def __init__(self, num_classes, train_layers=None, weights_path='DEFAULT'):
"""Create the graph of the densenet_161 model.
"""
# Parse input arguments into class variables
if weights_path == 'DEFAULT':
self.WEIGHTS_PATH = "./pre_trained_models/densenet_161.ckpt"
else:
self.WEIGHTS_PATH = weights_path
self.train_layers = train_layers
with tf.variable_scope("input"):
self.image_size = densenet.densenet161.default_image_size
self.x_input = tf.placeholder(tf.float32, [None, self.image_size, self.image_size, 3], name="x_input")
self.y_input = tf.placeholder(tf.float32, [None, num_classes], name="y_input")
self.learning_rate = tf.placeholder(tf.float32, name="learning_rate")
# train
with arg_scope(densenet.densenet_arg_scope()):
self.logits, _ = densenet.densenet161(self.x_input,
num_classes=num_classes,
is_training=True,
reuse=tf.AUTO_REUSE
)
# validation
with arg_scope(densenet.densenet_arg_scope()):
self.logits_val, _ = densenet.densenet161(self.x_input,
num_classes=num_classes,
is_training=False,
reuse=tf.AUTO_REUSE,
)
with tf.name_scope("loss"):
self.loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=self.logits, labels=self.y_input))
self.loss_val = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=self.logits_val, labels=self.y_input))
with tf.name_scope("train"):
self.global_step = tf.Variable(0, name="global_step", trainable=False)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
var_list = [v for v in tf.trainable_variables() if v.name.split('/')[-2] in train_layers or v.name.split('/')[-3] in train_layers ]
gradients = tf.gradients(self.loss, var_list)
self.grads_and_vars = list(zip(gradients, var_list))
optimizer = tf.train.GradientDescentOptimizer(self.learning_rate)
with tf.control_dependencies(update_ops):
self.train_op = optimizer.apply_gradients(grads_and_vars=self.grads_and_vars, global_step=self.global_step)
with tf.name_scope("probability"):
self.probability = tf.nn.softmax(self.logits_val, name="probability")
with tf.name_scope("prediction"):
self.prediction = tf.argmax(self.logits_val, 1, name="prediction")
with tf.name_scope("accuracy"):
correct_prediction = tf.equal(self.prediction, tf.argmax(self.y_input, 1))
self.accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"), name="accuracy")
def apply_network_img(data_dense, batch_size, is_training_net,
is_training_drop, net):
if net == 'D': #DenseNet-161
with slim.arg_scope(densenet.densenet_arg_scope()):
logits, _ = densenet.densenet161(data_dense,
num_classes=None,
is_training_batch=is_training_net,
is_training_drop=is_training_drop)
elif net == 'V': #VGG-16
with slim.arg_scope(vgg.vgg_arg_scope()):
logits, _ = vgg.vgg_16(data_dense,
num_classes=None,
is_training=is_training_drop,
spatial_squeeze=False,
classification=False,
max_pool=False)
return logits
def load_model_tf(parameters):
# Setup model params
if (parameters["device_type"] == "gpu") and tf.test.is_gpu_available():
device_str = "/device:GPU:{}".format(parameters["gpu_number"])
else:
device_str = "/cpu:0"
# Setup Graph
graph = tf.Graph()
with graph.as_default():
with tf.device(device_str):
x = tf.placeholder(tf.float32, [None, 256, 256, 3])
with slim.arg_scope(
densenet_arg_scope(weight_decay=0.0, data_format='NHWC')):
densenet121_net, end_points = densenet121(
x,
num_classes=parameters["number_of_classes"],
data_format='NHWC',
is_training=False,
)
y_logits = densenet121_net[:, 0, 0, :]
y = tf.nn.softmax(y_logits)
# Load weights
sess = tf.Session(graph=graph,
config=tf.ConfigProto(gpu_options=tf.GPUOptions(
per_process_gpu_memory_fraction=0.333)))
with open(parameters["tf_torch_weights_map_path"]) as f:
tf_torch_weights_map = json.loads(f.read())
with sess.as_default():
torch_weights = torch.load(parameters["initial_parameters"])
match_dict = construct_densenet_match_dict(
tf_variables=tf_utils.get_tf_variables(graph,
batch_norm_key="BatchNorm"),
torch_weights=torch_weights,
tf_torch_weights_map=tf_torch_weights_map)
sess.run(tf_utils.construct_weight_assign_ops(match_dict))
return sess, x, y
def testAllEndPointsShapes(self):
batch_size = 5
height, width = 224, 224
num_classes = 1000
inputs = tf.random_uniform((batch_size, height, width, 3))
with slim.arg_scope(densenet.densenet_arg_scope()):
logits, end_points = densenet.densenet(inputs, num_classes)
endpoints_shapes = {'input_layer': [batch_size, 56, 56, 48],
'dense_block1': [batch_size, 56, 56, 192],
'transition_layer1': [batch_size, 28, 28, 96],
'dense_block2': [batch_size, 28, 28, 384],
'transition_layer2': [batch_size, 14, 14, 192],
'dense_block3': [batch_size, 14, 14, 768],
'transition_layer3': [batch_size, 7, 7, 384],
'dense_block4': [batch_size, 7, 7, 768],
'classify_layer': [batch_size, num_classes]}
self.assertItemsEqual(endpoints_shapes.keys(), end_points.keys())
for endpoint_name in endpoints_shapes:
expected_shape = endpoints_shapes[endpoint_name]
self.assertTrue(endpoint_name in end_points)
self.assertListEqual(end_points[endpoint_name].get_shape().as_list(),
expected_shape)
def _extract_proposal_features(self, preprocessed_inputs, scope):
"""Extracts first stage RPN features.
Extracts features using the first half of the Inception Resnet v2 network.
We construct the network in `align_feature_maps=True` mode, which means
that all VALID paddings in the network are changed to SAME padding so that
the feature maps are aligned.
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]
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.
"""
if len(preprocessed_inputs.get_shape().as_list()) != 4:
raise ValueError(
'`preprocessed_inputs` must be 4 dimensional, got a '
'tensor of shape %s' % preprocessed_inputs.get_shape())
with slim.arg_scope(
densenet.densenet_arg_scope(weight_decay=self._weight_decay)):
# Forces is_training to False to disable batch norm update.
with slim.arg_scope([slim.batch_norm], is_training=False):
with tf.variable_scope('DenseNet',
reuse=self._reuse_weights) as scope:
rpn_feature_map = densenet.Dense_net(preprocessed_inputs,
nb_blocks=2,
filters=24,
is_training=True,
dropout_ratio=0.2)
return rpn_feature_map
def main(_):
if not FLAGS.dataset_dir:
raise ValueError(
'You must supply the dataset directory with --dataset_dir')
tf.logging.set_verbosity(tf.logging.INFO)
with tf.Graph().as_default():
######################
# Select the dataset #
######################
dataset = dataset_factory.get_dataset(FLAGS.dataset_name,
FLAGS.dataset_split_name,
FLAGS.dataset_dir)
####################
# Select the model #
####################
network_fn = nets_factory.get_network_fn(
FLAGS.model_name,
num_classes=(dataset.num_classes - FLAGS.labels_offset),
is_training=False)
##############################################################
# Create a dataset provider that loads data from the dataset #
##############################################################
provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
shuffle=False,
common_queue_capacity=2 * FLAGS.batch_size,
common_queue_min=FLAGS.batch_size)
[image, file_name] = provider.get(['image', 'file_name'])
#####################################
# Select the preprocessing function #
#####################################
preprocessing_name = FLAGS.preprocessing_name or FLAGS.model_name
image_preprocessing_fn = preprocessing_factory.get_preprocessing(
preprocessing_name, is_training=False)
eval_image_size = FLAGS.eval_image_size or network_fn.default_image_size
image = image_preprocessing_fn(image, eval_image_size, eval_image_size)
images, file_names = tf.train.batch(
[image, file_name],
batch_size=FLAGS.batch_size,
num_threads=FLAGS.num_preprocessing_threads,
capacity=5 * FLAGS.batch_size)
####################
# Define the model #
####################
# logits, end_points = network_fn(images)
# predictions = tf.argmax(logits, 1)
# TODO(sguada) use num_epochs=1
num_batches_per_epoch = math.ceil(dataset.num_samples /
float(FLAGS.batch_size))
num_steps_per_epoch = num_batches_per_epoch
with slim.arg_scope(densenet_arg_scope()):
logits, end_points = network_fn(images)
print(FLAGS.checkpoint_path)
if tf.gfile.IsDirectory(FLAGS.checkpoint_path):
checkpoint_path = tf.train.latest_checkpoint(FLAGS.checkpoint_path)
else:
checkpoint_path = FLAGS.checkpoint_path
tf.logging.info('Evaluating %s' % checkpoint_path)
# #get all the variables to restore from the checkpoint file and create the saver function to restore
variables_to_restore = slim.get_variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
def restore_fn(sess):
return saver.restore(sess, checkpoint_path)
#Just define the metrics to track without the loss or whatsoever
# predictions = tf.argmax(logits, 1)
# predictions = end_points[]
top_k_pred = tf.nn.top_k(end_points['predictions'], k=5)
# top_k_pred = tf.nn.top_k(logits, k=5)
#Create the global step and an increment op for monitoring
global_step = get_or_create_global_step()
global_step_op = tf.assign(
global_step, global_step + 1
) #no apply_gradient method so manually increasing the global_step
file_names_all = []
predictions_all = []
all_key = {}
c_key = {}
#Create a evaluation step function
def eval_step(sess, top_k_pred, file_names, global_step):
'''
Simply takes in a session, runs the metrics op and some logging information.
'''
start_time = time.time()
# global_step_count, predictions_ = sess.run([global_step_op, predictions])
p_file_name = sess.run(file_names)
global_step_count = sess.run(global_step_op)
time_elapsed = time.time() - start_time
#Log some information
logging.info('Global Step %s: Streaming Accuracy: (%.2f sec/step)',
global_step_count, time_elapsed)
return p_file_name
#Get your supervisor
sv = tf.train.Supervisor(logdir=FLAGS.test_dir,
summary_op=None,
saver=None,
init_fn=restore_fn)
#Now we are ready to run in one session
#config = tf.ConfigProto(device_count={'GPU':0}) # mask GPUs visible to the session so it falls back on CPU
with sv.managed_session() as sess:
for step in range(num_steps_per_epoch * num_epochs):
sess.run(sv.global_step)
#print vital information every start of the epoch as always
file_names_ = eval_step(sess,
top_k_pred,
file_names,
global_step=sv.global_step)
my_file_name = str(file_names_[0], 'utf-8')
logging.info('my_file_name=' + my_file_name)
if my_file_name not in all_key:
all_key[my_file_name] = my_file_name
else:
c_key[my_file_name] = my_file_name
#At the end of all the evaluation, show the final accuracy
logging.info('总处理不重复的文件数:' + str(len(all_key)))
logging.info('总处理重复的文件数:' + str(len(c_key)))
def _extract_box_classifier_features(self, proposal_feature_maps, scope):
"""Extracts second stage box classifier features.
This function reconstructs the "second half" of the Densenet
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.
"""
with tf.variable_scope('DenseNet', reuse=self._reuse_weights):
with slim.arg_scope(
densenet.densenet_arg_scope(
weight_decay=self._weight_decay)):
# Forces is_training to False to disable batch norm update.
with slim.arg_scope([slim.batch_norm], is_training=False):
with slim.arg_scope(
[slim.conv2d, slim.max_pool2d, slim.avg_pool2d],
stride=1,
padding='SAME'):
with tf.variable_scope('Mixed_7a'):
with tf.variable_scope('Branch_0'):
tower_conv = slim.conv2d(proposal_feature_maps,
256,
1,
scope='Conv2d_0a_1x1')
tower_conv_1 = slim.conv2d(
tower_conv,
384,
3,
stride=2,
padding='VALID',
scope='Conv2d_1a_3x3')
with tf.variable_scope('Branch_1'):
tower_conv1 = slim.conv2d(
proposal_feature_maps,
256,
1,
scope='Conv2d_0a_1x1')
tower_conv1_1 = slim.conv2d(
tower_conv1,
288,
3,
stride=2,
padding='VALID',
scope='Conv2d_1a_3x3')
with tf.variable_scope('Branch_2'):
tower_conv2 = slim.conv2d(
proposal_feature_maps,
256,
1,
scope='Conv2d_0a_1x1')
tower_conv2_1 = slim.conv2d(
tower_conv2, 288, 3, scope='Conv2d_0b_3x3')
tower_conv2_2 = slim.conv2d(
tower_conv2_1,
320,
3,
stride=2,
padding='VALID',
scope='Conv2d_1a_3x3')
with tf.variable_scope('Branch_3'):
tower_pool = slim.max_pool2d(
proposal_feature_maps,
3,
stride=2,
padding='VALID',
scope='MaxPool_1a_3x3')
net = tf.concat([
tower_conv_1, tower_conv1_1, tower_conv2_2,
tower_pool
], 3)
net = slim.repeat(net,
9,
inception_resnet_v2.block8,
scale=0.20)
net = inception_resnet_v2.block8(net,
activation_fn=None)
proposal_classifier_features = slim.conv2d(
net, 1536, 1, scope='Conv2d_7b_1x1')
return proposal_classifier_features
if not data:
break
image_files.append(data[:-1])
# image_files = random.sample(image_names, sample_num)
for i in image_files:
# GT_dict[i] = i.split('/')[3]
GT_num[i] = i.split('\\')[3]
image_input = tf.read_file(i)
image = tf.image.decode_jpeg(image_input, channels=3)
user_images.append(image)
processed_image = densenet_preprocessing.preprocess_image(image, image_size, image_size, is_training=False)
user_processed_images.append(processed_image)
processed_images = tf.expand_dims(processed_image, 0)
with slim.arg_scope(densenet.densenet_arg_scope()):
logits, _ = densenet.densenet121(user_processed_images, num_classes=5, is_training=False)
probabilities = tf.nn.softmax(logits)
init_fn = slim.assign_from_checkpoint_fn(
os.path.join(checkpoints_dir, 'model.ckpt-21031'),
slim.get_model_variables('densenet121'))
with tf.Session() as sess:
init_fn(sess)
probabilities = sess.run(probabilities)
# names = os.listdir("tmp/captcha/test_photos2")
names = ['bus', 'car', 'cat', 'dog', 'ship']
names.sort()
#names=['normal', 'adenoma', 'adenocarcinoma']
def __init__(self, options):
num_classes = options.NUM_CLASSES
with tf.variable_scope("input"):
self.image_size = options.IMAGE_SIZE
self.x_input = tf.placeholder(
tf.float32, [None, self.image_size, self.image_size, 3],
name="x_input")
self.y_input = tf.placeholder(tf.float32, [None, num_classes],
name="y_input")
self.learning_rate = tf.placeholder(tf.float32,
name="learning_rate")
self.keep_prob = None
if options.PHASE == 'train':
if train_layers == 'default':
self.train_layers = self.DEFAULT_TRAIN_LAYERS
else:
self.train_layers = train_layers
# train
with arg_scope(densenet.densenet_arg_scope()):
self.logits, _ = densenet.densenet121(self.x_input,
num_classes=num_classes,
is_training=True,
reuse=tf.AUTO_REUSE)
self.logits = tf.squeeze(self.logits, [1, 2])
# validation
with arg_scope(densenet.densenet_arg_scope()):
self.logits_val, _ = densenet.densenet121(
self.x_input,
num_classes=num_classes,
is_training=False,
reuse=tf.AUTO_REUSE)
self.logits_val = tf.squeeze(self.logits_val, [1, 2])
with tf.name_scope("loss"):
self.loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits_v2(
logits=self.logits, labels=self.y_input))
self.loss_val = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits_v2(
logits=self.logits_val, labels=self.y_input))
with tf.name_scope("train"):
self.global_step = tf.Variable(0,
name="global_step",
trainable=False)
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
var_list = [
v for v in tf.trainable_variables()
if v.name.split('/')[-2] in self.train_layers
or v.name.split('/')[-3] in self.train_layers
]
gradients = tf.gradients(self.loss, var_list)
self.grads_and_vars = list(zip(gradients, var_list))
# optimizer = tf.train.GradientDescentOptimizer(self.learning_rate)
opt_name = options.OPTIMIZER
if opt_name == 'sgd':
optimizer = tf.train.GradientDescentOptimizer(
self.learning_rate)
elif opt_name == 'adam':
optimizer = tf.train.AdamOptimizer(self.learning_rate)
else:
raise ValueError('Optimizer not supported')
with tf.control_dependencies(update_ops):
self.train_op = optimizer.apply_gradients(
grads_and_vars=self.grads_and_vars,
global_step=self.global_step)
else:
with arg_scope(densenet.densenet_arg_scope()):
self.logits_val, _ = densenet.densenet121(
self.x_input,
num_classes=num_classes,
is_training=False,
reuse=tf.AUTO_REUSE)
self.logits_val = tf.squeeze(self.logits_val, [1, 2])
with tf.name_scope("probability"):
self.probability = tf.nn.softmax(self.logits_val,
name="probability")
with tf.name_scope("prediction"):
self.prediction = tf.argmax(self.logits_val, 1, name="prediction")
with tf.name_scope("accuracy"):
correct_prediction = tf.equal(self.prediction,
tf.argmax(self.y_input, 1))
self.accuracy = tf.reduce_mean(tf.cast(correct_prediction,
"float"),
name="accuracy")
print(self.logits.shape.as_list())
print(self.logits_val.shape.as_list())
