質問編集履歴
2
参考サイトを追記しました
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -1,5 +1,7 @@
|
|
|
1
1
|
下記サイトを参考に、kerasでpix2pixが動くようにしました。
|
|
2
2
|
|
|
3
|
+
https://qiita.com/mine820/items/36ffc3c0aea0b98027fd
|
|
4
|
+
|
|
3
5
|
タイトルの通り、学習済みのgeneratorモデルを保存し、トレーニング画像以外のテスト画像で変換を行いたいと考えております。
|
|
4
6
|
|
|
5
7
|
def train()のブロックの最下段にmodel.saveを追記しました。
|
1
コード情報の追記
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -168,6 +168,168 @@
|
|
|
168
168
|
|
|
169
169
|
|
|
170
170
|
|
|
171
|
-
generator_model.save('C:/フォルダーの場所/
|
|
171
|
+
generator_model.save('C:/フォルダーの場所/model.h5')
|
|
172
172
|
|
|
173
173
|
```
|
|
174
|
+
|
|
175
|
+
|
|
176
|
+
|
|
177
|
+
以下、追記
|
|
178
|
+
|
|
179
|
+
「保存されたモデル(.h5)をロードして画像変換」は下記になります。
|
|
180
|
+
|
|
181
|
+
```python
|
|
182
|
+
|
|
183
|
+
import keras
|
|
184
|
+
|
|
185
|
+
from keras import optimizers
|
|
186
|
+
|
|
187
|
+
from matplotlib import pyplot as plt
|
|
188
|
+
|
|
189
|
+
from keras.preprocessing.image import ImageDataGenerator
|
|
190
|
+
|
|
191
|
+
from keras.models import Sequential, Model
|
|
192
|
+
|
|
193
|
+
from keras.layers import Dense, Dropout, Activation, Flatten
|
|
194
|
+
|
|
195
|
+
from keras.layers import Conv2D, MaxPooling2D, Add, Input, Multiply
|
|
196
|
+
|
|
197
|
+
from keras.callbacks import ModelCheckpoint, CSVLogger
|
|
198
|
+
|
|
199
|
+
from keras import optimizers
|
|
200
|
+
|
|
201
|
+
from keras.layers import GlobalAveragePooling2D
|
|
202
|
+
|
|
203
|
+
import cv2
|
|
204
|
+
|
|
205
|
+
import os, glob, random
|
|
206
|
+
|
|
207
|
+
import os, tkinter, tkinter.filedialog, tkinter.messagebox
|
|
208
|
+
|
|
209
|
+
from PIL import Image
|
|
210
|
+
|
|
211
|
+
import pickle
|
|
212
|
+
|
|
213
|
+
import numpy as np
|
|
214
|
+
|
|
215
|
+
import time
|
|
216
|
+
|
|
217
|
+
import datetime
|
|
218
|
+
|
|
219
|
+
import random
|
|
220
|
+
|
|
221
|
+
import keras.backend as K
|
|
222
|
+
|
|
223
|
+
|
|
224
|
+
|
|
225
|
+
def psnr(y_true, y_pred):
|
|
226
|
+
|
|
227
|
+
return -10*K.log(K.mean(K.flatten((y_true - y_pred))**2)) / np.log(10)
|
|
228
|
+
|
|
229
|
+
|
|
230
|
+
|
|
231
|
+
random.seed(0)
|
|
232
|
+
|
|
233
|
+
|
|
234
|
+
|
|
235
|
+
img_width = 256
|
|
236
|
+
|
|
237
|
+
img_height = 256
|
|
238
|
+
|
|
239
|
+
|
|
240
|
+
|
|
241
|
+
print('学習済みモデルを指定')
|
|
242
|
+
|
|
243
|
+
root = tkinter.Tk()
|
|
244
|
+
|
|
245
|
+
root.withdraw()
|
|
246
|
+
|
|
247
|
+
fTyp = [("","*")]
|
|
248
|
+
|
|
249
|
+
iDir = os.path.abspath(os.path.dirname(__file__))
|
|
250
|
+
|
|
251
|
+
file_path = tkinter.filedialog.askopenfilename(filetypes = fTyp,initialdir = iDir)
|
|
252
|
+
|
|
253
|
+
print(file_path)
|
|
254
|
+
|
|
255
|
+
|
|
256
|
+
|
|
257
|
+
from keras.models import load_model
|
|
258
|
+
|
|
259
|
+
model = load_model(file_path, custom_objects={'psnr': psnr, 'val_psnr': psnr})
|
|
260
|
+
|
|
261
|
+
model.summary()
|
|
262
|
+
|
|
263
|
+
|
|
264
|
+
|
|
265
|
+
|
|
266
|
+
|
|
267
|
+
print('データセットのフォルダを指定')
|
|
268
|
+
|
|
269
|
+
root = tkinter.Tk()
|
|
270
|
+
|
|
271
|
+
root.withdraw()
|
|
272
|
+
|
|
273
|
+
fTyp = [("","*")]
|
|
274
|
+
|
|
275
|
+
iDir = os.path.abspath(os.path.dirname(__file__))
|
|
276
|
+
|
|
277
|
+
file_path = tkinter.filedialog.askdirectory(initialdir = iDir)
|
|
278
|
+
|
|
279
|
+
print(file_path+'/*.jpg')
|
|
280
|
+
|
|
281
|
+
|
|
282
|
+
|
|
283
|
+
x_val = []
|
|
284
|
+
|
|
285
|
+
fn=glob.glob(file_path+'/*.jpg')
|
|
286
|
+
|
|
287
|
+
for nm in fn:
|
|
288
|
+
|
|
289
|
+
img = cv2.imread(nm)
|
|
290
|
+
|
|
291
|
+
x_val.append(img)
|
|
292
|
+
|
|
293
|
+
x_val = np.asarray(x_val)
|
|
294
|
+
|
|
295
|
+
x_val = x_val.astype('float32')
|
|
296
|
+
|
|
297
|
+
x_val = x_val / 255.0
|
|
298
|
+
|
|
299
|
+
#x_val = x_val.reshape((len(x_val), 256, 256, 3))
|
|
300
|
+
|
|
301
|
+
|
|
302
|
+
|
|
303
|
+
|
|
304
|
+
|
|
305
|
+
decoded_imgs = model.predict(x_val)
|
|
306
|
+
|
|
307
|
+
|
|
308
|
+
|
|
309
|
+
if not os.path.exists('C:/フォルダの場所'):
|
|
310
|
+
|
|
311
|
+
os.makedirs('C:/フォルダの場所')
|
|
312
|
+
|
|
313
|
+
|
|
314
|
+
|
|
315
|
+
i = -1
|
|
316
|
+
|
|
317
|
+
for nm in fn:
|
|
318
|
+
|
|
319
|
+
nm = nm.split('\')[1]
|
|
320
|
+
|
|
321
|
+
i = i + 1
|
|
322
|
+
|
|
323
|
+
ax = decoded_imgs[i].reshape(img_height, img_width, 3)
|
|
324
|
+
|
|
325
|
+
cv2.imwrite('C:/フォルダの場所/'+nm,ax*255)
|
|
326
|
+
|
|
327
|
+
|
|
328
|
+
|
|
329
|
+
print('結果が保存')
|
|
330
|
+
|
|
331
|
+
|
|
332
|
+
|
|
333
|
+
|
|
334
|
+
|
|
335
|
+
```
|