def build_nasnet_large(images, num_classes,
is_training=True,
final_endpoint=None,
config=None,
current_step=None):
"""Build NASNet Large model for the ImageNet Dataset."""
hparams = (large_imagenet_config() if config is None
else copy.deepcopy(config))
_update_hparams(hparams, is_training)
if tf.test.is_gpu_available() and hparams.data_format == 'NHWC':
tf.compat.v1.logging.info(
'A GPU is available on the machine, consider using NCHW '
'data format for increased speed on GPU.')
if hparams.data_format == 'NCHW':
images = tf.transpose(a=images, perm=[0, 3, 1, 2])
# Calculate the total number of cells in the network
# Add 2 for the reduction cells
total_num_cells = hparams.num_cells + 2
# If ImageNet, then add an additional two for the stem cells
total_num_cells += 2
normal_cell = nasnet_utils.NasNetANormalCell(
hparams.num_conv_filters, hparams.drop_path_keep_prob,
total_num_cells, hparams.total_training_steps,
hparams.use_bounded_activation)
reduction_cell = nasnet_utils.NasNetAReductionCell(
hparams.num_conv_filters, hparams.drop_path_keep_prob,
total_num_cells, hparams.total_training_steps,
hparams.use_bounded_activation)
with arg_scope([slim.dropout, nasnet_utils.drop_path, slim.batch_norm],
is_training=is_training):
with arg_scope([slim.avg_pool2d,
slim.max_pool2d,
slim.conv2d,
slim.batch_norm,
slim.separable_conv2d,
nasnet_utils.factorized_reduction,
nasnet_utils.global_avg_pool,
nasnet_utils.get_channel_index,
nasnet_utils.get_channel_dim],
data_format=hparams.data_format):
return _build_nasnet_base(images,
normal_cell=normal_cell,
reduction_cell=reduction_cell,
num_classes=num_classes,
hparams=hparams,
is_training=is_training,
stem_type='imagenet',
final_endpoint=final_endpoint,
current_step=current_step)
def build_nasnet_cifar(images, num_classes,
is_training=True,
config=None):
"""Build NASNet model for the Cifar Dataset."""
hparams = cifar_config() if config is None else copy.deepcopy(config)
_update_hparams(hparams, is_training)
if tf.test.is_gpu_available() and hparams.data_format == 'NHWC':
tf.logging.info('A GPU is available on the machine, consider using NCHW '
'data format for increased speed on GPU.')
if hparams.data_format == 'NCHW':
images = tf.transpose(images, [0, 3, 1, 2])
# Calculate the total number of cells in the network
# Add 2 for the reduction cells
total_num_cells = hparams.num_cells + 2
normal_cell = nasnet_utils.NasNetANormalCell(
hparams.num_conv_filters, hparams.drop_path_keep_prob,
total_num_cells, hparams.total_training_steps)
reduction_cell = nasnet_utils.NasNetAReductionCell(
hparams.num_conv_filters, hparams.drop_path_keep_prob,
total_num_cells, hparams.total_training_steps)
with arg_scope([slim.dropout, nasnet_utils.drop_path, slim.batch_norm],
is_training=is_training):
with arg_scope([slim.avg_pool2d,
slim.max_pool2d,
slim.conv2d,
slim.batch_norm,
slim.separable_conv2d,
nasnet_utils.factorized_reduction,
nasnet_utils.global_avg_pool,
nasnet_utils.get_channel_index,
nasnet_utils.get_channel_dim],
data_format=hparams.data_format):
return _build_nasnet_base(images,
normal_cell=normal_cell,
reduction_cell=reduction_cell,
num_classes=num_classes,
hparams=hparams,
is_training=is_training,
stem_type='cifar')
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
"""Extracts second stage box classifier features.
This function reconstructs the "second half" of the NASNet-A
network after the part defined in `_extract_proposal_features`.
Args:
proposal_feature_maps: A 4-D float tensor with shape
[batch_size * self.max_num_proposals, crop_height, crop_width, depth]
representing the feature map cropped to each proposal.
scope: A scope name.
Returns:
proposal_classifier_features: A 4-D float tensor with shape
[batch_size * self.max_num_proposals, height, width, depth]
representing box classifier features for each proposal.
"""
del scope
# Note that we always feed into 2 layers of equal depth
# where the first N channels corresponds to previous hidden layer
# and the second N channels correspond to the final hidden layer.
hidden_previous, hidden = tf.split(proposal_feature_maps, 2, axis=3)
# Note that what follows is largely a copy of build_nasnet_large() within
# nasnet.py. We are copying to minimize code pollution in slim.
# TODO(shlens,skornblith): Determine the appropriate drop path schedule.
# For now the schedule is the default (1.0->0.7 over 250,000 train steps).
hparams = nasnet.large_imagenet_config()
if not self._is_training:
hparams.set_hparam('drop_path_keep_prob', 1.0)
# Calculate the total number of cells in the network
# -- Add 2 for the reduction cells.
total_num_cells = hparams.num_cells + 2
# -- And add 2 for the stem cells for ImageNet training.
total_num_cells += 2
normal_cell = nasnet_utils.NasNetANormalCell(
hparams.num_conv_filters, hparams.drop_path_keep_prob,
total_num_cells, hparams.total_training_steps)
reduction_cell = nasnet_utils.NasNetAReductionCell(
hparams.num_conv_filters, hparams.drop_path_keep_prob,
total_num_cells, hparams.total_training_steps)
with arg_scope([slim.dropout, nasnet_utils.drop_path],
is_training=self._is_training):
with arg_scope([slim.batch_norm],
is_training=self._train_batch_norm):
with arg_scope([
slim.avg_pool2d, slim.max_pool2d, slim.conv2d,
slim.batch_norm, slim.separable_conv2d,
nasnet_utils.factorized_reduction,
nasnet_utils.global_avg_pool,
nasnet_utils.get_channel_index,
nasnet_utils.get_channel_dim
],
data_format=hparams.data_format):
# This corresponds to the cell number just past 'Cell_11' used by
# by _extract_proposal_features().
start_cell_num = 12
# Note that this number equals:
# start_cell_num + 2 stem cells + 1 reduction cell
true_cell_num = 15
with slim.arg_scope(nasnet.nasnet_large_arg_scope()):
net = _build_nasnet_base(hidden_previous,
hidden,
normal_cell=normal_cell,
reduction_cell=reduction_cell,
hparams=hparams,
true_cell_num=true_cell_num,
start_cell_num=start_cell_num)
proposal_classifier_features = net
return proposal_classifier_features
