def build(self, input_shape):
params = dict()
mu = K.variable(0)
sigma = K.variable(0.1, constraint=keras.constraints.non_neg()) # , constraint=lambda x: max(0, x))
x_up = K.variable(0.1, constraint=keras.constraints.non_neg()) # , constraint=lambda x: max(0, x))
x_down = K.variable(0.1, constraint=keras.constraints.non_neg()) # , constraint=lambda x: max(0, x))
alpha_up = K.variable(0.5, constraint=keras.constraints.non_neg()) # , constraint=lambda x: max(0, x))
alpha_down = K.variable(0.5, constraint=keras.constraints.non_neg()) # , constraint=lambda x: max(0, x))
self.params = dict(mu=mu, sigma=sigma, x_up=x_up, x_down=x_down, alpha_up=alpha_up, alpha_down=alpha_down)
super().build(input_shape)
def get_total_loss(content_losses, style_losses, total_var_loss,
content_weights, style_weights, tv_weights, class_targets):
total_loss = K.variable(0.)
# Compute content losses
for loss in content_losses:
weighted_loss = K.mean(K.gather(content_weights, class_targets) * loss)
weighted_content_losses.append(weighted_loss)
total_loss += weighted_loss
# Compute style losses
for loss in style_losses:
weighted_loss = K.mean(K.gather(style_weights, class_targets) * loss)
weighted_style_losses.append(weighted_loss)
total_loss += weighted_loss
# Compute tv loss
weighted_tv_loss = K.mean(
K.gather(tv_weights, class_targets) * total_var_loss)
total_loss += weighted_tv_loss
return (total_loss, weighted_content_losses, weighted_style_losses,
weighted_tv_loss)
def make_loss_function_wcc(weights):
""" make loss function: weighted categorical crossentropy
Args:
* weights<ktensor|nparray|list>: crossentropy weights
Returns:
* weighted categorical crossentropy function
"""
if isinstance(weights, list) or isinstance(weights, np.ndarray):
weights = K.variable(weights)
def loss(target, output, from_logits=False):
if not from_logits:
output /= tf.reduce_sum(output, len(output.get_shape()) - 1, True)
_epsilon = tf.convert_to_tensor(K.epsilon(),
dtype=output.dtype.base_dtype)
output = tf.clip_by_value(output, _epsilon, 1. - _epsilon)
weighted_losses = target * tf.log(output) * weights
return -tf.reduce_sum(weighted_losses, len(output.get_shape()) - 1)
else:
raise ValueError(
'WeightedCategoricalCrossentropy: not valid with logits')
return loss
def build(self, input_shape):
input_shape = to_tuple(input_shape)
if self.data_format == 'channels_first':
stack_size = input_shape[1]
self.kernel_shape = (self.filters, stack_size, self.nb_row,
self.nb_col)
self.kernel_norm_shape = (1, stack_size, self.nb_row, self.nb_col)
elif self.data_format == 'channels_last':
stack_size = input_shape[3]
self.kernel_shape = (self.nb_row, self.nb_col, stack_size,
self.filters)
self.kernel_norm_shape = (self.nb_row, self.nb_col, stack_size, 1)
else:
raise ValueError('Invalid data_format:', self.data_format)
self.W = self.add_weight(shape=self.kernel_shape,
initializer=partial(self.kernel_initializer),
name='{}_W'.format(self.name),
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint)
kernel_norm_name = '{}_kernel_norm'.format(self.name)
self.kernel_norm = K.variable(np.ones(self.kernel_norm_shape),
name=kernel_norm_name)
if self.use_bias:
self.b = self.add_weight(shape=(self.filters, ),
initializer='zero',
name='{}_b'.format(self.name),
regularizer=self.bias_regularizer,
constraint=self.bias_constraint)
else:
self.b = None
if self.initial_weights is not None:
self.set_weights(self.initial_weights)
del self.initial_weights
self.built = True
def __init__(self,
input_shape,
classes,
epochs,
batch_size,
extract_length=32 * 32 * 3 * 1,
extract_shape=(1, 32, 32, 3),
optimizer='adam',
dataset='cifar'):
self.input_shape = input_shape
self.classes = classes
self.model = self.get_model()
self.epochs = epochs
self.batch_size = batch_size
self.extract_length = extract_length
self.extract_shape = extract_shape
self.optimizer = optimizer
self.dataset = dataset
self.loss_value = None
self.acc_value = None
self.train_data, self.test_data = load_data(name=self.dataset,
classes=self.classes)
# Original design
#self.extracted_data = self.train_data[0].flatten()[:self.extract_length]
self.extracted_data = self.train_data[0].flatten(
)[:self.extract_length * 3]
self.extracted_data = rgb_to_grayscale(
self.extracted_data.reshape(self.extract_shape)).flatten()
#self.total_weights = self.model.get_weights()
self.total_weights = K.variable(
extract_params(self.extract_length, self.model.get_weights()))
self.model = self.compile_model()
def __init__(self,
input_dim,
train_ds,
val_ds=None,
filename="weightsvgg.h5",
coverage=0.8,
alpha=0.5,
train=True,
baseline=False):
self.lamda = coverage
self.alpha = alpha
self.mc_dropout_rate = K.variable(value=0)
self.num_classes = 2
self.weight_decay = 0.0005
self.x_shape = input_dim
self.filename = filename
self.model = self.build_model()
if baseline:
self.alpha = 0
if (train & os.path.isfile("saved_data/{}".format(self.filename))):
self.model.load_weights("saved_data/{}".format(self.filename))
def get_style_loss():
'''
这不是通常意义上的函数，这是一个函数调用。因为不用向其传递引用。
x_feature是一个tensor，shape未知
'''
layer_names = ['block3_conv1', 'block4_conv1']
style_loss = K.variable(0)
for layer_name in layer_names:
x_feature = get_feature_map(layer_name)[0][0]
s_feature = get_feature_val(layer_name, s_im)
# (56,56,256)
size, size1, channels = x_feature.shape
for i in range(channels):
C = x_feature[:, :, i]
S = s_feature[:, :, i].transpose()
el = K.sum(K.square(S - C))
el = el / (4.0 * int(channels) * int(channels) * int(size) *
int(size))
style_loss = style_loss + el
return style_loss
def __init__(self, optimizer, steps_per_update=1, **kwargs):
super(AccumOptimizer, self).__init__(**kwargs)
self.optimizer = optimizer
with K.name_scope(self.__class__.__name__):
self.steps_per_update = steps_per_update
self.iterations = K.variable(0, dtype='int64', name='iterations')
self.cond = K.equal(self.iterations % self.steps_per_update, 0)
self.lr = self.optimizer.lr
self.optimizer.lr = K.switch(self.cond, self.optimizer.lr, 0.)
for attr in ['momentum', 'rho', 'beta_1', 'beta_2']:
if hasattr(self.optimizer, attr):
value = getattr(self.optimizer, attr)
setattr(self, attr, value)
setattr(self.optimizer, attr,
K.switch(self.cond, value, 1 - 1e-7))
for attr in self.optimizer.get_config():
if not hasattr(self, attr):
value = getattr(self.optimizer, attr)
setattr(self, attr, value)
# Cover the original get_gradients method with accumulative gradients.
def get_gradients(loss, params):
return [ag / self.steps_per_update for ag in self.accum_grads]
self.optimizer.get_gradients = get_gradients
def weighted_categorical_crossentropy(weights):
"""
A weighted version of keras.objectives.categorical_crossentropy
Variables:
weights: numpy array of shape (C,) where C is the number of classes
Usage:
weights = np.array([0.5,2,10]) # Class one at 0.5, class 2 twice the normal weights, class 3 10x.
loss = weighted_categorical_crossentropy(weights)
model.compile(loss=loss,optimizer='adam')
"""
weights = K.variable(weights)
def loss(y_true, y_pred):
# scale predictions so that the class probas of each sample sum to 1
y_pred /= K.sum(y_pred, axis=-1, keepdims=True)
# clip to prevent NaN's and Inf's
y_pred = K.clip(y_pred, K.epsilon(), 1 - K.epsilon())
# calc
loss = y_true * K.log(y_pred) * weights
loss = -K.sum(loss, -1)
return loss
return loss
def make_soft(y_true, fragment_length, nb_output_bins, train_with_soft_target_stdev, with_prints=False):
receptive_field, _ = compute_receptive_field()
n_outputs = fragment_length - receptive_field + 1
# Make a gaussian kernel.
kernel_v = scipy.signal.gaussian(9, std=train_with_soft_target_stdev)
print(kernel_v)
kernel_v = np.reshape(kernel_v, [1, 1, -1, 1])
kernel = K.variable(kernel_v)
if with_prints:
y_true = print_t(y_true, 'y_true initial')
# y_true: [batch, timesteps, input_dim]
y_true = K.reshape(y_true, (-1, 1, nb_output_bins, 1)) # Same filter for all output; combine with batch.
# y_true: [batch*timesteps, n_channels=1, input_dim, dummy]
y_true = K.conv2d(y_true, kernel, padding='same')
y_true = K.reshape(y_true, (-1, n_outputs, nb_output_bins)) # Same filter for all output; combine with batch.
# y_true: [batch, timesteps, input_dim]
y_true /= K.sum(y_true, axis=-1, keepdims=True)
if with_prints:
y_true = print_t(y_true, 'y_true after')
return y_true
def test_magnitude_to_decibel(dynamic_range, dtype: str):
"""test for backend_keras.magnitude_to_decibel"""
x = np.array(
[[1e-20, 1e-5, 1e-3, 5e-2], [0.3, 1.0, 20.5, 9999]], dtype=dtype
) # random positive numbers
amin = 1e-5
x_decibel_ref = np.stack(
(
librosa.power_to_db(x[0], amin=amin, ref=1.0, top_db=dynamic_range),
librosa.power_to_db(x[1], amin=amin, ref=1.0, top_db=dynamic_range),
),
axis=0,
)
x_var = K.variable(x)
x_decibel_kapre = magnitude_to_decibel(
x_var, ref_value=1.0, amin=amin, dynamic_range=dynamic_range
)
if dtype == 'float16':
np.testing.assert_allclose(K.eval(x_decibel_kapre), x_decibel_ref, rtol=1e-3, atol=TOL)
else:
np.testing.assert_allclose(K.eval(x_decibel_kapre), x_decibel_ref, atol=TOL)
def __init__(self,
axis=-1,
momentum=0.99,
epsilon=1e-3,
center=True,
scale=True,
beta_initializer='zeros',
gamma_initializer='ones',
moving_mean_initializer='zeros',
moving_variance_initializer='ones',
beta_regularizer=None,
gamma_regularizer=None,
beta_constraint=None,
gamma_constraint=None,
**kwargs):
self._trainable = True
self._trainable_tensor = K.variable(1,
dtype='float32',
name='trainable')
super(BatchNormalization, self).__init__(**kwargs)
self.supports_masking = True
self.axis = axis
self.momentum = momentum
self.epsilon = epsilon
self.center = center
self.scale = scale
self.beta_initializer = initializers.get(beta_initializer)
self.gamma_initializer = initializers.get(gamma_initializer)
self.moving_mean_initializer = initializers.get(
moving_mean_initializer)
self.moving_variance_initializer = initializers.get(
moving_variance_initializer)
self.beta_regularizer = regularizers.get(beta_regularizer)
self.gamma_regularizer = regularizers.get(gamma_regularizer)
self.beta_constraint = constraints.get(beta_constraint)
self.gamma_constraint = constraints.get(gamma_constraint)
def weighted_loss(y_true, y_pred):
"""
Will be used as the metric in model.compile()
---------------------------------------------
Similar to the custom loss function 'weighted_log_loss()' above
but with normalized weights, which should be very similar
to the official competition metric:
https://www.kaggle.com/kambarakun/lb-probe-weights-n-of-positives-scoring
and hence:
sklearn.metrics.log_loss with sample weights
"""
class_weights = K.variable([2., 1., 1., 1., 1., 1.])
eps = K.epsilon()
y_pred = K.clip(y_pred, eps, 1.0 - eps)
loss = -(y_true * K.log(y_pred) + (1.0 - y_true) * K.log(1.0 - y_pred))
loss_samples = _normalized_weighted_average(loss, class_weights)
return K.mean(loss_samples)
def __init__(self,
lr=0.001,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-6,
weight_decay=0.,
exclude_from_weight_decay=None,
**kwargs):
super().__init__(name='LambOptimizer', **kwargs)
with K.name_scope(self.__class__.__name__):
self.iterations = K.variable(0, dtype='int64', name='iterations')
self.lr = K.variable(lr, dtype='float32', name='lr')
self.beta_1 = K.variable(beta_1, dtype='float32', name='beta_1')
self.beta_2 = K.variable(beta_2, dtype='float32', name='beta_2')
self.epsilon = K.variable(epsilon, dtype='float32', name='epsilon')
self.weight_decay = K.variable(weight_decay,
dtype='float32',
name='weight_decay')
self.exclude_from_weight_decay = exclude_from_weight_decay
def build(self, input_shape):
params = dict()
mu = K.variable(0)
sigma = K.variable(0.1, constraint=keras.constraints.non_neg()
) # , constraint=lambda x: max(0, x))
x_up = K.variable(0.1, constraint=keras.constraints.non_neg()
) # , constraint=lambda x: max(0, x))
x_down = K.variable(0.1, constraint=keras.constraints.non_neg()
) # , constraint=lambda x: max(0, x))
alpha_up = K.variable(0.5, constraint=keras.constraints.non_neg()
) # , constraint=lambda x: max(0, x))
alpha_down = K.variable(0.5, constraint=keras.constraints.non_neg()
) # , constraint=lambda x: max(0, x))
self.params = dict(mu=mu,
sigma=sigma,
x_up=x_up,
x_down=x_down,
alpha_up=alpha_up,
alpha_down=alpha_down)
super().build(input_shape)
def __init__(self,
lr=0.001,
beta_1=0.9,
beta_2=0.999,
weight_decay=1e-4,
epsilon=1e-8,
decay=0.,
**kwargs):
# decoupled weight decay (1/6)
super(AdamW, self).__init__(**kwargs)
with K.name_scope(self.__class__.__name__):
self.iterations = K.variable(0, dtype='int64', name='iterations')
self.lr = K.variable(lr, name='lr')
# decoupled weight decay (2/6)
self.init_lr = lr
self.beta_1 = K.variable(beta_1, name='beta_1')
self.beta_2 = K.variable(beta_2, name='beta_2')
self.decay = K.variable(decay, name='decay')
# decoupled weight decay (3/6)
self.wd = K.variable(weight_decay, name='weight_decay')
self.epsilon = epsilon
self.initial_decay = decay
def __init__(self,
lr=0.001,
beta_1=0.9,
beta_2=0.999,
beta_3=0.999,
epsilon=None,
decay=0.,
#amsgrad=False,
**kwargs):
super(AdaMod, self).__init__(name='AdaMod', **kwargs)
with K.name_scope(self.__class__.__name__):
self.iterations = K.variable(0, dtype='int64', name='iterations')
self.lr = K.variable(lr, name='lr')
self.beta_1 = K.variable(beta_1, name='beta_1')
self.beta_2 = K.variable(beta_2, name='beta_2')
self.beta_3 = K.variable(beta_3, name='beta_3')
self.decay = K.variable(decay, name='decay')
#self.amsgrad = K.variable(amsgrad, name='amsgrad')
if epsilon is None:
epsilon = K.epsilon()
self.epsilon = epsilon
self.initial_decay = decay
def __init__(
self,
learning_rate=0.001,
beta_1=0.9,
beta_2=0.999,
weight_decay=1e-4, # decoupled weight decay (1/4)
epsilon=1e-8,
decay=0.,
name='adamw',
**kwargs):
super(AdamW, self).__init__(name, **kwargs)
with K.name_scope(self.__class__.__name__):
self.iterations = K.variable(0, dtype='int64', name='iterations')
self.learning_rate = K.variable(learning_rate,
name='learning_rate')
self.beta_1 = K.variable(beta_1, name='beta_1')
self.beta_2 = K.variable(beta_2, name='beta_2')
self.decay = K.variable(decay, name='decay')
self.wd = K.variable(
weight_decay,
name='weight_decay') # decoupled weight decay (2/4)
self.epsilon = epsilon
self.initial_decay = decay
get_style_fun = K.function([model.input], style_features)
content_targets = get_content_fun([content_image])
# List of list of features
style_targets_list = [get_style_fun([img]) for img in style_images]
# List of batched features
style_targets = []
for l in range(len(args.style_layers)):
batched_features = []
for i in range(nb_styles):
batched_features.append(style_targets_list[i][l][None])
style_targets.append(np.concatenate(batched_features))
if args.init == 'content':
pastiche_image = K.variable(np.repeat(content_image, nb_styles,
axis=0))
else:
if args.init != 'random':
print(
'Could not recognize init arg \'%s\'. Falling back to random.'
% args.init)
pastiche_image = K.variable(
args.std_init *
np.random.randn(nb_styles, *content_image.shape[1:]))
# Store targets as variables
content_targets_dict = {
k: K.variable(v)
for k, v in zip(args.content_layers, content_targets)
}
style_targets_dict = {
def loss(self, y_true, y_pred):
total_loss = K.variable(0)
for idx, loss in enumerate(self.losses):
total_loss += self.loss_wts[idx] * loss(y_true, y_pred)
return total_loss
def build(self, input_shape):
self.input_spec = [InputSpec(shape=input_shape)]
self.conv_layers = {c: [] for c in ['i', 'f', 'c', 'o', 'a', 'ahat']}
for l in range(self.nb_layers):
for c in ['i', 'f', 'c', 'o']:
act = self.LSTM_activation if c == 'c' else self.LSTM_inner_activation
self.conv_layers[c].append(
Conv2D(self.R_stack_sizes[l],
self.R_filt_sizes[l],
padding='same',
activation=act,
data_format=self.data_format))
act = 'relu' if l == 0 else self.A_activation
self.conv_layers['ahat'].append(
Conv2D(self.stack_sizes[l],
self.Ahat_filt_sizes[l],
padding='same',
activation=act,
data_format=self.data_format))
if l < self.nb_layers - 1:
self.conv_layers['a'].append(
Conv2D(self.stack_sizes[l + 1],
self.A_filt_sizes[l],
padding='same',
activation=self.A_activation,
data_format=self.data_format))
self.upsample = UpSampling2D(data_format=self.data_format)
self.pool = MaxPooling2D(data_format=self.data_format)
self._trainable_weights = []
nb_row, nb_col = (
input_shape[-2],
input_shape[-1]) if self.data_format == 'channels_first' else (
input_shape[-3], input_shape[-2])
for c in sorted(self.conv_layers.keys()):
for l in range(len(self.conv_layers[c])):
ds_factor = 2**l
if c == 'ahat':
nb_channels = self.R_stack_sizes[l]
elif c == 'a':
nb_channels = 2 * self.R_stack_sizes[l]
else:
nb_channels = self.stack_sizes[l] * 2 + self.R_stack_sizes[
l]
if l < self.nb_layers - 1:
nb_channels += self.R_stack_sizes[l + 1]
in_shape = (input_shape[0], nb_channels, nb_row // ds_factor,
nb_col // ds_factor)
if self.data_format == 'channels_last':
in_shape = (in_shape[0], in_shape[2], in_shape[3],
in_shape[1])
with K.name_scope('layer_' + c + '_' + str(l)):
self.conv_layers[c][l].build(in_shape)
self._trainable_weights += self.conv_layers[c][
l].trainable_weights
self.states = [None] * self.nb_layers * 3
if self.extrap_start_time is not None:
self.t_extrap = K.variable(
self.extrap_start_time,
int if K.backend() != 'tensorflow' else 'int32')
self.states += [None] * 2 # [previous frame prediction, timestep]
self.built = True
def call(self, x, mask=None):
# TODO: validate input shape
assert (len(x) == 3)
L_flat = x[0]
mu = x[1]
a = x[2]
if self.mode == 'full':
# Create L and L^T matrix, which we use to construct the positive-definite matrix P.
L = None
LT = None
if K.backend() == 'theano':
import theano.tensor as T
import theano
def fn(x, L_acc, LT_acc):
x_ = K.zeros((self.nb_actions, self.nb_actions))
x_ = T.set_subtensor(x_[np.tril_indices(self.nb_actions)], x)
diag = K.exp(T.diag(x_)) + K.epsilon()
x_ = T.set_subtensor(x_[np.diag_indices(self.nb_actions)], diag)
return x_, x_.T
outputs_info = [
K.zeros((self.nb_actions, self.nb_actions)),
K.zeros((self.nb_actions, self.nb_actions)),
]
results, _ = theano.scan(fn=fn, sequences=L_flat, outputs_info=outputs_info)
L, LT = results
elif K.backend() == 'tensorflow':
import tensorflow as tf
# Number of elements in a triangular matrix.
nb_elems = (self.nb_actions * self.nb_actions + self.nb_actions) // 2
# Create mask for the diagonal elements in L_flat. This is used to exponentiate
# only the diagonal elements, which is done before gathering.
diag_indeces = [0]
for row in range(1, self.nb_actions):
diag_indeces.append(diag_indeces[-1] + (row + 1))
diag_mask = np.zeros(1 + nb_elems) # +1 for the leading zero
diag_mask[np.array(diag_indeces) + 1] = 1
diag_mask = K.variable(diag_mask)
# Add leading zero element to each element in the L_flat. We use this zero
# element when gathering L_flat into a lower triangular matrix L.
nb_rows = tf.shape(L_flat)[0]
zeros = tf.expand_dims(tf.tile(K.zeros((1,)), [nb_rows]), 1)
try:
# Old TF behavior.
L_flat = tf.concat(1, [zeros, L_flat])
except (TypeError, ValueError):
# New TF behavior
L_flat = tf.concat([zeros, L_flat], 1)
# Create mask that can be used to gather elements from L_flat and put them
# into a lower triangular matrix.
tril_mask = np.zeros((self.nb_actions, self.nb_actions), dtype='int32')
tril_mask[np.tril_indices(self.nb_actions)] = range(1, nb_elems + 1)
# Finally, process each element of the batch.
init = [
K.zeros((self.nb_actions, self.nb_actions)),
K.zeros((self.nb_actions, self.nb_actions)),
]
def fn(a, x):
# Exponentiate everything. This is much easier than only exponentiating
# the diagonal elements, and, usually, the action space is relatively low.
x_ = K.exp(x) + K.epsilon()
# Only keep the diagonal elements.
x_ *= diag_mask
# Add the original, non-diagonal elements.
x_ += x * (1. - diag_mask)
# Finally, gather everything into a lower triangular matrix.
L_ = tf.gather(x_, tril_mask)
return [L_, tf.transpose(L_)]
tmp = tf.scan(fn, L_flat, initializer=init)
if isinstance(tmp, (list, tuple)):
# TensorFlow 0.10 now returns a tuple of tensors.
L, LT = tmp
else:
# Old TensorFlow < 0.10 returns a shared tensor.
L = tmp[:, 0, :, :]
LT = tmp[:, 1, :, :]
else:
raise RuntimeError('Unknown Keras backend "{}".'.format(K.backend()))
assert L is not None
assert LT is not None
P = K.batch_dot(L, LT)
elif self.mode == 'diag':
if K.backend() == 'theano':
import theano.tensor as T
import theano
def fn(x, P_acc):
x_ = K.zeros((self.nb_actions, self.nb_actions))
x_ = T.set_subtensor(x_[np.diag_indices(self.nb_actions)], x)
return x_
outputs_info = [
K.zeros((self.nb_actions, self.nb_actions)),
]
P, _ = theano.scan(fn=fn, sequences=L_flat, outputs_info=outputs_info)
elif K.backend() == 'tensorflow':
import tensorflow as tf
# Create mask that can be used to gather elements from L_flat and put them
# into a diagonal matrix.
diag_mask = np.zeros((self.nb_actions, self.nb_actions), dtype='int32')
diag_mask[np.diag_indices(self.nb_actions)] = range(1, self.nb_actions + 1)
# Add leading zero element to each element in the L_flat. We use this zero
# element when gathering L_flat into a lower triangular matrix L.
nb_rows = tf.shape(L_flat)[0]
zeros = tf.expand_dims(tf.tile(K.zeros((1,)), [nb_rows]), 1)
try:
# Old TF behavior.
L_flat = tf.concat(1, [zeros, L_flat])
except (TypeError, ValueError):
# New TF behavior
L_flat = tf.concat([zeros, L_flat], 1)
# Finally, process each element of the batch.
def fn(a, x):
x_ = tf.gather(x, diag_mask)
return x_
P = tf.scan(fn, L_flat, initializer=K.zeros((self.nb_actions, self.nb_actions)))
else:
raise RuntimeError('Unknown Keras backend "{}".'.format(K.backend()))
assert P is not None
assert K.ndim(P) == 3
# Combine a, mu and P into a scalar (over the batches). What we compute here is
# -.5 * (a - mu)^T * P * (a - mu), where * denotes the dot-product. Unfortunately
# TensorFlow handles vector * P slightly suboptimal, hence we convert the vectors to
# 1xd/dx1 matrices and finally flatten the resulting 1x1 matrix into a scalar. All
# operations happen over the batch size, which is dimension 0.
prod = K.batch_dot(K.expand_dims(a - mu, 1), P)
prod = K.batch_dot(prod, K.expand_dims(a - mu, -1))
A = -.5 * K.batch_flatten(prod)
assert K.ndim(A) == 2
return A
def styleTransfer(cData, sData, tData):
print(" Building transfer model.")
contentTensor = K.variable(cData)
styleTensor = K.variable(sData)
genTensor = K.placeholder((1, CONTENT_IMG_H, CONTENT_IMG_W, 3))
inputTensor = K.concatenate([contentTensor, styleTensor, genTensor],
axis=0)
model = vgg19.VGG19(include_top=False,
weights="imagenet",
input_tensor=inputTensor) ######
outputDict = dict([(layer.name, layer.output) for layer in model.layers])
print(" VGG19 model loaded.")
loss = 0.0
styleLayerNames = [
"block1_conv1", "block2_conv1", "block3_conv1", "block4_conv1",
"block5_conv1"
]
contentLayerName = "block5_conv2"
print(" Calculating content loss.")
contentLayer = outputDict[contentLayerName]
contentOutput = contentLayer[0, :, :, :]
genOutput = contentLayer[2, :, :, :]
loss += CONTENT_WEIGHT * contentLoss(contentOutput, genOutput)
print(" Calculating style loss.")
for layerName in styleLayerNames:
layer_features = outputDict[layerName]
style_reference_features = layer_features[1, :, :, :]
gen_features = layer_features[2, :, :, :]
loss += (STYLE_WEIGHT / len(styleLayerNames)) * styleLoss(
style_reference_features, gen_features)
loss += TOTAL_WEIGHT * totalLoss(genTensor)
# TODO: Setup gradients or use K.gradients().###########################
grads = K.gradients(loss, genTensor)
outputs = [loss]
if isinstance(grads, (list, tuple)):
outputs += grads
else:
outputs.append(grads)
f_outputs = K.function([genTensor], outputs)
def eval_loss(x):
if K.image_data_format() == "channels_first":
x = x.reshape((1, 3, CONTENT_IMG_H, CONTENT_IMG_W))
else:
x = x.reshape((1, CONTENT_IMG_H, CONTENT_IMG_W, 3))
outs = f_outputs([x])
loss_val = outs[0]
return loss_val
def eval_grads(x):
if K.image_data_format() == "channels_first":
x = x.reshape((1, 3, CONTENT_IMG_H, CONTENT_IMG_W))
else:
x = x.reshape((1, CONTENT_IMG_H, CONTENT_IMG_W, 3))
outs = f_outputs([x])
if len(outs[1:]) == 1:
grad_vals = outs[1].flatten().astype("float64")
else:
grad_vals = np.array(outs[1:]).flatten().astype("float64")
return grad_vals
loadedImg = load_img(CONTENT_IMG_PATH)
x = preprocessData((loadedImg, CONTENT_IMG_H, CONTENT_IMG_W))
print(" Beginning transfer.")
for i in range(TRANSFER_ROUNDS):
print(" Step %d." % i)
#TODO: perform gradient descent using fmin_l_bfgs_b.#######################
x, min, info = fmin_l_bfgs_b(eval_loss,
x.flatten(),
fprime=eval_grads,
maxfun=20)
print(" Loss: %f." % min)
img = deprocessImage(x.copy())
saveFile = "/Users/alex_p/Desktop/CS390NIP/Lab2/" + "transfer_round_%d.png" % i
imsave(saveFile, img) #Uncomment when everything is working right.
print(" Image saved to \"%s\"." % saveFile)
print(" Transfer complete.")
def yolo_head(feats, anchors, num_classes):
"""Convert final layer features to bounding box parameters.
Parameters
----------
feats : tensor
Final convolutional layer features.
anchors : array-like
Anchor box widths and heights.
num_classes : int
Number of target classes.
Returns
-------
box_xy : tensor
x, y box predictions adjusted by spatial location in conv layer.
box_wh : tensor
w, h box predictions adjusted by anchors and conv spatial resolution.
box_conf : tensor
Probability estimate for whether each box contains any object.
box_class_pred : tensor
Probability distribution estimate for each box over class labels.
"""
num_anchors = len(anchors)
# Reshape to batch, height, width, num_anchors, box_params.
anchors_tensor = K.reshape(K.variable(anchors), [1, 1, 1, num_anchors, 2])
# Static implementation for fixed models.
# TODO: Remove or add option for static implementation.
# _, conv_height, conv_width, _ = K.int_shape(feats)
# conv_dims = K.variable([conv_width, conv_height])
# Dynamic implementation of conv dims for fully convolutional model.
conv_dims = K.shape(feats)[1:3] # assuming channels last
# In YOLO the height index is the inner most iteration.
conv_height_index = K.arange(0, stop=conv_dims[0])
conv_width_index = K.arange(0, stop=conv_dims[1])
conv_height_index = K.tile(conv_height_index, [conv_dims[1]])
# TODO: Repeat_elements and tf.split doesn't support dynamic splits.
# conv_width_index = K.repeat_elements(conv_width_index, conv_dims[1], axis=0)
conv_width_index = K.tile(K.expand_dims(conv_width_index, 0),
[conv_dims[0], 1])
conv_width_index = K.flatten(K.transpose(conv_width_index))
conv_index = K.transpose(K.stack([conv_height_index, conv_width_index]))
conv_index = K.reshape(conv_index, [1, conv_dims[0], conv_dims[1], 1, 2])
# print(feats.dtype)
conv_index = K.cast(conv_index,
feats.dtype) # 原本是K.dtype(feats),但是不知道为什么报错
feats = K.reshape(
feats, [-1, conv_dims[0], conv_dims[1], num_anchors, num_classes + 5])
conv_dims = K.cast(K.reshape(conv_dims, [1, 1, 1, 1, 2]), K.dtype(feats))
# Static generation of conv_index:
# conv_index = np.array([_ for _ in np.ndindex(conv_width, conv_height)])
# conv_index = conv_index[:, [1, 0]] # swap columns for YOLO ordering.
# conv_index = K.variable(
# conv_index.reshape(1, conv_height, conv_width, 1, 2))
# feats = Reshape(
# (conv_dims[0], conv_dims[1], num_anchors, num_classes + 5))(feats)
box_xy = K.sigmoid(feats[..., :2])
box_wh = K.exp(feats[..., 2:4])
box_confidence = K.sigmoid(feats[..., 4:5])
box_class_probs = K.softmax(feats[..., 5:])
# Adjust preditions to each spatial grid point and anchor size.
# Note: YOLO iterates over height index before width index.
box_xy = (box_xy + conv_index) / conv_dims
box_wh = box_wh * anchors_tensor / conv_dims
return box_confidence, box_xy, box_wh, box_class_probs
def build(self, input_shape):
self.input_spec = [InputSpec(shape=input_shape)]
gamma = self.gamma_init * np.ones((input_shape[self.axis], ))
self.gamma = K.variable(gamma, name='{}_gamma'.format(self.name))
self.trainable_weights.append([self.gamma])
super(L2Normalization, self).build(input_shape)
# Aqui carregamos e alteramos o tamanho da imagem de estilo
h, w = content_img.shape[1:3]
style_img = load_img_and_preprocess('pintura.jpg', (h, w))
batch_shape = content_img.shape
shape = content_img.shape[1:]
# Aqui criamos o modelo vgg16 com camadas de maxpooling alteradas.
vgg = vgg16_avgpool(shape)
# Aqui criamos o modelo que irá lidar com a imagem principal e selecionamos
# a camada que queremos usar, quando mais profunda a camada, mais borrada a imagem,
# quando mais raza mais nítida a imagem.
# também definimos o alvo(target).
content_model = Model(vgg.input, vgg.layers[10].get_output_at(0))
content_target = K.variable(content_model.predict(content_img))
# Aqui criamos o modelo para a imagem de estilo
# nesse modelo, diferentemente do modelo para a imagem principal onde temos
# apenas 1 output, teremos varios outputs.
symbolic_conv_outputs = [
layer.get_output_at(1) for layer in vgg.layers
if layer.name.endswith('conv1')
]
# Criando o modelo que terá multíplos outputs.
style_model = Model(vgg.input, symbolic_conv_outputs)
# Calculando os alvos que são outputs de cada camada
style_layers_output = [K.variable(y) for y in style_model.predict(style_img)]
def __init__(self):
self.gamma = K.variable(2.)
x = preprocess_input(x)
# we'll use this throughout the rest of the script
batch_shape = x.shape
shape = x.shape[1:]
# see the image
plt.imshow(img)
plt.show()
# make a content model
# try different cutoffs to see the images that result
content_model = VGG16_AvgPool_CutOff(shape, 11)
# make the target
target = K.variable(content_model.predict(x))
# try to match the image
# define our loss in keras
loss = K.mean(K.square(target - content_model.output))
# gradients which are needed by the optimizer
grads = K.gradients(loss, content_model.input)
# just like theano.function
get_loss_and_grads = K.function(inputs=[content_model.input],
outputs=[loss] + grads)
def get_loss_and_grads_wrapper(x_vec):
# scipy's minimizer allows us to pass back
def get_suggested_scheduler(init_lr=5e-5, total_steps=10000, warmup_ratio=0.1): opt_lr = K.variable(init_lr) warmup_steps = int(warmup_ratio * total_steps) warmup = WarmupCallback(opt_lr, init_lr, total_steps, warmup_steps) return warmup, opt_lr
def main():
'''pydot.Dot.create(pydot.Dot())'''
print tf.__version__
samples = list()
for i in range(0,2000):
samples.append((random_process(1.0,0.0),0))
for i in range(0,2000):
adjustCoeff = 0.88+random.random()/10.0
samples.append((random_process(adjustCoeff,0.0),1))
for i in range(0,2000):
drift = 0.0001+(random.random()/10000.0)
samples.append((random_process(1.0,drift),2))
random.shuffle(samples)
test = list()
for i in range(0, 100):
test.append((random_process(1.0, 0.0), 0))
for i in range(0, 100):
adjustCoeff = 0.88 + random.random() / 10.0
test.append((random_process(adjustCoeff, 0.0), 1))
for i in range(0, 100):
drift = 0.0001 + (random.random() / 10000.0)
test.append((random_process(1.0, drift), 2))
j = 0
classes=["RANDOM_WALK","MEAN_REVERTING","DRIFT"]
training = [s[0] for s in samples]
labels = [s[1] for s in samples]
t = K.variable(training)
l = K.variable(labels)
tst = K.variable([ts[0] for ts in test])
tstLabel = K.variable([ts[1] for ts in test])
print t
print t.shape
plt.figure(figsize=(10, 10))
for i in range(25):
j = int(random.random()*3000)
plt.subplot(5, 5, i + 1)
plt.xticks([])
plt.yticks([])
plt.grid(False)
plt.plot(training[j])
plt.xlabel(classes[labels[j]])
plt.show()
model = keras.Sequential([
keras.layers.Dropout(0.2, input_shape=(500,)),
keras.layers.Dense(128, activation='relu'),
keras.layers.Dropout(0.2),
keras.layers.Dense(10, activation='relu'),
keras.layers.Dense(3, activation='softmax')
])
'''keras.utils.plot_model(model, 'model.png')'''
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
history = model.fit(t, l, epochs=10,steps_per_epoch=100)
# Plot training & validation accuracy values
plt.plot(history.history['loss'])
plt.plot(history.history['accuracy'])
plt.title('Model accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Loss', 'Accuracy'], loc='upper left')
plt.show()
test_loss, test_acc = model.evaluate(tst, tstLabel,steps=10)
print('\nTest accuracy:', test_acc)
def build(self, input_shape):
self.input_spec = [InputSpec(shape=input_shape)]
shape = (input_shape[self.axis],)
init_gamma = self.scale * np.ones(shape)
self.gamma = K.variable(init_gamma, name='{}_gamma'.format(self.name))
