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
| #-*-coding:cp1251-*-
from keras import models
from keras.layers import Input, Convolution2D, MaxPooling2D, Dense, Dropout, Flatten
from keras.models import Model
from keras.preprocessing.image import ImageDataGenerator
import pickle
import theano
theano.config.optimizer="None" # чтобы работали свертки theano (не через numpy и нет g++ у меня)
import os
from PIL import Image
import numpy as np
"""
->
На google colab (c tensorflow) такой код с генератароми не сработает - там баг, что-то
про steps_per_epoch, batch_size, epochs
"""
"""
Общая структура проекта:
D:.
. app.py
. learned.pkl
. learn_log.txt
.
.....vscode
. launch.json
. settings.json
.
....data
. ....train
. . ....krug
. . . krug1.png
. . .
. . ....kvadr
. . . kvadr1.png
. . .
. . ....treug
. . treug1.png
. .
. ....val
. ....krug
. . krug1.png
. .
. ....kvadr
. . kvadr1.png
. .
. ....treug
. treug1.png
.
....test_folder
krug1.png
kvadr1.png
treug1.png
"""
# параметры обучения
nbatch_size = 32
nepochs = 10
nsteps_per_epoch=5
nvalidation_steps=5
# /параметры обучения
kernel_size = 3
pool_size = 2
conv_depth_1 = 8
hidden_size = 480
height=64
width=64
depth=3
num_classes=3
drop_prob_6=0.3
fpickle_learned='learned.pkl'
image_directory = 'test_folder/' # для спрашивания
# для google colab
# from google.colab import drive
# drive.mount('/content/gdrive')
# fold_images_train = '/content/gdrive/MyDrive/Colab Notebooks/data'
# fold_images_val = '/content/gdrive/MyDrive/Colab Notebooks/val'
# /для google colab
# для компа
fold_images_train = './data/train'
fold_images_val = './data/val'
# /для компа
def save_model(model):
"""
->
Пользуюсь таким так как на google colab всегда tensorfow,
у меня theano для win32
"""
len_layers=len(model.layers)
weights={}
for i in range(len_layers):
weights[i]=model.layers[i].get_weights()
with open(fpickle_learned, 'wb') as f:
pickle.dump(weights, f)
print(f'saved learned to {fpickle_learned}')
def get_model():
inp = Input(shape=(height, width, depth))
conv_1 = Convolution2D(conv_depth_1, (kernel_size, kernel_size), padding='valid', activation='relu')(inp)
pool_2 = MaxPooling2D(pool_size=(pool_size, pool_size), padding='valid')(conv_1)
flat = Flatten()(pool_2)
hidden = Dense(hidden_size, activation='relu')(flat)
# drop_6 = Dropout(drop_prob_6)(hidden) # Не знаю как восстанавливать такое после сохранения весов
out = Dense(num_classes, activation='softmax')(hidden)
model = Model(inputs=inp, outputs=out)
return model
def from_file(file=None):
model= get_model()
if file:
params = {}
with open(file, 'rb') as f:
params = pickle.load(f)
len_layers = len(model.layers)
for i in range(len_layers):
model.layers[i].set_weights(params[i])
return model
def learn():
train_datagen = ImageDataGenerator(
rotation_range=45, #Random rotation between 0 and 45
width_shift_range=0.2, #% shift
height_shift_range=0.2,
rescale=1./255, # изменение масштаба всех значений пикселей с 0 до 255, после этого шага они будут находится в диапазоне (0,1)
shear_range=0.2, # применение случайных преобразований
zoom_range=0.2, # увеличение масштаба
horizontal_flip=True,
fill_mode='constant', cval=125) # Заполнение (цветом) после уменьшения изображения областей - Also try nearest, constant, reflect, wrap) # горизонтальный поворот
test_datagen = ImageDataGenerator(rescale=1./255)
training_set = train_datagen.flow_from_directory(
fold_images_train,
target_size=(64, 64),
batch_size=nbatch_size,
class_mode='categorical',
shuffle = False)
test_set = test_datagen.flow_from_directory(
fold_images_val,
target_size=(64, 64),
batch_size=nbatch_size,
class_mode='categorical',
shuffle = False )
model=get_model()
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy']) # reporting the accuracy
try: # Имеем возможность прервать обучение и сохранить веса
model.fit_generator(
training_set,
steps_per_epoch=nsteps_per_epoch,
epochs=nepochs,
validation_data=test_set,
validation_steps=nvalidation_steps,
)
except KeyboardInterrupt:
save_model(model)
model.evaluate_generator(test_datagen, steps=5, verbose=1) # Evaluate the trained model on the test set!
finally:
save_model(model)
model.evaluate_generator(test_set, steps=5, verbose=1)
def ask():
model=from_file(fpickle_learned)
my_images = os.listdir(image_directory)
for i, image_name in enumerate(my_images):
if (image_name.split('.')[1] == 'png'):
print(f'file {image_name} and class {i}', ' ')
image = Image.open(image_directory + image_name)
image = image.resize((height, width)) # не обязательно 64x64px должно быть, ресайзим
image=np.array(image)
image = Image.fromarray(image, 'RGB') # читаем как 3d
x=np.array(image) # 3d
x=x / 255
print(model.predict(np.array([x]))) # требует как 4d
def main():
# learn()
ask()
main() |
