image-multi-task-learning-/multi_task_mnf.py at master · st70712/image-multi-task-learning- · GitHub

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

# coding: utf-8

from __future__ import print_function
import keras
from keras.datasets import cifar10, mnist
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential, Model
from keras.layers import Dense, Dropout, Activation, Flatten, Input
from keras.layers import Conv2D, MaxPooling2D,GlobalAveragePooling2D,GlobalAveragePooling1D
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.normalization import BatchNormalization
from keras.callbacks import ModelCheckpoint,LearningRateScheduler
from keras.initializers import RandomUniform
from keras.utils import plot_model
import os
import pickle
import numpy as np
import tensorflow as tf
import math

from img_process import *
from lsuv_init import LSUVinit
import mnist_reader
flags = tf.app.flags
flags.DEFINE_integer("epochs", 25, "Epoch to train")
flags.DEFINE_float("learning_rate", 0.001, "Learning rate of for adam[0.001] ")
flags.DEFINE_boolean("load_model", False, "loading model weights")
FLAGS = flags.FLAGS

num_classes = 10
batch_size = 100
cy_train = np.zeros((60000,10))
cy_test = np.zeros((10000,10))

(mx_train, my_train), (mx_test, my_test) = mnist.load_data()
mx_train = np.asarray(mx_train)
mx_test = np.asarray(mx_test)
mx_train = np.reshape(mx_train,[-1,28,28,1])
mx_test = np.reshape(mx_test,[-1,28,28,1])
mx_train = mx_train.astype('float32')
mx_test = mx_test.astype('float32')
mx_train=re_scale(mx_train)
mx_test=re_scale(mx_test)
#mx_train=i2b(mx_train)
#mx_test=i2b(mx_test)#image to binary

my_train = keras.utils.to_categorical(my_train, num_classes)
my_test = keras.utils.to_categorical(my_test, num_classes)
my_train = np.concatenate((my_train,cy_train,cy_train),axis=1)
my_test = np.concatenate((my_test,cy_test,cy_test),axis=1)

fx_train, fy_train = mnist_reader.load_mnist('data/fashion', kind='train')
fx_test, fy_test = mnist_reader.load_mnist('data/fashion', kind='t10k')
fx_train = np.asarray(fx_train)
fx_test = np.asarray(fx_test)
fx_train = np.reshape(fx_train,[-1,28,28,1])
fx_test = np.reshape(fx_test,[-1,28,28,1])
fx_train = fx_train.astype('float32')
fx_test = fx_test.astype('float32')
fx_train=re_scale(fx_train)
fx_test=re_scale(fx_test)
#fx_train=i2b(fx_train)
#fx_test=i2b(fx_test)#image to binary

fy_train = keras.utils.to_categorical(fy_train, num_classes)
fy_test = keras.utils.to_categorical(fy_test, num_classes)
fy_train = np.concatenate((cy_train, fy_train,cy_train),axis=1)
fy_test = np.concatenate((cy_test, fy_test,cy_test),axis=1)

nx_train, ny_train = mnist_reader.load_mnist('data/notMNIST', kind='train')
nx_test, ny_test = mnist_reader.load_mnist('data/notMNIST', kind='t10k')
nx_train = np.asarray(nx_train)
nx_test = np.asarray(nx_test)
nx_train = np.reshape(nx_train,[-1,28,28,1])
nx_test = np.reshape(nx_test,[-1,28,28,1])
nx_train = nx_train.astype('float32')
nx_test = nx_test.astype('float32')
nx_train=re_scale(nx_train)
nx_test=re_scale(nx_test)
#nx_train=i2b(nx_train)
#nx_test=i2b(nx_test)#image to binary

ny_train = keras.utils.to_categorical(ny_train, num_classes)
ny_test = keras.utils.to_categorical(ny_test, num_classes)
ny_train = np.concatenate((cy_train, cy_train,ny_train),axis=1)
ny_test = np.concatenate((cy_test, cy_test,ny_test),axis=1)
kx_train = np.concatenate((mx_train, fx_train,nx_train),axis=0)
ky_train = np.concatenate((my_train, fy_train,ny_train),axis=0)

print('x_train shape:', kx_train.shape)
print('y_train shape:', ky_train.shape)

model = Sequential()
model.add(Conv2D(96, (3, 3), padding='same',input_shape=kx_train.shape[1:]))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Conv2D(96, (3, 3), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Conv2D(96, (3, 3), padding='same',strides=(2,2)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dropout(0.5))

model.add(Conv2D(192, (3, 3), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Conv2D(192, (3, 3), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Conv2D(192, (3, 3), padding='same',strides=(2,2)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dropout(0.5))

model.add(Conv2D(192, (3, 3), padding='valid'))
model.add(BatchNormalization())
model.add(Activation('relu'))

model.add(Conv2D(192, (1, 1), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Conv2D(30, (1, 1), padding='same'))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(GlobalAveragePooling2D(data_format='channels_last'))
model.add(BatchNormalization())
model.add(Activation('softmax'))

decay_rate = FLAGS.learning_rate / FLAGS.epochs
opt = keras.optimizers.Adam(lr=FLAGS.learning_rate,decay=decay_rate)
model.compile(loss='categorical_crossentropy',optimizer=opt,metrics=['accuracy'])

if (FLAGS.load_model):
    model.load_weights('./checkpoint/learning_decay_weights.h5')
model.fit(kx_train,ky_train,batch_size=100,epochs=FLAGS.epochs,validation_data=(fx_test,fy_test),verbose=2)
i = 0
for layer in model.layers:
    weights = layer.get_weights()
    np.save('./layer_weight/weights{:.2f}'.format(i),weights)
    i +=1
if (FLAGS.load_model == False):
    model.save_weights('./checkpoint/learning_decay_weights.h5')
model.save_weights('./checkpoint/learning_decay_weights.h5')