def build_model(data_tensor, reuse, training):
"""Create the hgru from Learning long-range..."""
with tf.variable_scope('cnn', reuse=reuse):
with tf.variable_scope('input', reuse=reuse):
conv_aux = {
'pretrained': os.path.join('weights',
'gabors_for_contours_7.npy'),
'pretrained_key': 's1',
'nonlinearity': 'square'
}
x = conv.conv_layer(bottom=data_tensor,
name='gabor_input',
stride=[1, 1, 1, 1],
padding='SAME',
trainable=training,
use_bias=True,
aux=conv_aux)
layer_hgru = hgru.hGRU('hgru_1',
x_shape=x.get_shape().as_list(),
timesteps=8,
h_ext=15,
strides=[1, 1, 1, 1],
padding='SAME',
aux={
'reuse': False,
'constrain': False
},
train=training)
h2 = layer_hgru.build(x)
h2 = normalization.batch(
bottom=h2,
reuse=reuse,
# renorm=True,
name='hgru_bn',
training=training)
with tf.variable_scope('readout_1', reuse=reuse):
activity = conv.conv_layer(bottom=h2,
name='pre_readout_conv',
num_filters=2,
kernel_size=1,
trainable=training,
use_bias=False)
pool_aux = {'pool_type': 'max'}
activity = pooling.global_pool(bottom=activity,
name='pre_readout_pool',
aux=pool_aux)
activity = normalization.batch(
bottom=activity,
reuse=reuse,
# renorm=True,
name='readout_1_bn',
training=training)
with tf.variable_scope('readout_2', reuse=reuse):
activity = tf.layers.flatten(activity, name='flat_readout')
activity = tf.layers.dense(inputs=activity, units=2)
return activity, h2
def build_model(data_tensor,
reuse,
training,
output_shape,
data_format='NHWC'):
"""Create the hgru from Learning long-range..."""
if isinstance(output_shape, list):
output_shape = output_shape[-1]
elif isinstance(output_shape, dict):
output_shape = output_shape['output']
output_normalization_type = 'batch_norm_original'
ff_kernel_size = (5, 5)
ff_nl = tf.nn.elu
data_tensor, long_data_format = tf_fun.interpret_data_format(
data_tensor=data_tensor, data_format=data_format)
# Build model
with tf.variable_scope('gammanet', reuse=reuse):
conv_aux = {
'pretrained': os.path.join('weights',
'gabors_for_contours_11.npy'),
'pretrained_key': 's1',
'nonlinearity': 'square'
}
activity = conv.conv_layer(bottom=data_tensor,
name='gabor_input',
stride=[1, 1, 1, 1],
padding='SAME',
trainable=training,
use_bias=True,
aux=conv_aux)
layer_hgru = hgru.hGRU('hgru_1',
x_shape=activity.get_shape().as_list(),
timesteps=8,
h_ext=15,
strides=[1, 1, 1, 1],
padding='SAME',
aux={
'reuse': False,
'constrain': False
},
train=training)
h2 = layer_hgru.build(activity)
h2 = normalization.batch_contrib(bottom=h2,
name='hgru_bn',
training=training)
mask = np.load('weights/cardena_mask.npy')[None, :, :, None]
activity = h2 * mask
with tf.variable_scope('cv_readout', reuse=reuse):
activity = tf.reduce_mean(activity, reduction_indices=[1, 2])
activity = tf.layers.dense(activity, output_shape)
if long_data_format is 'channels_first':
activity = tf.transpose(activity, (0, 2, 3, 1))
extra_activities = {}
if activity.dtype != tf.float32:
activity = tf.cast(activity, tf.float32)
# return [activity, h_deep], extra_activities
return activity, extra_activities
def build_model(data_tensor, reuse, training, output_shape):
"""Create the hgru from Learning long-range..."""
if isinstance(output_shape, list):
output_shape = output_shape[0]
with tf.variable_scope('cnn', reuse=reuse):
# Add input
so_filters = np.load(os.path.join('weights',
'so_filters.npy')).squeeze().reshape(
11, 11, 3, 8 * 4)[..., 0:32:4]
so_filter_tensor = tf.get_variable(name='so_filters',
initializer=so_filters,
trainable=training)
so_bias = tf.get_variable(name='so_bias',
initializer=tf.zeros(so_filters.shape[-1]),
trainable=training)
in_emb = tf.nn.conv2d(input=data_tensor,
filter=so_filter_tensor,
strides=[1, 1, 1, 1],
padding='SAME',
name='so')
in_emb = tf.nn.bias_add(in_emb, so_bias)
# in_emb = in_emb ** 2
layer_hgru = hgru.hGRU('hgru_1',
x_shape=in_emb.get_shape().as_list(),
timesteps=8,
h_ext=21,
strides=[1, 1, 1, 1],
padding='SAME',
aux={
'reuse': False,
'constrain': False,
'nonnegative': True,
'while_loop': False,
'mirror_horizontal': True,
'horizontal_dilations': [1, 2, 2, 1]
},
train=training)
h2 = layer_hgru.build(in_emb)
h2 = normalization.batch(bottom=h2,
renorm=False,
name='hgru_bn',
training=training)
activity = conv.readout_layer(
activity=h2,
reuse=reuse,
training=training,
pool_type='max', # 'select',
output_shape=output_shape,
features=so_filters.shape[-1])
extra_activities = {'activity': h2}
return activity, extra_activities
def build_model(data_tensor, reuse, training, output_shape):
"""Create the hgru from Learning long-range..."""
if isinstance(output_shape, list):
output_shape = output_shape[0]
with tf.variable_scope('cnn', reuse=reuse):
# Add input
so_filters = np.load(
'/media/data_cifs/clicktionary/pretrained_weights/alexnet.npy'
).item()
so_filter_tensor = tf.get_variable(name='so_filters',
initializer=so_filters['conv1'][0],
trainable=training)
so_bias = tf.get_variable(name='so_bias',
initializer=so_filters['conv1'][1],
trainable=training)
in_emb = tf.nn.conv2d(input=data_tensor,
filter=so_filter_tensor,
strides=[1, 1, 1, 1],
padding='SAME',
name='so')
in_emb = tf.nn.bias_add(in_emb, so_bias)
# in_emb = in_emb ** 2
layer_hgru = hgru.hGRU('hgru_1',
x_shape=in_emb.get_shape().as_list(),
timesteps=8,
h_ext=11,
strides=[1, 1, 1, 1],
padding='SAME',
aux={
'reuse': False,
'constrain': False,
'nonnegative': True,
'while_loop': False,
'horizontal_dilations': [1, 2, 2, 1]
},
train=training)
h2 = layer_hgru.build(in_emb)
h2 = normalization.batch(bottom=h2,
renorm=False,
name='hgru_bn',
training=training)
activity = conv.readout_layer(
activity=h2,
reuse=reuse,
training=training,
pool_type='max', # 'select',
output_shape=output_shape,
features=output_shape)
extra_activities = {'activity': h2}
return activity, extra_activities
def build_model(data_tensor, reuse, training, output_shape):
"""Create the hgru from Learning long-range..."""
if isinstance(output_shape, list):
output_shape = output_shape[0]
reduction_filters = 16
with tf.variable_scope('cnn', reuse=reuse):
# Add input
so_filters = np.load(os.path.join('weights',
'so_filters.npy')).squeeze().reshape(
11, 11, 3, 8 * 4)
in_emb = tf.nn.conv2d(input=data_tensor,
filter=so_filters,
strides=[1, 1, 1, 1],
padding='SAME',
name='so')
in_emb = tf.layers.conv2d(inputs=in_emb,
filters=reduction_filters,
kernel_size=(1, 1))
in_emb = in_emb**2
layer_hgru = hgru.hGRU('hgru_1',
x_shape=in_emb.get_shape().as_list(),
timesteps=8,
h_ext=11,
strides=[1, 1, 1, 1],
padding='SAME',
aux={
'reuse': False,
'constrain': False,
'nonnegative': True,
'while_loop': False
},
train=training)
h2 = layer_hgru.build(in_emb)
h2 = normalization.batch(bottom=h2,
renorm=False,
name='hgru_bn',
training=training)
activity = conv.readout_layer(
activity=h2,
reuse=reuse,
training=training,
pool_type='max', # 'select',
output_shape=output_shape,
features=reduction_filters)
extra_activities = {'activity': h2}
return activity, extra_activities
def build_model(data_tensor, reuse, training, output_shape):
"""Create the hgru from Learning long-range..."""
if isinstance(output_shape, list):
output_shape = output_shape[0]
with tf.variable_scope('cnn', reuse=reuse):
# Add input
conv_aux = {
'pretrained': os.path.join('weights',
'gabors_for_contours_11.npy'),
'pretrained_key': 's1',
'nonlinearity': 'square'
}
activity = conv.conv_layer(bottom=data_tensor,
name='gabor_input',
stride=[1, 1, 1, 1],
padding='SAME',
trainable=training,
use_bias=True,
aux=conv_aux)
layer_hgru = hgru.hGRU('hgru_1',
x_shape=activity.get_shape().as_list(),
timesteps=8,
h_ext=15,
strides=[1, 1, 1, 1],
padding='SAME',
aux={
'reuse': False,
'constrain': False
},
train=training)
h2 = layer_hgru.build(activity)
h2 = normalization.batch(bottom=h2,
renorm=False,
name='hgru_bn',
training=training)
activity = conv.readout_layer(activity=h2,
reuse=reuse,
training=training,
output_shape=output_shape)
extra_activities = {'activity': h2}
return activity, extra_activities
def build_model(data_tensor, reuse, training, output_shape):
"""Create the hgru from Learning long-range..."""
if isinstance(output_shape, list):
output_shape = output_shape[0]
with tf.variable_scope('cnn', reuse=reuse):
# Add input
in_emb = conv.skinny_input_layer(X=data_tensor,
reuse=reuse,
training=training,
features=24,
conv_activation=tf.nn.elu,
conv_kernel_size=7,
pool=False,
name='l0')
layer_hgru = hgru.hGRU('hgru_1',
x_shape=in_emb.get_shape().as_list(),
timesteps=8,
h_ext=9,
strides=[1, 1, 1, 1],
padding='SAME',
aux={
'reuse': False,
'constrain': False,
'nonnegative': True
},
train=training)
h2 = layer_hgru.build(in_emb)
h2 = normalization.batch(bottom=h2,
renorm=True,
name='hgru_bn',
training=training)
activity = conv.conv_layer(bottom=h2,
name='pre_readout_conv',
num_filters=output_shape['output'],
kernel_size=1,
trainable=training,
use_bias=True)
extra_activities = {}
return activity, extra_activities
def build_model(data_tensor, reuse, training, output_shape):
"""Create the hgru from Learning long-range..."""
if isinstance(output_shape, list):
output_shape = output_shape[0]
with tf.variable_scope('cnn', reuse=reuse):
with tf.variable_scope('input', reuse=reuse):
x = tf.layers.conv2d(
inputs=data_tensor,
filters=24,
kernel_size=11,
name='l0',
strides=(1, 1),
padding='same',
activation=tf.nn.relu,
trainable=training,
use_bias=True)
gauss_mask = generate_gaussian_masks(x, [20, 40], num_centers_list=[30, 15])
o_x = masked_substitution_noise(x, gauss_mask)
layer_hgru = hgru.hGRU(
'hgru_1',
x_shape=o_x.get_shape().as_list(),
timesteps=8,
h_ext=15,
strides=[1, 1, 1, 1],
padding='SAME',
aux={'reuse': False, 'constrain': False},
train=training)
h2 = layer_hgru.build(o_x)
# h2 = normalization.batch(
# bottom=h2,
# renorm=False,
# name='hgru_bn',
# training=training)
return h2 - o_x, {}