-
Notifications
You must be signed in to change notification settings - Fork 0
/
nn_infer.py
209 lines (164 loc) · 7.66 KB
/
nn_infer.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
import h5py
import tensorflow.keras as keras
import tensorflow as tf
import numpy as np
import pickle
import argparse
import util
np.random.seed(1)
tf.random.set_seed(1)
import PIL, PIL.Image
WORK_PATH = './infer-task/'
EPOCHS = 60
"""
exp3
run command: python nn_infer.py -i fonts-jpn-all -e 1 -t -dir inf-exp-3 -w ./gen-task/fonts-all-2908_exp2_d2019-12-01_2h-3m_weights
add -p to predict only
python nn_infer.py -i fonts-system -e 0 -p -dir inf-sysfonts-task -w ./infer-task/fonts-jpn-all_exp1_d2019-12-01_20h-25m_weights
add -s to save weights
"""
def generate_task(infile, experiment, run_test_set, save_weights, weightfile, tmp_dir, predict_only):
"""
Helper functions
"""
namespace = util.namespace(infile, experiment)
def display_picture(arr, name):
"""
Displays 46 hiragana.
"""
img = PIL.Image.fromarray(np.hstack([arr[idx] for idx in range(46)]))
if img.mode != 'L':
img = img.convert('L')
# img.show() # Debug (disable when running)
img.save('./{}/{}{}.png'.format(tmp_dir, namespace, name))
def display_basis(arr, name):
"""
Displays basis letters.
"""
img = PIL.Image.fromarray(np.hstack([arr[idx] for idx in range(4)]))
if img.mode != 'L':
img = img.convert('L')
# img.show() # Debug (disable when running)
img.save('./{}/{}{}.png'.format(tmp_dir, namespace, name))
def dump_history(history):
with open('{}/{}history.pickle'.format(WORK_PATH, namespace), 'wb') as f:
pickle.dump(history, f)
print('Dumped history.')
"""
Model definition: tower network on basis letters to get autoencoding, feed
through shared dense layers to retrieve all 26 letters (caps)
"""
# Input
X = keras.layers.Input(shape=(4, 64, 64), name='input')
# Lambda layers pull out the 4 basis characters
x1 = keras.layers.Lambda(lambda x: x[:, 0, :, :], output_shape=(64, 64), name='x1')(X)
x2 = keras.layers.Lambda(lambda x: x[:, 1, :, :], output_shape=(64, 64), name='x2')(X)
x3 = keras.layers.Lambda(lambda x: x[:, 2, :, :], output_shape=(64, 64), name='x3')(X)
x4 = keras.layers.Lambda(lambda x: x[:, 3, :, :], output_shape=(64, 64), name='x4')(X)
# Flatten the images into 64 * 64 dimensional vectors
x1 = keras.layers.Flatten()(x1)
x2 = keras.layers.Flatten()(x2)
x3 = keras.layers.Flatten()(x3)
x4 = keras.layers.Flatten()(x4)
# The towers consist of a fully connected layer
neurons = 32
x1 = keras.layers.Dense(neurons, activation='relu')(x1)
x2 = keras.layers.Dense(neurons, activation='relu')(x2)
x3 = keras.layers.Dense(neurons, activation='relu')(x3)
x4 = keras.layers.Dense(neurons, activation='relu')(x4)
# Concatenates the towers together and feed through fully connected layers
added = keras.layers.Concatenate()([x1, x2, x3, x4])
neurons = 400
num_chars = 46
fc = keras.layers.Dense(neurons, activation='relu')(added)
fc = keras.layers.Dense(neurons, activation='relu')(fc)
if (experiment == 1):
fc = keras.layers.Dense(26 * 64 * 64, activation='relu')(fc) # Use same layer as in original gen task
fc = keras.layers.Dense(num_chars * 64 * 64, activation='relu', name='hiragana_dense')(fc)
# Reshape for 2D convolution and upsample
fc = keras.layers.Reshape((num_chars, 64, 64))(fc)
fc = keras.layers.Conv2DTranspose(num_chars, data_format='channels_first', kernel_size=(4, 4), padding='same', activation='relu', name='hiragana_transpose')(fc)
out = fc
model = keras.models.Model(inputs=X, outputs=out)
model.compile(optimizer='adam', loss='mean_squared_error', metrics=[])
model.summary()
# Plotting
# keras.utils.plot_model(model, to_file='model.png')
# exit()
if weightfile is not None:
print('Loading weightfile...')
model.load_weights('{}.hdf5'.format(weightfile), by_name=True)
if predict_only:
"""
Does just prediciton on the data.
"""
train = h5py.File('./infer-task-dsets/infer-task-{}.hdf5'.format(infile), 'r')
basis = train['basis'][:]
predictions = model.predict(basis)
for i, (p, b) in enumerate(zip(predictions, basis)):
display_picture(p, 'predict{}'.format(i))
display_basis(b, 'basis{}'.format(i))
exit()
# Open and prepare training set
train = h5py.File('./infer-task-dsets/infer-task-{}-train.hdf5'.format(infile), 'r')
outputs = train['outputs'][:]
basis = train['basis'][:]
print('Training model on {} fonts...'.format(train['basis'].shape[0]))
class ImageHistory(keras.callbacks.Callback):
"""
Runs predict on the model and test set to visualize how the
NN is learning. In a Keras callback, we have access to model
and params as class properties.
"""
def __init__(self):
super() # Parent class constructor
self.image_idx = 0
def on_train_begin(self, logs={}):
predictions = self.model.predict(basis[:1])
display_picture(predictions[0], 'train-viz-{}'.format(self.image_idx))
self.image_idx += 1
def on_batch_end(self, batch, logs={}):
predictions = self.model.predict(basis[:1])
display_picture(predictions[0], 'train-viz-{}'.format(self.image_idx))
self.image_idx += 1
history = model.fit(x=basis, y=outputs, epochs=EPOCHS, batch_size=512, callbacks=[ImageHistory()]) # See Keras docs for the history object
dump_history(history.history)
if save_weights:
model.save_weights('./infer-task/{}weights.hdf5'.format(namespace))
# Open and prepare val set
test = h5py.File('./infer-task-dsets/infer-task-{}-val.hdf5'.format(infile), 'r')
outputs = test['outputs'][:]
basis = test['basis'][:]
print('Validating model on {} fonts...'.format(test['basis'].shape[0]))
loss = model.evaluate(x=basis, y=outputs)
# View classified examples from the validation set
predictions = model.predict(basis)
display_picture(predictions[0], 'val')
display_basis(basis[0], 'basis-val')
if run_test_set:
# Open and prepare test set
test = h5py.File('./infer-task-dsets/infer-task-{}-test.hdf5'.format(infile), 'r')
outputs = test['outputs'][:]
basis = test['basis'][:]
print('Testing model on {} fonts...'.format(test['basis'].shape[0]))
loss = model.evaluate(x=basis, y=outputs)
# View classified examples from the validation set
predictions = model.predict(basis)
display_picture(predictions[0], 'test')
display_basis(basis[0], 'basis-test')
def parse_args():
DEFAULT_FILENAME = 'fonts-50'
parser = argparse.ArgumentParser(description='Run generative task.')
parser.add_argument('--infile', '-i', default=DEFAULT_FILENAME, help='Name of data infile.')
parser.add_argument('--experiment', '-e', default=0, type=int, help='Experiment number.')
parser.add_argument('--test', '-t', action='store_true', help='Run test set (default: False).')
parser.add_argument('--save_weights', '-s', action='store_true', help='Save weights (default: False).')
parser.add_argument('--load_weights', '-w', help='Load weights from hdf5 file (default: None).')
parser.add_argument('--tmp_dir', '-dir', default='tmp', help='Temp dir to dump info (default: tmp).')
parser.add_argument('--predict', '-p', action='store_true', help='Predict only.')
return parser.parse_args()
def main():
args = parse_args()
generate_task(args.infile, args.experiment, args.test, args.save_weights, args.load_weights, args.tmp_dir, args.predict)
if __name__ == '__main__':
main()