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 conv_defs is not provided or from_layer does not meet the
size requirement.
"""
if not self._conv_defs:
raise ValueError('Must provide backbone conv defs.')
if len(self._from_layer) != 2:
raise ValueError('SSD input feature names are not provided.')
preprocessed_inputs = shape_utils.check_min_image_dim(
33, preprocessed_inputs)
feature_map_layout = {
'from_layer':
[self._from_layer[0], self._from_layer[1], '', '', '', ''],
'layer_depth': [-1, -1, 512, 256, 256, 128],
'use_depthwise':
self._use_depthwise,
'use_explicit_padding':
self._use_explicit_padding,
}
with tf.variable_scope(self._scope_name,
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v3.training_scope(is_training=None, bn_decay=0.9997)), \
slim.arg_scope(
[mobilenet.depth_multiplier], min_depth=self._min_depth):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
_, image_features = mobilenet_v3.mobilenet_base(
ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
conv_defs=self._conv_defs,
final_endpoint=self._from_layer[1],
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.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 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)
feature_map_layout = {
'from_layer':
['layer_15/expansion_output', 'layer_19', '', '', '', ''],
'layer_depth': [-1, -1, 512, 256, 256, 128],
'use_depthwise': self._use_depthwise,
'use_explicit_padding': self._use_explicit_padding,
}
if self._num_layers == 7:
feature_map_layout['from_layer'] += ['']
feature_map_layout['layer_depth'] += [64]
with tf.variable_scope('MobilenetV3',
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v3.training_scope(is_training=None, bn_decay=0.9997)), \
slim.arg_scope(
[mobilenet.depth_multiplier], min_depth=self._min_depth):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
_, image_features = mobilenet_v3.mobilenet_base(
ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
final_endpoint='layer_19',
depth_multiplier=self._depth_multiplier,
use_explicit_padding=self._use_explicit_padding,
squeeze_excitation=self._squeeze_excitation,
scope=scope)
with slim.arg_scope(self._conv_hyperparams_fn()):
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 _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('MobilenetV3', reuse=self._reuse_weights):
with slim.arg_scope(
mobilenet_v3.training_scope(
is_training=None, 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_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 = shape_utils.check_min_image_dim(
33, preprocessed_inputs)
with tf.variable_scope('MobilenetV3',
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v3.training_scope(is_training=None, bn_decay=0.9997)), \
slim.arg_scope(
[mobilenet.depth_multiplier], min_depth=self._min_depth):
_, activations = mobilenet_v3.mobilenet_base(
preprocessed_inputs,
conv_defs=mobilenet_v3.V3_LARGE_DETECTION,
final_endpoint='layer_17',
depth_multiplier=self._depth_multiplier,
scope=scope)
return activations['layer_17'], activations
tf.reset_default_graph()
# For simplicity we just decode jpeg inside tensorflow.
# But one can provide any input obviously.
file_input = tf.placeholder(tf.string, ())
image = tf.image.decode_jpeg(tf.read_file(file_input))
images = tf.expand_dims(image, 0)
images = tf.cast(images, tf.float32) / 128. - 1
images.set_shape((None, None, None, 3))
images = tf.image.resize_images(images, (224, 224))
# 测试MobileNetV2,MobileNetV3
# Note: arg_scope is optional for inference.
with tf.contrib.slim.arg_scope(mobilenet_v3.training_scope(is_training=False)):
logits, endpoints = mobilenet_v3.mobilenet(images)
# 测试MobileNetV1
#with tf.contrib.slim.arg_scope(mobilenet_v1.mobilenet_v1_arg_scope(is_training=False)):
#logits, endpoints = mobilenet_v1.mobilenet_v1(images)
# Restore using exponential moving average since it produces (1.5-2%) higher
# accuracy
ema = tf.train.ExponentialMovingAverage(0.999)
vars = ema.variables_to_restore()
saver = tf.train.Saver(vars)
from datasets import imagenet
import os
import time
(nets.MobileNet100v3largemini, (224, 224, 3),
'mobilenet_100_v3_large_mini', mobilenet_v3.large_minimalistic, 1.0,
'v3-large-minimalistic_224_1.0_float/ema/model-342500'),
(nets.MobileNet100v3smallmini, (224, 224, 3),
'mobilenet_100_v3_small_mini', mobilenet_v3.small_minimalistic, 1.0,
'v3-small-minimalistic_224_1.0_float/ema/model-498000'),
]
for (net, shape, model_name, net_slim, alpha, checkpoint) in models_list:
with tf.Graph().as_default():
inputs = tf.compat.v1.placeholder(tf.float32, [None] + list(shape))
with tf.contrib.slim.arg_scope(
mobilenet_v3.training_scope(is_training=False)):
logits, endpoints = net_slim(inputs, depth_multiplier=alpha)
saver = tf.compat.v1.train.Saver()
weights_tfslim = tf.compat.v1.get_collection(
tf.compat.v1.GraphKeys.GLOBAL_VARIABLES)
model = net(inputs, scope='a')
img = nets.utils.load_img('/home/taehoonlee/tensornets/cat.png',
target_size=int(shape[0] * 8 / 7),
crop_size=shape[0])
with tf.compat.v1.Session() as sess: