y_test_branch9 = divideLabels(y_test)
# Neural Net em si
# Modelo 1 --> Soprado da documentação do keras
inputs = tf.keras.layers.Input(shape=(300,300,3))
pre_treinada_saida = inputs
main_branch = tf.keras.applications.EfficientNetB0(weights="imagenet", input_tensor=pre_treinada_saida, include_top=False)
main_branch.trainable = True # O modelo base treina?
main_branch = tf.keras.layers.GlobalAveragePooling2D()(main_branch.output)
main_branch = tf.keras.layers.BatchNormalization()(main_branch)
main_branch = tf.keras.layers.Dense(1024, activation='relu')(main_branch)
main_branch = tf.keras.layers.Dense(512, activation='relu')(main_branch)
main_branch = tf.keras.layers.Dense(256, activation='relu')(main_branch)
main_branch = tf.keras.layers.Dense(128, activation='relu')(main_branch)
out1 = create_branch(main_branch, 'out1')
out2 = create_branch(main_branch, 'out2')
out3 = create_branch(main_branch, 'out3')
out4 = create_branch(main_branch, 'out4')
out5 = create_branch(main_branch, 'out5')
out6 = create_branch(main_branch, 'out6')
out7 = create_branch(main_branch, 'out7')
out8 = create_branch(main_branch, 'out8')
out9 = create_branch(main_branch, 'out9')
# Construção do modelo
neuralNet = tf.keras.Model(inputs, outputs=[out1, out2, out3,
out4, out5, out6,
out7, out8, out9])
#neuralNet.summary() # Resumo da rede
neuralNet.compile(loss={
'out1' : 'categorical_crossentropy',
'out2' : 'categorical_crossentropy',
'out3' : 'categorical_crossentropy',
'out4' : 'categorical_crossentropy',
'out5' : 'categorical_crossentropy',
'out6' : 'categorical_crossentropy',
'out7' : 'categorical_crossentropy',
'out8' : 'categorical_crossentropy',
'out9' : 'categorical_crossentropy'
},
loss_weights={
'out1' : 0.11,
'out2' : 0.11,
'out3' : 0.11,
'out4' : 0.11,
'out5' : 0.11,
'out6' : 0.11,
'out7' : 0.11,
'out8' : 0.11,
'out9' : 0.11
},
optimizer=tf.keras.optimizers.Adam(),
metrics=["accuracy"])
neuralNet.fit(x_train,
{
'out1' : y_train_branch1,
'out2' : y_train_branch2,
'out3' : y_train_branch3,
'out4' : y_train_branch4,
'out5' : y_train_branch5,
'out6' : y_train_branch6,
'out7' : y_train_branch7,
'out8' : y_train_branch8,
'out9' : y_train_branch9,
},
epochs=40,
validation_data=(x_test, {
'out1' : y_test_branch1,
'out2' : y_test_branch2,
'out3' : y_test_branch3,
'out4' : y_test_branch4,
'out5' : y_test_branch5,
'out6' : y_test_branch6,
'out7' : y_test_branch7,
'out8' : y_test_branch8,
'out9' : y_test_branch9,
}),
verbose=1)
neuralNet.save("./contact-lens-model-multitask_b0")