def testBatchNormScopeDoesHasIsTrainingWhenItsNotNone(self):
sc = mobilenet_v1.mobilenet_v1_arg_scope(is_training=True)
self.assertIn('is_training',
sc[slim.arg_scope_func_key(slim.batch_norm)])
sc = mobilenet_v1.mobilenet_v1_arg_scope(is_training=False)
self.assertIn('is_training',
sc[slim.arg_scope_func_key(slim.batch_norm)])
sc = mobilenet_v1.mobilenet_v1_arg_scope()
self.assertIn('is_training',
sc[slim.arg_scope_func_key(slim.batch_norm)])
def build_model():
"""Build the mobilenet_v1 model for evaluation.
Returns:
g: graph with rewrites after insertion of quantization ops and batch norm
folding.
eval_ops: eval ops for inference.
variables_to_restore: List of variables to restore from checkpoint.
"""
g = tf.Graph()
with g.as_default():
inputs, labels = imagenet_input(is_training=False)
scope = mobilenet_v1.mobilenet_v1_arg_scope(
is_training=False, weight_decay=0.0)
with slim.arg_scope(scope):
logits, _ = mobilenet_v1.mobilenet_v1(
inputs,
is_training=False,
depth_multiplier=FLAGS.depth_multiplier,
num_classes=FLAGS.num_classes)
if FLAGS.quantize:
tf.contrib.quantize.create_eval_graph()
eval_ops = metrics(logits, labels)
return g, eval_ops
def extract_features(self, preprocessed_inputs):
"""Extract features from preprocessed inputs.
Args:
preprocessed_inputs: a [batch, height, width, channels] float tensor
representing a batch of images.
Returns:
feature_maps: a list of tensors where the ith tensor has shape
[batch, height_i, width_i, depth_i]
"""
preprocessed_inputs = shape_utils.check_min_image_dim(
min_dim=33, image_tensor=preprocessed_inputs)
feature_map_layout = {
'from_layer':
['Conv2d_11_pointwise', 'Conv2d_13_pointwise', '', '', '', ''],
'layer_depth': [-1, -1, 512, 256, 256, 128],
'use_explicit_padding':
self._use_explicit_padding,
'use_depthwise':
self._use_depthwise,
}
with tf.variable_scope('MobilenetV1',
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=None, regularize_depthwise=True)):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
# TODO(skligys): Enable fused batch norm once quantization supports it.
with slim.arg_scope([slim.batch_norm], fused=False):
_, image_features = mobilenet_v1.mobilenet_v1_base(
ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
final_endpoint='Conv2d_13_pointwise',
min_depth=self._min_depth,
depth_multiplier=self._depth_multiplier,
use_explicit_padding=self._use_explicit_padding,
scope=scope)
with slim.arg_scope(self._conv_hyperparams_fn()):
# TODO(skligys): Enable fused batch norm once quantization supports it.
with slim.arg_scope([slim.batch_norm], fused=False):
feature_maps = feature_map_generators.multi_resolution_feature_maps(
feature_map_layout=feature_map_layout,
depth_multiplier=self._depth_multiplier,
min_depth=self._min_depth,
insert_1x1_conv=True,
image_features=image_features)
return feature_maps.values()
def build_model():
"""Builds graph for model to train with rewrites for quantization.
Returns:
g: Graph with fake quantization ops and batch norm folding suitable for
training quantized weights.
train_tensor: Train op for execution during training.
"""
g = tf.Graph()
with g.as_default(), tf.device(
tf.train.replica_device_setter(FLAGS.ps_tasks)):
inputs, labels = imagenet_input(is_training=True)
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(is_training=True)):
logits, _ = mobilenet_v1.mobilenet_v1(
inputs,
is_training=True,
depth_multiplier=FLAGS.depth_multiplier,
num_classes=FLAGS.num_classes)
tf.losses.softmax_cross_entropy(labels, logits)
# Call rewriter to produce graph with fake quant ops and folded batch norms
# quant_delay delays start of quantization till quant_delay steps, allowing
# for better model accuracy.
if FLAGS.quantize:
tf.contrib.quantize.create_training_graph(
quant_delay=get_quant_delay())
total_loss = tf.losses.get_total_loss(name='total_loss')
# Configure the learning rate using an exponential decay.
num_epochs_per_decay = 2.5
imagenet_size = 1271167
decay_steps = int(imagenet_size / FLAGS.batch_size *
num_epochs_per_decay)
learning_rate = tf.train.exponential_decay(
get_learning_rate(),
tf.train.get_or_create_global_step(),
decay_steps,
_LEARNING_RATE_DECAY_FACTOR,
staircase=True)
opt = tf.train.GradientDescentOptimizer(learning_rate)
train_tensor = slim.learning.create_train_op(total_loss, optimizer=opt)
slim.summaries.add_scalar_summary(total_loss, 'total_loss', 'losses')
slim.summaries.add_scalar_summary(learning_rate, 'learning_rate',
'training')
return g, train_tensor
def _extract_proposal_features(self, preprocessed_inputs, scope):
"""Extracts first stage RPN features.
Args:
preprocessed_inputs: A [batch, height, width, channels] float32 tensor
representing a batch of images.
scope: A scope name.
Returns:
rpn_feature_map: A tensor with shape [batch, height, width, depth]
activations: A dictionary mapping feature extractor tensor names to
tensors
Raises:
InvalidArgumentError: If the spatial size of `preprocessed_inputs`
(height or width) is less than 33.
ValueError: If the created network is missing the required activation.
"""
preprocessed_inputs.get_shape().assert_has_rank(4)
preprocessed_inputs = shape_utils.check_min_image_dim(
min_dim=33, image_tensor=preprocessed_inputs)
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=self._train_batch_norm,
weight_decay=self._weight_decay)):
with tf.variable_scope('MobilenetV1',
reuse=self._reuse_weights) as scope:
params = {}
if self._skip_last_stride:
params[
'conv_defs'] = _get_mobilenet_conv_no_last_stride_defs(
conv_depth_ratio_in_percentage=self.
_conv_depth_ratio_in_percentage)
_, activations = mobilenet_v1.mobilenet_v1_base(
preprocessed_inputs,
final_endpoint='Conv2d_11_pointwise',
min_depth=self._min_depth,
depth_multiplier=self._depth_multiplier,
scope=scope,
**params)
return activations['Conv2d_11_pointwise'], activations
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
"""Extracts second stage box classifier 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 (unused).
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.
"""
net = proposal_feature_maps
conv_depth = 1024
if self._skip_last_stride:
conv_depth_ratio = float(
self._conv_depth_ratio_in_percentage) / 100.0
conv_depth = int(float(conv_depth) * conv_depth_ratio)
depth = lambda d: max(int(d * 1.0), 16)
with tf.variable_scope('MobilenetV1', reuse=self._reuse_weights):
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=self._train_batch_norm,
weight_decay=self._weight_decay)):
with slim.arg_scope([slim.conv2d, slim.separable_conv2d],
padding='SAME'):
net = slim.separable_conv2d(net,
depth(conv_depth), [3, 3],
depth_multiplier=1,
stride=2,
scope='Conv2d_12_pointwise')
return slim.separable_conv2d(net,
depth(conv_depth), [3, 3],
depth_multiplier=1,
stride=1,
scope='Conv2d_13_pointwise')
def extract_features(self, preprocessed_inputs):
"""Extract features from preprocessed inputs.
Args:
preprocessed_inputs: a [batch, height, width, channels] float tensor
representing a batch of images.
Returns:
feature_maps: a list of tensors where the ith tensor has shape
[batch, height_i, width_i, depth_i]
"""
preprocessed_inputs = shape_utils.check_min_image_dim(
33, preprocessed_inputs)
with tf.variable_scope('MobilenetV1',
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=None, regularize_depthwise=True)):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
_, image_features = mobilenet_v1.mobilenet_v1_base(
ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
final_endpoint='Conv2d_13_pointwise',
min_depth=self._min_depth,
depth_multiplier=self._depth_multiplier,
use_explicit_padding=self._use_explicit_padding,
scope=scope)
with slim.arg_scope(self._conv_hyperparams_fn()):
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=0,
num_layers=6,
image_features={
'image_features': image_features['Conv2d_11_pointwise']
})
return feature_maps.values()
def testBatchNormScopeDoesNotHaveIsTrainingWhenItsSetToNone(self):
sc = mobilenet_v1.mobilenet_v1_arg_scope(is_training=None)
self.assertNotIn('is_training',
sc[slim.arg_scope_func_key(slim.batch_norm)])
def extract_features(self, preprocessed_inputs):
"""Extract features from preprocessed inputs.
Args:
preprocessed_inputs: a [batch, height, width, channels] float tensor
representing a batch of images.
Returns:
feature_maps: a list of tensors where the ith tensor has shape
[batch, height_i, width_i, depth_i]
Raises:
ValueError: if image height or width are not 256 pixels.
"""
image_shape = preprocessed_inputs.get_shape()
image_shape.assert_has_rank(4)
image_height = image_shape[1].value
image_width = image_shape[2].value
if image_height is None or image_width is None:
shape_assert = tf.Assert(
tf.logical_and(tf.equal(tf.shape(preprocessed_inputs)[1], 256),
tf.equal(tf.shape(preprocessed_inputs)[2],
256)),
['image size must be 256 in both height and width.'])
with tf.control_dependencies([shape_assert]):
preprocessed_inputs = tf.identity(preprocessed_inputs)
elif image_height != 256 or image_width != 256:
raise ValueError(
'image size must be = 256 in both height and width;'
' image dim = %d,%d' % (image_height, image_width))
feature_map_layout = {
'from_layer':
['Conv2d_11_pointwise', 'Conv2d_13_pointwise', '', '', ''],
'layer_depth': [-1, -1, 512, 256, 256],
'conv_kernel_size': [-1, -1, 3, 3, 2],
'use_explicit_padding':
self._use_explicit_padding,
'use_depthwise':
self._use_depthwise,
}
with tf.variable_scope('MobilenetV1',
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(is_training=None)):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
# TODO(skligys): Enable fused batch norm once quantization supports it.
with slim.arg_scope([slim.batch_norm], fused=False):
_, image_features = mobilenet_v1.mobilenet_v1_base(
ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
final_endpoint='Conv2d_13_pointwise',
min_depth=self._min_depth,
depth_multiplier=self._depth_multiplier,
use_explicit_padding=self._use_explicit_padding,
scope=scope)
with slim.arg_scope(self._conv_hyperparams_fn()):
# TODO(skligys): Enable fused batch norm once quantization supports it.
with slim.arg_scope([slim.batch_norm], fused=False):
feature_maps = feature_map_generators.multi_resolution_feature_maps(
feature_map_layout=feature_map_layout,
depth_multiplier=self._depth_multiplier,
min_depth=self._min_depth,
insert_1x1_conv=True,
image_features=image_features)
#return feature_maps.values()
return list(feature_maps.values())
def extract_features(self, preprocessed_inputs):
"""Extract features from preprocessed inputs.
Args:
preprocessed_inputs: a [batch, height, width, channels] float tensor
representing a batch of images.
Returns:
feature_maps: a list of tensors where the ith tensor has shape
[batch, height_i, width_i, depth_i]
"""
preprocessed_inputs = shape_utils.check_min_image_dim(
33, preprocessed_inputs)
with tf.variable_scope('MobilenetV1',
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=None, regularize_depthwise=True)):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams
else context_manager.IdentityContextManager()):
_, image_features = mobilenet_v1.mobilenet_v1_base(
ops.pad_to_multiple(preprocessed_inputs, self._pad_to_multiple),
final_endpoint='Conv2d_13_pointwise',
min_depth=self._min_depth,
depth_multiplier=self._depth_multiplier,
conv_defs=self._conv_defs,
use_explicit_padding=self._use_explicit_padding,
scope=scope)
depth_fn = lambda d: max(int(d * self._depth_multiplier), self._min_depth)
with slim.arg_scope(self._conv_hyperparams_fn()):
with tf.variable_scope('fpn', reuse=self._reuse_weights):
feature_blocks = [
'Conv2d_3_pointwise', 'Conv2d_5_pointwise', 'Conv2d_11_pointwise',
'Conv2d_13_pointwise'
]
base_fpn_max_level = min(self._fpn_max_level, 5)
feature_block_list = []
for level in range(self._fpn_min_level, base_fpn_max_level + 1):
feature_block_list.append(feature_blocks[level - 2])
fpn_features = feature_map_generators.fpn_top_down_feature_maps(
[(key, image_features[key]) for key in feature_block_list],
depth=depth_fn(self._additional_layer_depth),
use_depthwise=self._use_depthwise,
use_explicit_padding=self._use_explicit_padding)
feature_maps = []
for level in range(self._fpn_min_level, base_fpn_max_level + 1):
feature_maps.append(fpn_features['top_down_{}'.format(
feature_blocks[level - 2])])
last_feature_map = fpn_features['top_down_{}'.format(
feature_blocks[base_fpn_max_level - 2])]
# Construct coarse features
padding = 'VALID' if self._use_explicit_padding else 'SAME'
kernel_size = 3
for i in range(base_fpn_max_level + 1, self._fpn_max_level + 1):
if self._use_depthwise:
conv_op = functools.partial(
slim.separable_conv2d, depth_multiplier=1)
else:
conv_op = slim.conv2d
if self._use_explicit_padding:
last_feature_map = ops.fixed_padding(
last_feature_map, kernel_size)
last_feature_map = conv_op(
last_feature_map,
num_outputs=depth_fn(self._additional_layer_depth),
kernel_size=[kernel_size, kernel_size],
stride=2,
padding=padding,
scope='bottom_up_Conv2d_{}'.format(i - base_fpn_max_level + 13))
feature_maps.append(last_feature_map)
return feature_maps
