def build_net(self):
assert 'num_action' in self.params.keys() and isinstance(self.params['num_action'], int), \
'param num_action (type int) needed ... '
# load base model
self.base_model = model_generator(model_name=self.params['model_name']) \
if 'model_name' in self.params.keys() else None
self.action_net = self.base_model. \
add(keras.layers.Dense(units=self.params['num_action'])). \
add(keras.layers.Softmax())
self.value_net = self.base_model. \
add(keras.layers.Dense(units=1))
pickle.dump(mylist, file_pi) # Load the index to the samples of the dataset for 10 folds mylist = pickle.load(open(os.path.join(dir_of_file,"data","FOLDS_ATTACK.obj"), 'rb' )) #Train the model on folds 1 to 9 and test on fold 0 (default) train_index, test_index = mylist[fold] #checkpoints str1="weightsurban_ATTACK_SINCNET+VGG19_" str2=""+str(0)+".best.hdf5" filepath_model_weights=str1+str2 print(filepath_model_weights) model_end = model_generator() #train & test set X_train, X_test = X[train_index], X[test_index] Y_train, Y_test = Y[train_index], Y[test_index] #Fit the model checkpoint = ModelCheckpoint(filepath_model_weights, monitor='val_acc', verbose=1, save_best_only=True, mode='max') callbacks_list = [checkpoint] model_end.fit( X_train, Y_train, batch_size=20,nb_epoch=100, verbose=1,callbacks=callbacks_list,validation_split=0.10) #Load weights of the model model_end.load_weights(filepath_model_weights)
def example_gan(result_dir="output", data_dir="data"):
input_shape = (128, 128, 3)
local_shape = (64, 64, 3)
batch_size = 32
n_epoch = 10
#tc = int(n_epoch * 0.18)
#td = int(n_epoch * 0.02)
tc = 2
td = 2
alpha = 0.0004
train_datagen = DataGenerator(input_shape[:2], local_shape[:2])
generator = model_generator(input_shape)
discriminator = model_discriminator(input_shape, local_shape)
optimizer = Adadelta()
# build model
org_img = Input(shape=input_shape)
mask = Input(shape=(input_shape[0], input_shape[1], 1))
in_img = merge([org_img, mask],
mode=lambda x: x[0] * (1 - x[1]),
output_shape=input_shape)
imitation = generator(in_img)
completion = merge([imitation, org_img, mask],
mode=lambda x: x[0] * x[2] + x[1] * (1 - x[2]),
output_shape=input_shape)
cmp_container = Container([org_img, mask], completion, name='g_container')
cmp_out = cmp_container([org_img, mask])
cmp_model = Model([org_img, mask], cmp_out)
cmp_model.compile(loss='mse',
optimizer=optimizer)
local_img = Input(shape=local_shape)
d_container = Container([org_img, local_img], discriminator([org_img, local_img]),
name='d_container')
d_model = Model([org_img, local_img], d_container([org_img, local_img]))
d_model.compile(loss='binary_crossentropy',
optimizer=optimizer)
'''
'''
cmp_model.summary()
d_model.summary()
from keras.utils import plot_model
plot_model(cmp_model, to_file='cmp_model.png', show_shapes=True)
plot_model(d_model, to_file='d_model.png', show_shapes=True)
def random_cropping(x, x1, y1, x2, y2):
out = []
for idx in range(batch_size):
out.append(x[idx, y1[idx]:y2[idx], x1[idx]:x2[idx], :])
return K.stack(out, axis=0)
cropping = Lambda(random_cropping, output_shape=local_shape)
for n in range(n_epoch):
''''''
org_img = Input(shape=input_shape)
mask = Input(shape=(input_shape[0], input_shape[1], 1))
in_img = merge([org_img, mask],
mode=lambda x: x[0] * (1 - x[1]),
output_shape=input_shape)
imitation = generator(in_img)
completion = merge([imitation, org_img, mask],
mode=lambda x: x[0] * x[2] + x[1] * (1 - x[2]),
output_shape=input_shape)
cmp_container = Container([org_img, mask], completion, name='g_container')
cmp_out = cmp_container([org_img, mask])
cmp_model = Model([org_img, mask], cmp_out)
cmp_model.compile(loss='mse',
optimizer=optimizer)
local_img = Input(shape=local_shape)
d_container = Container([org_img, local_img], discriminator([org_img, local_img]),
name='d_container')
d_model = Model([org_img, local_img], d_container([org_img, local_img]))
d_model.compile(loss='binary_crossentropy',
optimizer=optimizer)
cmp_model = Model([org_img, mask], cmp_out)
local_img = Input(shape=local_shape)
d_container = Container([org_img, local_img], discriminator([org_img, local_img]),
name='d_container')
d_model = Model([org_img, local_img], d_container([org_img, local_img]))
'''
for inputs, points, masks in train_datagen.flow_from_directory(data_dir, batch_size,
hole_min=48, hole_max=64):
cmp_image = cmp_model.predict([inputs, masks])
local = []
local_cmp = []
for i in range(batch_size):
x1, y1, x2, y2 = points[i]
local.append(inputs[i][y1:y2, x1:x2, :])
local_cmp.append(cmp_image[i][y1:y2, x1:x2, :])
valid = np.ones((batch_size, 1))
fake = np.zeros((batch_size, 1))
g_loss = 0.0
d_loss = 0.0
if n < tc:
g_loss = cmp_model.train_on_batch([inputs, masks], inputs)
print("epoch: %d < %d [D loss: %e] [G mse: %e]" % (n,tc, d_loss, g_loss))
else:
#d_model.trainable = True
d_loss_real = d_model.train_on_batch([inputs, np.array(local)], valid)
d_loss_fake = d_model.train_on_batch([cmp_image, np.array(local_cmp)], fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
print('train D',n,(tc+td),'|',d_loss,'|',g_loss)
if n >= tc + td:
d_container.trainable = False
cropping.arguments = {'x1': points[:, 0], 'y1': points[:, 1],
'x2': points[:, 2], 'y2': points[:, 3]}
all_model = Model([org_img, mask],
[cmp_out, d_container([cmp_out, cropping(cmp_out)])])
all_model.compile(loss=['mse', 'binary_crossentropy'],
loss_weights=[1.0, alpha], optimizer=optimizer)
g_loss = all_model.train_on_batch([inputs, masks],
[inputs, valid])
#print("epoch: %d [D loss: %e] [G all: %e]" % (n, d_loss, g_loss))
print(all_model.metrics_names)
print('train ALL',n,'|',d_loss,'|',g_loss)
'''
if n < tc:
print("epoch: %d < %d [D loss: %e] [G mse: %e]" % (n,tc, d_loss, g_loss))
else:
print('train D',n,(tc+td),'|',d_loss,'|',g_loss)
if n >= tc + td:
print(all_model.metrics_names)
print('train ALL',n,'|',d_loss,'|',g_loss)
'''
num_img = min(5, batch_size)
fig, axs = plt.subplots(num_img, 3)
for i in range(num_img):
axs[i, 0].imshow(inputs[i] * (1 - masks[i]))
axs[i, 0].axis('off')
axs[i, 0].set_title('Input')
axs[i, 1].imshow(cmp_image[i])
axs[i, 1].axis('off')
axs[i, 1].set_title('Output')
axs[i, 2].imshow(inputs[i])
axs[i, 2].axis('off')
axs[i, 2].set_title('Ground Truth')
fig.savefig(os.path.join(result_dir, "result_%d.png" % n))
plt.close()
# save model
generator.save(os.path.join(result_dir, "generator_%d.h5" % n))
discriminator.save(os.path.join(result_dir, "discriminator_%d.h5" % n))
K.clear_session()
def example_gan(result_dir="output", data_dir="data"):
input_shape = (256, 256, 3)
local_shape = (128, 128, 3)
batch_size = 4
n_epoch = 100
tc = int(n_epoch * 0.18)
td = int(n_epoch * 0.02)
alpha = 0.0004
train_datagen = DataGenerator(data_dir, input_shape[:2], local_shape[:2])
generator = model_generator(input_shape)
discriminator = model_discriminator(input_shape, local_shape)
optimizer = Adadelta()
# build model
org_img = Input(shape=input_shape)
mask = Input(shape=(input_shape[0], input_shape[1], 1))
in_img = merge([org_img, mask],
mode=lambda x: x[0] * (1 - x[1]),
output_shape=input_shape)
imitation = generator(in_img)
completion = merge([imitation, org_img, mask],
mode=lambda x: x[0] * x[2] + x[1] * (1 - x[2]),
output_shape=input_shape)
cmp_container = Container([org_img, mask], completion)
cmp_out = cmp_container([org_img, mask])
cmp_model = Model([org_img, mask], cmp_out)
cmp_model.compile(loss='mse', optimizer=optimizer)
cmp_model.summary()
in_pts = Input(shape=(4, ), dtype='int32')
d_container = Container([org_img, in_pts], discriminator([org_img,
in_pts]))
d_model = Model([org_img, in_pts], d_container([org_img, in_pts]))
d_model.compile(loss='binary_crossentropy', optimizer=optimizer)
d_model.summary()
d_container.trainable = False
all_model = Model([org_img, mask, in_pts],
[cmp_out, d_container([cmp_out, in_pts])])
all_model.compile(loss=['mse', 'binary_crossentropy'],
loss_weights=[1.0, alpha],
optimizer=optimizer)
all_model.summary()
for n in range(n_epoch):
progbar = generic_utils.Progbar(len(train_datagen))
for inputs, points, masks in train_datagen.flow(batch_size):
cmp_image = cmp_model.predict([inputs, masks])
valid = np.ones((batch_size, 1))
fake = np.zeros((batch_size, 1))
g_loss = 0.0
d_loss = 0.0
if n < tc:
g_loss = cmp_model.train_on_batch([inputs, masks], inputs)
else:
d_loss_real = d_model.train_on_batch([inputs, points], valid)
d_loss_fake = d_model.train_on_batch([cmp_image, points], fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
if n >= tc + td:
g_loss = all_model.train_on_batch([inputs, masks, points],
[inputs, valid])
g_loss = g_loss[0] + alpha * g_loss[1]
progbar.add(inputs.shape[0],
values=[("D loss", d_loss), ("G mse", g_loss)])
num_img = min(5, batch_size)
fig, axs = plt.subplots(num_img, 3)
for i in range(num_img):
axs[i, 0].imshow(inputs[i] * (1 - masks[i]))
axs[i, 0].axis('off')
axs[i, 0].set_title('Input')
axs[i, 1].imshow(cmp_image[i])
axs[i, 1].axis('off')
axs[i, 1].set_title('Output')
axs[i, 2].imshow(inputs[i])
axs[i, 2].axis('off')
axs[i, 2].set_title('Ground Truth')
fig.savefig(os.path.join(result_dir, "result_%d.png" % n))
plt.close()
# save model
generator.save(os.path.join(result_dir, "generator.h5"))
discriminator.save(os.path.join(result_dir, "discriminator.h5"))
def example_gan(result_dir="output", data_dir="data"):
input_shape = (256, 256, 3)
local_shape = (128, 128, 3)
batch_size = 4
n_epoch = 100
train_datagen = DataGenerator(input_shape[:2], local_shape[:2])
generator = model_generator(input_shape)
discriminator = model_discriminator(input_shape, local_shape)
optimizer = Adam(0.0002, 0.5)
# build model
org_img = Input(shape=input_shape)
mask = Input(shape=(input_shape[0], input_shape[1], 1))
in_img = merge([org_img, mask],
mode=lambda x: x[0] * (1 - x[1]),
output_shape=input_shape)
imitation = generator(in_img)
completion = merge([imitation, org_img, mask],
mode=lambda x: x[0] * x[2] + x[1] * (1 - x[2]),
output_shape=input_shape)
cmp_model = Model([org_img, mask], completion)
cmp_model.compile(loss='mse', optimizer=optimizer)
discriminator.compile(loss='binary_crossentropy', optimizer=optimizer)
for n in range(n_epoch):
for inputs, points, masks in train_datagen.flow_from_directory(
data_dir, batch_size):
cmp_image = cmp_model.predict([inputs, masks])
local = []
local_cmp = []
for i in range(batch_size):
x1, y1, x2, y2 = points[i]
local.append(inputs[i][y1:y2, x1:x2, :])
local_cmp.append(cmp_image[i][y1:y2, x1:x2, :])
valid = np.ones((batch_size, 1))
fake = np.zeros((batch_size, 1))
# Train the discriminator
d_loss_real = discriminator.train_on_batch(
[inputs, np.array(local)], valid)
d_loss_fake = discriminator.train_on_batch(
[cmp_image, np.array(local_cmp)], fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
g_loss = cmp_model.train_on_batch([inputs, masks], inputs)
print("%d [D loss: %f] [G mse: %f]" % (n, d_loss, g_loss))
num_img = min(5, batch_size)
fig, axs = plt.subplots(num_img, 3)
for i in range(num_img):
axs[i, 0].imshow(inputs[i] * (1 - masks[i]))
axs[i, 0].axis('off')
axs[i, 0].set_title('Input')
axs[i, 1].imshow(cmp_image[i])
axs[i, 1].axis('off')
axs[i, 1].set_title('Output')
axs[i, 2].imshow(inputs[i])
axs[i, 2].axis('off')
axs[i, 2].set_title('Ground Truth')
fig.savefig(os.path.join(result_dir, "result_%d.png" % n))
plt.close()
# save model
generator.save(os.path.join(result_dir, "generator.h5"))
discriminator.save(os.path.join(result_dir, "discriminator.h5"))
def example_gan(result_dir="output", data_dir="data"):
input_shape = (256, 256, 3)
local_shape = (128, 128, 3)
batch_size = 4
n_epoch = 100
tc = int(n_epoch * 0.18)
td = int(n_epoch * 0.02)
alpha = 0.0004
train_datagen = DataGenerator(data_dir, input_shape[:2], local_shape[:2])
generator = model_generator(input_shape)
discriminator = model_discriminator(input_shape, local_shape)
optimizer = Adadelta()
# build model
org_img = Input(shape=input_shape)
mask = Input(shape=(input_shape[0], input_shape[1], 1))
in_img = merge([org_img, mask],
mode=lambda x: x[0] * (1 - x[1]),
output_shape=input_shape)
imitation = generator(in_img)
completion = merge([imitation, org_img, mask],
mode=lambda x: x[0] * x[2] + x[1] * (1 - x[2]),
output_shape=input_shape)
cmp_container = Container([org_img, mask], completion)
cmp_out = cmp_container([org_img, mask])
cmp_model = Model([org_img, mask], cmp_out)
cmp_model.compile(loss='mse',
optimizer=optimizer)
cmp_model.summary()
in_pts = Input(shape=(4,), dtype='int32')
d_container = Container([org_img, in_pts], discriminator([org_img, in_pts]))
d_model = Model([org_img, in_pts], d_container([org_img, in_pts]))
d_model.compile(loss='binary_crossentropy',
optimizer=optimizer)
d_model.summary()
d_container.trainable = False
all_model = Model([org_img, mask, in_pts],
[cmp_out, d_container([cmp_out, in_pts])])
all_model.compile(loss=['mse', 'binary_crossentropy'],
loss_weights=[1.0, alpha], optimizer=optimizer)
all_model.summary()
for n in range(n_epoch):
progbar = generic_utils.Progbar(len(train_datagen))
for inputs, points, masks in train_datagen.flow(batch_size):
cmp_image = cmp_model.predict([inputs, masks])
valid = np.ones((batch_size, 1))
fake = np.zeros((batch_size, 1))
g_loss = 0.0
d_loss = 0.0
if n < tc:
g_loss = cmp_model.train_on_batch([inputs, masks], inputs)
else:
d_loss_real = d_model.train_on_batch([inputs, points], valid)
d_loss_fake = d_model.train_on_batch([cmp_image, points], fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
if n >= tc + td:
g_loss = all_model.train_on_batch([inputs, masks, points],
[inputs, valid])
g_loss = g_loss[0] + alpha * g_loss[1]
progbar.add(inputs.shape[0], values=[("D loss", d_loss), ("G mse", g_loss)])
num_img = min(5, batch_size)
fig, axs = plt.subplots(num_img, 3)
for i in range(num_img):
axs[i, 0].imshow(inputs[i] * (1 - masks[i]))
axs[i, 0].axis('off')
axs[i, 0].set_title('Input')
axs[i, 1].imshow(cmp_image[i])
axs[i, 1].axis('off')
axs[i, 1].set_title('Output')
axs[i, 2].imshow(inputs[i])
axs[i, 2].axis('off')
axs[i, 2].set_title('Ground Truth')
fig.savefig(os.path.join(result_dir, "result_%d.png" % n))
plt.close()
# save model
generator.save(os.path.join(result_dir, "generator.h5"))
discriminator.save(os.path.join(result_dir, "discriminator.h5"))
def build_model():
optimizer = Adadelta()
# build Completion Network model
org_img = Input(shape=input_shape, dtype='float32')
mask = Input(shape=(input_shape[0], input_shape[1], 1))
generator, completion_out = model_generator(org_img, mask)
completion_model = generator.compile(loss='mse', optimizer=optimizer)
# build Discriminator model
in_pts = Input(shape=(4, ), dtype='int32') # [y1,x1,y2,x2]
discriminator = model_discriminator(input_shape, local_shape)
d_container = Network(inputs=[org_img, in_pts],
outputs=discriminator([org_img, in_pts]))
d_out = d_container([org_img, in_pts])
d_model = Model([org_img, in_pts], d_out)
d_model.compile(loss='binary_crossentropy', optimizer=optimizer)
d_container.trainable = False
# build Discriminator & Completion Network models
all_model = Model([org_img, mask, in_pts], [completion_out, d_out])
all_model.compile(loss=['mse', 'binary_crossentropy'],
loss_weights=[1.0, alpha],
optimizer=optimizer)
X_train = filenames[:5000]
valid = np.ones((batch_size, 1)) ## label
fake = np.zeros((batch_size, 1)) ## label
for n in range(n_epoch):
progbar = generic_utils.Progbar(len(X_train))
for i in range(int(len(X_train) // batch_size)):
X_batch = X_train[i * batch_size:(i + 1) * batch_size]
inputs = np.array([
process_image(filename, input_shape[:2])
for filename in X_batch
])
points, masks = get_points(batch_size)
completion_image = generator.predict([inputs, masks])
g_loss = 0.0
d_loss = 0.0
if n < tc:
g_loss = generator.train_on_batch([inputs, masks], inputs)
else:
d_loss_real = d_model.train_on_batch([inputs, points], valid)
d_loss_fake = d_model.train_on_batch(
[completion_image, points], fake)
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
if n >= tc + td:
g_loss = all_model.train_on_batch([inputs, masks, points],
[inputs, valid])
g_loss = g_loss[0] + alpha * g_loss[1]
progbar.add(inputs.shape[0],
values=[("Epoch", int(n + 1)), ("D loss", d_loss),
("G mse", g_loss)])
# show_image
show_image(batch_size, n, inputs, masks, completion_image)
# save model
generator.save("model/generator.h5")
discriminator.save("model/discriminator.h5")