forked from avishekbh/D2SCGAN
-
Notifications
You must be signed in to change notification settings - Fork 0
/
d2scgan_train.py
460 lines (387 loc) · 17.7 KB
/
d2scgan_train.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
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
import keras
from keras.models import Model
from keras.layers import Conv2D, Input, Deconvolution2D, merge
from keras.optimizers import SGD, adam
import prepare_data as pd
import numpy
import math
from keras.utils import plot_model
from keras.models import Sequential
from keras.layers import Dense, Dropout, Input, GaussianDropout, GaussianNoise, Add
from keras.layers import Reshape
from keras.layers.core import Activation
from keras.layers.advanced_activations import LeakyReLU, PReLU
from keras.layers.normalization import BatchNormalization
from keras.layers.convolutional import UpSampling2D
from keras.layers.convolutional import Conv2D, MaxPooling2D, Deconv2D, Conv2DTranspose, AveragePooling2D
from keras.layers.core import Flatten
from keras.optimizers import SGD, adam, adadelta
from keras.datasets import mnist
from keras.models import Model
import numpy as np
import numpy.linalg as LA
import numpy.linalg as LA
from PIL import Image
import argparse
import os
import os.path
import math
from keras import backend as K
import tensorflow as tf
import matplotlib.pyplot as plt
import scipy as misc
from keras.callbacks import TensorBoard
import random
# import Image
import math
from scipy import misc
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.per_process_gpu_memory_fraction = 0.99
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
K.set_session(sess)
def _tf_fspecial_gauss(size, sigma):
"""Function to mimic the 'fspecial' gaussian MATLAB function
"""
x_data, y_data = np.mgrid[-size // 2 + 1:size // 2 + 1, -size // 2 + 1:size // 2 + 1]
x_data = np.expand_dims(x_data, axis=-1)
x_data = np.expand_dims(x_data, axis=-1)
y_data = np.expand_dims(y_data, axis=-1)
y_data = np.expand_dims(y_data, axis=-1)
x = tf.constant(x_data, dtype=tf.float32)
y = tf.constant(y_data, dtype=tf.float32)
g = tf.exp(-((x ** 2 + y ** 2) / (2.0 * sigma ** 2)))
return g / tf.reduce_sum(g)
def tf_ssim(img1, img2, cs_map=False, mean_metric=True, size=5, sigma=1.5):
window = _tf_fspecial_gauss(size, sigma) # window shape [size, size]
# print window
K1 = 0.01
K2 = 0.03
L = 255 # depth of image (255 in case the image has a differnt scale)
C1 = (K1 * L) ** 2
C2 = (K2 * L) ** 2
mu1 = tf.nn.conv2d(img1, window, strides=[1, 1, 1, 1], padding='VALID')
mu2 = tf.nn.conv2d(img2, window, strides=[1, 1, 1, 1], padding='VALID')
mu1_sq = mu1 * mu1
mu2_sq = mu2 * mu2
mu1_mu2 = mu1 * mu2
sigma1_sq = tf.nn.conv2d(img1 * img1, window, strides=[1, 1, 1, 1], padding='VALID') - mu1_sq
sigma2_sq = tf.nn.conv2d(img2 * img2, window, strides=[1, 1, 1, 1], padding='VALID') - mu2_sq
sigma12 = tf.nn.conv2d(img1 * img2, window, strides=[1, 1, 1, 1], padding='VALID') - mu1_mu2
if cs_map:
value = (((2 * mu1_mu2 + C1) * (2 * sigma12 + C2)) / ((mu1_sq + mu2_sq + C1) *
(sigma1_sq + sigma2_sq + C2)),
(2.0 * sigma12 + C2) / (sigma1_sq + sigma2_sq + C2))
else:
value = ((2 * mu1_mu2 + C1) * (2 * sigma12 + C2)) / ((mu1_sq + mu2_sq + C1) *
(sigma1_sq + sigma2_sq + C2))
if mean_metric:
value = tf.reduce_mean(value)
return value
def tf_ms_ssim(img1, img2, mean_metric=True, level=5):
weight = tf.constant([0.0448, 0.2856, 0.3001, 0.2363, 0.1333], dtype=tf.float32)
mssim = []
mcs = []
for l in range(level):
ssim_map, cs_map = tf_ssim(img1, img2, cs_map=True, mean_metric=False)
mssim.append(tf.reduce_mean(ssim_map))
mcs.append(tf.reduce_mean(cs_map))
filtered_im1 = tf.nn.avg_pool(img1, [1, 2, 2, 1], [1, 2, 2, 1], padding='SAME')
filtered_im2 = tf.nn.avg_pool(img2, [1, 2, 2, 1], [1, 2, 2, 1], padding='SAME')
img1 = filtered_im1
img2 = filtered_im2
# list to tensor of dim D+1
mssim = tf.stack(mssim, axis=0)
mcs = tf.stack(mcs, axis=0)
value = (tf.reduce_prod(mcs[0:level - 1] ** weight[0:level - 1]) *
(mssim[level - 1] ** weight[level - 1]))
if mean_metric:
value = tf.reduce_mean(value)
return value
def mse(y_true, y_pred):
return K.mean(K.square(y_pred - y_true), axis=-1)
def multires_patchwise_mse(y_true, y_pred):
ch1_t, ch2_t, ch3_t = tf.split(y_true, [1, 1, 1], axis=3) # Separating channels from the image tensor "y_true"
ch1_p, ch2_p, ch3_p = tf.split(y_pred, [1, 1, 1], axis=3) # Separating channels from the image tensor "y_pred"
final_loss_1, final_loss_2, final_loss_3 = 0.0, 0.0, 0.0
alpha = 0.0
for kernel_size in [(input_dim/4), (input_dim/2), input_dim]:
kernel = [1, kernel_size, kernel_size, 1]
if kernel_size == (input_dim/4):
strides = [1, 10, 10, 1]
alpha = 0.00001
if kernel_size == (input_dim/2):
strides = [1, 20, 20, 1]
alpha = 0.0001
if kernel_size == input_dim:
strides = [1, 10, 10, 1]
alpha = 0.001
patches_true_1 = tf.extract_image_patches(ch1_t, kernel, strides, [1, 1, 1, 1], padding='SAME')
patches_true_2 = tf.extract_image_patches(ch2_t, kernel, strides, [1, 1, 1, 1], padding='SAME')
patches_true_3 = tf.extract_image_patches(ch3_t, kernel, strides, [1, 1, 1, 1], padding='SAME')
patches_pred_1 = tf.extract_image_patches(ch1_p, kernel, strides, [1, 1, 1, 1], padding='SAME')
patches_pred_2 = tf.extract_image_patches(ch2_p, kernel, strides, [1, 1, 1, 1], padding='SAME')
patches_pred_3 = tf.extract_image_patches(ch3_p, kernel, strides, [1, 1, 1, 1], padding='SAME')
loss_1 = 0.0
loss_2 = 0.0
loss_3 = 0.0
patches = math.ceil((kernel_output - kernel_size) / strides[1])
if patches > 0:
for i in range(int(patches)):
for j in range(int(patches)):
loss_1 = loss_1 + (0.2989 * mse(patches_true_1[0, i, j], patches_pred_1[0, i, j]) * alpha)
loss_2 = loss_2 + (0.5870 * mse(patches_true_2[0, i, j], patches_pred_2[0, i, j]) * alpha)
loss_3 = loss_3 + (0.1141 * mse(patches_true_3[0, i, j], patches_pred_3[0, i, j]) * alpha)
loss_1 = loss_1 / (patches * patches)
loss_2 = loss_2 / (patches * patches)
loss_3 = loss_3 / (patches * patches)
else:
loss_1 = 0.2989 * mse(patches_true_1, patches_pred_1) * alpha
loss_2 = 0.5870 * mse(patches_true_2, patches_pred_2) * alpha
loss_3 = 0.1141 * mse(patches_true_3, patches_pred_3) * alpha
final_loss_1 = final_loss_1 + loss_1
final_loss_2 = final_loss_2 + loss_2
final_loss_3 = final_loss_3 + loss_3
total_loss = 0.000001 * (final_loss_1 + final_loss_2 + final_loss_3)
return total_loss
def kld(y_true, y_pred):
y_true = K.clip(y_true, K.epsilon(), 1)
y_pred = K.clip(y_pred, K.epsilon(), 1)
return K.sum(y_true * K.log(y_true / y_pred), axis=-1)
def jsd(y_true, y_pred):
img1 = tf.image.rgb_to_grayscale(y_true)
img2 = tf.image.rgb_to_grayscale(y_pred)
val_range = [0.0, 255.0]
hist1 = tf.cast(tf.histogram_fixed_width(img1, val_range, nbins=64), tf.float32)
hist2 = tf.cast(tf.histogram_fixed_width(img2, val_range, nbins=64), tf.float32)
m = tf.cast(tf.add(hist1, hist2), tf.float32)
m_mod = tf.divide(m, tf.constant(2.))
t1 = kld(hist1, m_mod)
t2 = kld(hist2, m_mod)
jsd = tf.div(tf.add(t1, t2), tf.constant(2.))
jsd = tf.where(tf.is_nan(jsd), 0., jsd)
return 0.05 * jsd
def chi_squared_distance(y_true, y_pred):
val_range = [-0.5, 0.5]
hist1 = tf.cast(tf.histogram_fixed_width(y_true, val_range, nbins=16), tf.float32)
hist2 = tf.cast(tf.histogram_fixed_width(y_pred, val_range, nbins=16), tf.float32)
hist1_normalize = K.l2_normalize(hist1)
hist2_normalize = K.l2_normalize(hist2)
elmw_sqd = tf.cast(tf.squared_difference(hist1_normalize, hist2_normalize), tf.float32)
elmw_div = tf.cast(tf.truediv(elmw_sqd, hist1_normalize), tf.float32)
loss = tf.cast(tf.reduce_sum(elmw_div), tf.float32) * 0.5
loss = tf.where(tf.is_nan(loss), 0., loss)
loss = tf.where(tf.is_inf(loss), 0., loss)
return 0.001 * (loss / 16)
def structural_loss(y_true, y_pred):
img1 = tf.image.rgb_to_grayscale(y_true)
img2 = tf.image.rgb_to_grayscale(y_pred)
ms_ssim_loss = 1 - tf_ms_ssim(img1, img2) # heuristics, gives better generation result
mse_loss = keras.losses.mse(y_true, y_pred)
mr_pmse_loss = multires_patchwise_mse(y_true, y_pred)
chi_squared_loss = chi_squared_distance(img1, img2)
jsd_loss = jsd(y_true, y_pred)
st_loss = K.abs(ms_ssim_loss + 0.1 * ms_pmse_loss + mse_loss + 0.1 * chi_squared_loss + 0.0001 * jsd_loss)
return st_loss
def generator(input_col, input_row):
_input = Input(shape=(input_col, input_row, 3), name='input')
# Channel 1: Shallow channel
x = Conv2D(1, (1, 1), input_shape=(input_row, input_row, 3), padding='same')(_input)
x = Flatten()(x)
x = Dense(2048)(x)
x = Activation('tanh')(x)
x = Dense(128 * (dim_output/4) * (dim_output/4))(x)
x = BatchNormalization()(x)
x = Activation('tanh')(x)
x = Reshape(((dim_output/4), (dim_output/4), 128), input_shape=(128 * (dim_output/4) * (dim_output/4),))(x)
x = UpSampling2D(size=(2, 2))(x)
x = Conv2D(64, (5, 5), padding='same')(x)
x = Activation('tanh')(x)
x = UpSampling2D(size=(2, 2))(x)
x = Conv2D(64, (5, 5), padding='same')(x)
out_1 = Conv2D(nb_filter=3, nb_row=5, nb_col=5, activation='tanh', border_mode='same')(x)
model1 = Model(input=_input, output=out_1)
g_optim = adam(lr=0.001, beta_1=0.9, beta_2=0.99, epsilon=1e-08, decay=1e-5)
model1.compile(loss=structural_loss, optimizer=g_optim)
plot_model(model1, to_file='channel1.png')
# Channel 2: Deep channel
x = Conv2D(3, (1, 1), input_shape=(input_row, input_row, 3), padding='same')(_input)
x = Activation('tanh')(x)
x = Flatten()(x)
x = Dense(2048)(x)
x = Activation('tanh')(x)
x = Dense(128 * (dim_output/4) * (dim_output/4))(x)
x = BatchNormalization()(x)
x = Activation('tanh')(x)
x = Reshape(((dim_output/4), (dim_output/4), 128), input_shape=(128 * (dim_output/4) * (dim_output/4),))(x)
x = UpSampling2D(size=(2, 2))(x)
x = Conv2D(128, (5, 5), padding='same')(x)
x = Activation('tanh')(x)
x = UpSampling2D(size=(2, 2))(x)
x = Conv2D(128, (5, 5), padding='same')(x)
x = Activation('tanh')(x)
x = Conv2D(64, (5, 5), padding='same')(x)
x = Activation('tanh')(x)
x = Conv2D(32, (5, 5), padding='same')(x)
x = Activation('tanh')(x)
out_2 = Conv2D(nb_filter=3, nb_row=5, nb_col=5, activation='tanh', border_mode='same')(x)
model2 = Model(input=_input, output=out_2)
g_optim = adam(lr=0.001, beta_1=0.9, beta_2=0.99, epsilon=1e-08, decay=1e-5)
model2.compile(loss=structural_loss, optimizer=g_optim)
plot_model(model2, to_file='channel2.png')
EES = model1(_input)
EED = model2(_input)
add_layer_output = keras.layers.average(inputs=[EED, EES])
model = Model(input=_input, output=add_layer_output)
Adam = adam(lr=0.001, beta_1=0.9, beta_2=0.99, epsilon=1e-08, decay=1e-5)
model.compile(optimizer=Adam, loss=structural_loss)
#
# plot_model(model, to_file='generator.png')
return model
def discriminator_model(input_row, input_col):
_input = Input(shape=(input_col, input_row, 3), name='input')
x = Conv2D(64, (5, 5), padding='same', input_shape=(input_row, input_col, 3))(_input)
x = Activation('tanh')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Conv2D(128, (5, 5))(x)
x = Activation('tanh')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Flatten()(x)
x = Dense(1024)(x)
out = Activation('tanh')(x)
x = Dense(1)(out)
x = Activation('sigmoid')(x)
y = Dense(51)(out)
y = Activation('sigmoid')(y)
model = Model(input=_input, output=[x, y])
# plot_model(model, to_file='model2.png')
return model
def generator_containing_discriminator(g, d):
d.trainable = False
model = Model(inputs=g.inputs, outputs=d(g.outputs))
return model
def train(BATCH_SIZE, input_row, input_col):
# Trainers
# adadelta = keras.optimizers.adadelta(lr=0.0001, decay=1e-5)
# Adam = adam(lr=0.0001, beta_1=0.9, beta_2=0.99, epsilon=1e-08, decay=1e-5)
# d_optim = SGD(lr=0.0001, momentum=0.9, nesterov=True)
d = discriminator_model(dim_output, dim_output)
g = generator(input_row, input_col)
Adam = adam(lr=0.0001, beta_1=0.9, beta_2=0.99, epsilon=1e-08, decay=1e-5)
g.compile(optimizer=Adam, loss=structural_loss)
print(g.summary())
# d.load_weights('/home/vplab/PycharmProjects/DSC-GAN/Models_CAD/D1/discriminator100')
# g.load_weights('/home/vplab/PycharmProjects/DSC-GAN/Models_CAD/G1/generator100')
Y_tr = np.load('trImages.npy') ## Load gallery data
X_tr = np.load('tsImages.npy') ## Load probe data
X_lb = np.load('trLabels.npy') ## Load gallery labels
Y_lb = np.load('tsLabels.npy') ## Load probe labels
result = []
X_trn = []
Y_trn = []
X_lbl = []
Y_lbl = []
for x in range(0, len(X_tr)):
num = random.randint(0, len(X_tr) - 1)
while num in result:
num = random.randint(0, len(X_tr) - 1)
result.append(num)
for i in result:
X_im = X_tr[i]
X_label = X_lb[i]
X_trn.append(X_im)
X_lbl.append(X_label)
Y_im = Y_tr[i]
Y_label = Y_lb[i]
Y_trn.append(Y_im)
Y_lbl.append(Y_label)
X_train = np.asarray(X_trn)
Y_train = np.asarray(Y_trn)
X_labels = np.asarray(X_lbl)
Y_labels = np.asarray(Y_lbl)
# print(Y_train.shape[0])
g_loss = 0.
d_loss = 0.00001
alpha_loss = 0
s_x = []
g_x = []
d_x = []
d_on_g = generator_containing_discriminator(g, d)
d_optim = SGD(lr=0.00001, momentum=0.9, nesterov=True)
g_optim = adam(lr=0.00001, beta_1=0.9, beta_2=0.99, epsilon=1e-08, decay=1e-5)
d_on_g.compile(loss=['binary_crossentropy', 'categorical_crossentropy'], optimizer=g_optim)
d.trainable = True
d.compile(loss=['binary_crossentropy', 'categorical_crossentropy'], optimizer=d_optim)
for epoch in range(num_epochs):
print("Epoch is", epoch)
print("Number of batches", int(X_train.shape[0] / BATCH_SIZE))
for index in range(int(X_train.shape[0] / BATCH_SIZE)):
noise = np.asarray(Y_train[index * BATCH_SIZE:(index + 1) * BATCH_SIZE])
labels_batch = np.asarray(Y_labels[index * BATCH_SIZE:(index + 1) * BATCH_SIZE])
image_batch = np.asarray(X_train[index * BATCH_SIZE:(index + 1) * BATCH_SIZE])
generated_images = g.predict(noise, verbose=0)
if index % 30 == 0:
image = combine_images(generated_images)
image_or = combine_images(image_batch)
image = image * 127.5 + 127.5
image_or = image_or * 127.5 + 127.5
Image.fromarray(image.astype(np.uint8)).save("imgs/" +
str(epoch) + "_" + str(index) + ".png")
Image.fromarray(image_or.astype(np.uint8)).save("imgs/" +
str(epoch) + "_" + str(index) + "_or.png")
if epoch > 6:
g_trend = sum(g_x[epoch-6:epoch])/5.
d_trend = sum(d_x[epoch-6:epoch])/5.
diff = g_trend - d_trend
ratio = LA.norm(g_loss)/LA.norm(d_loss)
if diff > 1 or ratio > 5:
d.trainable = False
# print('visited')
if diff < -1 or ratio < 0.01:
g.trainable = False
X = np.array(image_batch)
y = np.array([1] * BATCH_SIZE)
d.train_on_batch(X, [y, labels_batch])
X = generated_images
y = np.array([0] * BATCH_SIZE)
d_loss = d.train_on_batch(X, [y, labels_batch])
y = np.array([1] * BATCH_SIZE)
d.trainable = False
alpha_loss = g.train_on_batch(noise, image_batch)
g_loss = d_on_g.train_on_batch(noise, [y, labels_batch])
d.trainable = True
g.trainable = True
print("epoch %d batch %d d_loss : %f g_loss : %f s_loss : %f" % (epoch, index, LA.norm(d_loss), LA.norm(g_loss), alpha_loss))
if epoch % 200 == 99:
g.save('Models_CAD/G/generator_cad5_%d.h5' % epoch, True)
d.save('Models_CAD/D/discriminator_cad5_%d.h5' % epoch, True)
np.save('Models_CAD/generator_cad5_.npy', g_x, True)
np.save('Models_CAD/discriminator_cad5_.npy', d_x, True)
np.save('Models_CAD/structural_cad5_.npy', s_x, True)
g_x.append(LA.norm(g_loss))
d_x.append(LA.norm(d_loss))
s_x.append(alpha_loss)
def l1_norm(weight_matrix):
return K.sum(K.abs(weight_matrix))
def combine_images(generated_images):
num = generated_images.shape[0]
width = int(math.sqrt(num))
height = int(math.ceil(float(num) / width))
shape = generated_images.shape[1:4]
image = np.zeros((height * shape[0], width * shape[1], shape[2]),
dtype=generated_images.dtype)
for index, rgb in enumerate(generated_images):
img = rgb
i = int(index / width)
j = index % width
image[i * shape[0]:(i + 1) * shape[0], j * shape[1]:(j + 1) * shape[1], 0] = img[:, :, 0]
image[i * shape[0]:(i + 1) * shape[0], j * shape[1]:(j + 1) * shape[1], 1] = img[:, :, 1]
image[i * shape[0]:(i + 1) * shape[0], j * shape[1]:(j + 1) * shape[1], 2] = img[:, :, 2]
return image
## Model parameters
kernel_output = 140
dim_output = 140
input_dim = 35
num_epochs = 100000
train(10, 35, 35)