回答編集履歴
1
動作確認コードを追加
test
CHANGED
|
@@ -71,3 +71,99 @@
|
|
|
71
71
|
|
|
72
72
|
|
|
73
73
|
ただ、このように学習前に前処理として画像を増やしておくと、メモリを圧迫しますので、通常は``model.fit_generator()`` を使って学習時にリアルタイムで新規の画像を生成する方法が一般的ではないでしょうか。
|
|
74
|
+
|
|
75
|
+
|
|
76
|
+
|
|
77
|
+
---
|
|
78
|
+
|
|
79
|
+
|
|
80
|
+
|
|
81
|
+
**【動作確認】**
|
|
82
|
+
|
|
83
|
+
|
|
84
|
+
|
|
85
|
+
確認コード
|
|
86
|
+
|
|
87
|
+
```Python
|
|
88
|
+
|
|
89
|
+
from keras.preprocessing.image import ImageDataGenerator
|
|
90
|
+
|
|
91
|
+
import numpy as np
|
|
92
|
+
|
|
93
|
+
|
|
94
|
+
|
|
95
|
+
# ダミーデータを生成
|
|
96
|
+
|
|
97
|
+
X_train = np.random.randint(0, 255, (4200, 1, 28, 28))
|
|
98
|
+
|
|
99
|
+
y_train = np.random.randint(0, 1, (4200, 15))
|
|
100
|
+
|
|
101
|
+
|
|
102
|
+
|
|
103
|
+
datagen = ImageDataGenerator(
|
|
104
|
+
|
|
105
|
+
rotation_range=15,
|
|
106
|
+
|
|
107
|
+
width_shift_range=0.1,
|
|
108
|
+
|
|
109
|
+
height_shift_range=0.1,
|
|
110
|
+
|
|
111
|
+
shear_range=0.2,
|
|
112
|
+
|
|
113
|
+
zoom_range=[0.4, 0],
|
|
114
|
+
|
|
115
|
+
#horizontal_flip=False,
|
|
116
|
+
|
|
117
|
+
#vertical_flip=False
|
|
118
|
+
|
|
119
|
+
)
|
|
120
|
+
|
|
121
|
+
#データ拡張実施
|
|
122
|
+
|
|
123
|
+
X_gen = datagen.flow(X_train, y_train, batch_size=4200)
|
|
124
|
+
|
|
125
|
+
|
|
126
|
+
|
|
127
|
+
for _ in range(10):
|
|
128
|
+
|
|
129
|
+
#データ生成
|
|
130
|
+
|
|
131
|
+
X_gen_new, y_gen_new = X_gen.__next__()
|
|
132
|
+
|
|
133
|
+
#データを結合
|
|
134
|
+
|
|
135
|
+
X_train = np.concatenate([X_train, X_gen_new])
|
|
136
|
+
|
|
137
|
+
y_train = np.concatenate([y_train, y_gen_new])
|
|
138
|
+
|
|
139
|
+
#データサイズを表示
|
|
140
|
+
|
|
141
|
+
print(X_train.shape, y_train.shape)
|
|
142
|
+
|
|
143
|
+
```
|
|
144
|
+
|
|
145
|
+
結果
|
|
146
|
+
|
|
147
|
+
```Text
|
|
148
|
+
|
|
149
|
+
(8400, 1, 28, 28) (8400, 15)
|
|
150
|
+
|
|
151
|
+
(12600, 1, 28, 28) (12600, 15)
|
|
152
|
+
|
|
153
|
+
(16800, 1, 28, 28) (16800, 15)
|
|
154
|
+
|
|
155
|
+
(21000, 1, 28, 28) (21000, 15)
|
|
156
|
+
|
|
157
|
+
(25200, 1, 28, 28) (25200, 15)
|
|
158
|
+
|
|
159
|
+
(29400, 1, 28, 28) (29400, 15)
|
|
160
|
+
|
|
161
|
+
(33600, 1, 28, 28) (33600, 15)
|
|
162
|
+
|
|
163
|
+
(37800, 1, 28, 28) (37800, 15)
|
|
164
|
+
|
|
165
|
+
(42000, 1, 28, 28) (42000, 15)
|
|
166
|
+
|
|
167
|
+
(46200, 1, 28, 28) (46200, 15)
|
|
168
|
+
|
|
169
|
+
```
|