/
test_PixelRNN.py
277 lines (213 loc) · 11.7 KB
/
test_PixelRNN.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
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
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
"""
A Keras implementation of PixelRNN by Van Den Oord et al. Google
arxiv.org/pdf/1601.06759
Copyright Shir Gur, 2016
me@gurshir.com
"""
import theano
import theano.sandbox.cuda
theano.sandbox.cuda.use("gpu0")
from keras.datasets import cifar10
import matplotlib.pyplot as plt
import numpy as np
from keras.callbacks import LearningRateScheduler
from keras.models import Model
from keras.layers import Input, Convolution2D
from keras.engine import merge
from keras.optimizers import RMSprop
from keras.regularizers import l2
from Layers import MaskedConvolution2D, SoftmaxLayer, GetColors, DiagLSTM, PyramidSTM
from Utils import image_categorical_crossentropy
class PixelRNN(object):
def __init__(self):
self.img_rows, self.img_cols = 32, 32
self.img_channels = 3
self.margin = 0
self.batch_size = 10
self.h = 300
self.h_2 = self.h//2
rms = RMSprop(lr=(1e-6), clipnorm=1)
self.optimizer = rms
self.Xtrain = []
self.Rtrain = []
self.Gtrain = []
self.Btrain = []
self.comp_net = None
self.build_data_cifar10()
def build_data_cifar10(self):
print("Loading data...")
(self.Xtrain, _), (_, _) = cifar10.load_data()
self.Xtrain = self.Xtrain[:501]
self.extra = np.shape(self.Xtrain)[0] % self.batch_size
if self.extra > 0:
self.Xtrain = self.Xtrain[:-self.extra]
Ytrain = self.Xtrain.copy()
self.Xtrain[:,:,:,-self.margin:] = 0
RGB_Shapr = (np.shape(self.Xtrain)[0], np.shape(self.Xtrain)[2], np.shape(self.Xtrain)[3], 256)
RGB_Shapr_train = (np.shape(self.Xtrain)[0], np.shape(self.Xtrain)[2]*np.shape(self.Xtrain)[3], 256)
self.Rtrain = np.zeros(RGB_Shapr, dtype=self.Xtrain.dtype)
self.Gtrain = np.zeros(RGB_Shapr, dtype=self.Xtrain.dtype)
self.Btrain = np.zeros(RGB_Shapr, dtype=self.Xtrain.dtype)
for s in range(np.shape(self.Xtrain)[0]):
for j in range(3):
for i in range(self.img_rows):
for k in range(self.img_cols):
if j == 0:
self.Rtrain[s, i, k, Ytrain[s,j,i,k]] = 1
elif j == 1:
self.Gtrain[s, i, k, Ytrain[s,j,i,k]] = 1
else:
self.Btrain[s, i, k, Ytrain[s,j,i,k]] = 1
self.Rtrain = np.reshape(self.Rtrain,RGB_Shapr_train)
self.Gtrain = np.reshape(self.Gtrain,RGB_Shapr_train)
self.Btrain = np.reshape(self.Btrain,RGB_Shapr_train)
def build_net_DiagLSTM(self, load_weights = False):
img = Input(batch_shape=(10, self.img_channels, self.img_rows, self.img_cols), name='input_img')
model_in = MaskedConvolution2D(self.h,7,7,mask_type='a', direction='Right', border_mode='same', init='he_uniform')(img)
for _ in range(12):
model_LSTM_F = DiagLSTM(self.h_2,3, return_sequences=True, init='he_uniform', inner_init='he_uniform', direction='Right')(model_in)
model_LSTM_B = DiagLSTM(self.h_2,3, return_sequences=True, init='he_uniform', inner_init='he_uniform', direction='Right', reverse=True)(model_in)
model_LSTM = merge([model_LSTM_F, model_LSTM_B], mode='sum')
model_per = Convolution2D(self.h,1,1, init='he_normal')(model_LSTM)
model_in = merge([model_in, model_per], mode='sum')
model_out = MaskedConvolution2D(self.h,1,1,mask_type='b', direction='Right', border_mode='same', activation='relu', init='he_uniform')(model_in)
model_out = MaskedConvolution2D(256*3,1,1,mask_type='b', direction='Right', border_mode='same', activation='relu', init='he_uniform')(model_out)
Red = GetColors(0)(model_out)
Green = GetColors(1)(model_out)
Blue = GetColors(2)(model_out)
Red_out = SoftmaxLayer(name='Red_out')(Red)
Green_out = SoftmaxLayer(name='Green_out')(Green)
Blue_out = SoftmaxLayer(name='Blue_out')(Blue)
Col_Model = Model(img, [Red_out, Green_out, Blue_out])
if load_weights:
Col_Model.load_weights('Data/comp_model.h5')
print("Compiling...")
Col_Model.compile(optimizer=self.optimizer,
loss={'Red_out': image_categorical_crossentropy,
'Green_out': image_categorical_crossentropy,
'Blue_out': image_categorical_crossentropy},
metrics={'Red_out': 'accuracy',
'Green_out': 'accuracy',
'Blue_out': 'accuracy'})
self.comp_net = Col_Model
def build_net_PyramidSTM(self, load_weights = False):
img = Input(batch_shape=(10, self.img_channels, self.img_rows, self.img_cols), name='input_img')
model_in = MaskedConvolution2D(self.h,7,7,mask_type='a', direction='Right', border_mode='same', init='he_uniform', W_regularizer=l2(0.0005))(img)
for _ in range(12):
model_LSTM = PyramidSTM(self.h_2,3, return_sequences=True, init='he_uniform', inner_init='he_uniform', direction='Right', W_regularizer=l2(0.0005))(model_in)
model_per = Convolution2D(self.h,1,1, init='he_normal')(model_LSTM)
model_in = merge([model_in, model_per], mode='sum')
model_out = MaskedConvolution2D(self.h,1,1,mask_type='b', direction='Right', border_mode='same', activation='relu', init='he_uniform', W_regularizer=l2(0.0005))(model_in)
model_out = MaskedConvolution2D(256*3,1,1,mask_type='b', direction='Right', border_mode='same', activation='relu', init='he_uniform', W_regularizer=l2(0.0005))(model_out)
Red = GetColors(0)(model_out)
Green = GetColors(1)(model_out)
Blue = GetColors(2)(model_out)
Red_out = SoftmaxLayer(name='Red_out')(Red)
Green_out = SoftmaxLayer(name='Green_out')(Green)
Blue_out = SoftmaxLayer(name='Blue_out')(Blue)
Col_Model = Model(img, [Red_out, Green_out, Blue_out])
if load_weights:
Col_Model.load_weights('Data/comp_model.h5')
print("Compiling...")
Col_Model.compile(optimizer=self.optimizer,
loss={'Red_out': image_categorical_crossentropy,
'Green_out': image_categorical_crossentropy,
'Blue_out': image_categorical_crossentropy},
metrics={'Red_out': 'accuracy',
'Green_out': 'accuracy',
'Blue_out': 'accuracy'})
self.comp_net = Col_Model
def build_net_CNN(self, load_weights = False):
img = Input(batch_shape=(10, self.img_channels, self.img_rows, self.img_cols), name='input_img')
model_in = MaskedConvolution2D(self.h,7,7,mask_type='a', direction='Right', border_mode='same', init='he_uniform')(img)
for _ in range(15):
model_in = MaskedConvolution2D(self.h,3,3,mask_type='b', direction='Right', border_mode='same', init='he_uniform')(model_in)
model_out = MaskedConvolution2D(self.h,1,1,mask_type='b', direction='Right', border_mode='same', activation='relu', init='he_uniform')(model_in)
model_out = MaskedConvolution2D(256*3,1,1,mask_type='b', direction='Right', border_mode='same', activation='relu', init='he_uniform')(model_out)
Red = GetColors(0)(model_out)
Green = GetColors(1)(model_out)
Blue = GetColors(2)(model_out)
Red_out = SoftmaxLayer(name='Red_out')(Red)
Green_out = SoftmaxLayer(name='Green_out')(Green)
Blue_out = SoftmaxLayer(name='Blue_out')(Blue)
Col_Model = Model(img, [Red_out, Green_out, Blue_out])
if load_weights:
Col_Model.load_weights('Data/comp_model.h5')
print("Compiling...")
Col_Model.compile(optimizer=self.optimizer,
loss={'Red_out': image_categorical_crossentropy,
'Green_out': image_categorical_crossentropy,
'Blue_out': image_categorical_crossentropy},
metrics={'Red_out': 'accuracy',
'Green_out': 'accuracy',
'Blue_out': 'accuracy'})
self.comp_net = Col_Model
def fit_net_DiagLSTM(self, batch_size=None, nb_epoch=5):
if self.comp_net is None:
self.build_net_DiagLSTM(load_weights=False)
if batch_size is not None:
self.batch_size = batch_size
# Fit
print("Fitting...")
self.comp_net.fit({'input_img':self.Xtrain[:-self.extra]},
{'Red_out':self.Rtrain[:-self.extra],
'Green_out':self.Gtrain[:-self.extra],
'Blue_out':self.Btrain[:-self.extra]},
batch_size=self.batch_size, nb_epoch=nb_epoch, verbose=1,
callbacks=[LearningRateScheduler(self.scd)])
# Save weights
print("Saving weights...")
self.comp_net.save_weights('Data/comp_model.h5', overwrite=True)
def fit_net_PyramidSTM(self, batch_size=None, nb_epoch=3):
if self.comp_net is None:
self.build_net_PyramidSTM(load_weights=False)
if batch_size is not None:
self.batch_size = batch_size
# Fit
print("Fitting...")
self.comp_net.fit({'input_img':self.Xtrain},
{'Red_out':self.Rtrain,
'Green_out':self.Gtrain,
'Blue_out':self.Btrain},
batch_size=self.batch_size, nb_epoch=nb_epoch, verbose=1,
callbacks=[LearningRateScheduler(self.scd)])
# Save weights
print("Saving weights...")
self.comp_net.save_weights('Data/comp_model.h5', overwrite=True)
def fit_net_CNN(self, batch_size=None, nb_epoch=5):
if self.comp_net is None:
self.build_net_CNN(load_weights=False)
if batch_size is not None:
self.batch_size = batch_size
# Fit
print("Fitting...")
self.comp_net.fit({'input_img':self.Xtrain[:-self.extra]},
{'Red_out':self.Rtrain[:-self.extra],
'Green_out':self.Gtrain[:-self.extra],
'Blue_out':self.Btrain[:-self.extra]},
batch_size=self.batch_size, nb_epoch=nb_epoch, verbose=1,
callbacks=[LearningRateScheduler(self.scd)])
# Save weights
print("Saving weights...")
self.comp_net.save_weights('Data/comp_model.h5', overwrite=True)
def scd(self,x):
if x%5 == 0:
return float(0.1)
if x%3 == 0:
return float(1./x)
return (1e-6)
p = PixelRNN()
p.fit_net_PyramidSTM()
predictions = p.comp_net.predict({'input_img':p.Xtrain[0:p.batch_size]})
result = np.concatenate([np.reshape(np.argmax(np.reshape(
predictions[i][9], (p.img_rows,p.img_cols,256)),axis=-1), (p.img_rows,p.img_cols,1))
.astype(dtype=p.Xtrain.dtype)
for i in range(3)], axis=-1)
train = np.concatenate([np.reshape(np.argmax(np.reshape(p.Rtrain[9], (p.img_rows,p.img_cols,256)),axis=-1), (p.img_rows,p.img_cols,1)).astype(dtype=p.Xtrain.dtype),
np.reshape(np.argmax(np.reshape(p.Gtrain[9], (p.img_rows,p.img_cols,256)),axis=-1), (p.img_rows,p.img_cols,1)).astype(dtype=p.Xtrain.dtype),
np.reshape(np.argmax(np.reshape(p.Btrain[9], (p.img_rows,p.img_cols,256)),axis=-1), (p.img_rows,p.img_cols,1)).astype(dtype=p.Xtrain.dtype)], axis=-1)
plt.figure(1)
plt.imshow(result)
plt.figure(2)
plt.imshow(train)
plt.show()