def get_labels_from_annotation(annotation_tensor, class_labels):
"""Returns tensor of size (width, height, num_classes) derived from annotation tensor.
The function returns tensor that is of a size (width, height, num_classes) which
is derived from annotation tensor with sizes (width, height) where value at
each position represents a class. The functions requires a list with class
values like [0, 1, 2 ,3] -- they are used to derive labels. Derived values will
be ordered in the same way as the class numbers were provided in the list. Last
value in the aforementioned list represents a value that indicate that the pixel
should be masked out. So, the size of num_classes := len(class_labels) - 1.
Parameters
----------
annotation_tensor : Tensor of size (width, height)
Tensor with class labels for each element
class_labels : list of ints
List that contains the numbers that represent classes. Last
value in the list should represent the number that was used
for masking out.
Returns
-------
labels_2d_stacked : Tensor of size (width, height, num_classes).
Tensor with labels for each pixel.
"""
# Last value in the classes list should show
# which number was used in the annotation to mask out
# the ambigious regions or regions that should not be
# used for training.
# TODO: probably replace class_labels list with some custom object
valid_entries_class_labels = class_labels[:-1]
# Stack the binary masks for each class
labels_2d = map(lambda x: tf.equal(annotation_tensor, x),
valid_entries_class_labels)
# Perform the merging of all of the binary masks into one matrix
labels_2d_stacked = tf.stack(labels_2d, axis=2)
# Convert tf.bool to tf.float
# Later on in the labels and logits will be used
# in tf.softmax_cross_entropy_with_logits() function
# where they have to be of the float type.
labels_2d_stacked_float = tf.to_float(labels_2d_stacked)
return labels_2d_stacked_float
python类softmax_cross_entropy_with_logits()的实例源码
def get_valid_logits_and_labels(annotation_batch_tensor,
logits_batch_tensor,
class_labels):
"""Returns two tensors of size (num_valid_entries, num_classes).
The function converts annotation batch tensor input of the size
(batch_size, height, width) into label tensor (batch_size, height,
width, num_classes) and then selects only valid entries, resulting
in tensor of the size (num_valid_entries, num_classes). The function
also returns the tensor with corresponding valid entries in the logits
tensor. Overall, two tensors of the same sizes are returned and later on
can be used as an input into tf.softmax_cross_entropy_with_logits() to
get the cross entropy error for each entry.
Parameters
----------
annotation_batch_tensor : Tensor of size (batch_size, width, height)
Tensor with class labels for each batch
logits_batch_tensor : Tensor of size (batch_size, width, height, num_classes)
Tensor with logits. Usually can be achived after inference of fcn network.
class_labels : list of ints
List that contains the numbers that represent classes. Last
value in the list should represent the number that was used
for masking out.
Returns
-------
(valid_labels_batch_tensor, valid_logits_batch_tensor) : Two Tensors of size (num_valid_eintries, num_classes).
Tensors that represent valid labels and logits.
"""
labels_batch_tensor = get_labels_from_annotation_batch(annotation_batch_tensor=annotation_batch_tensor,
class_labels=class_labels)
valid_batch_indices = get_valid_entries_indices_from_annotation_batch(annotation_batch_tensor=annotation_batch_tensor,
class_labels=class_labels)
valid_labels_batch_tensor = tf.gather_nd(params=labels_batch_tensor, indices=valid_batch_indices)
valid_logits_batch_tensor = tf.gather_nd(params=logits_batch_tensor, indices=valid_batch_indices)
return valid_labels_batch_tensor, valid_logits_batch_tensor
def get_labels_from_annotation(annotation_tensor, class_labels):
"""Returns tensor of size (width, height, num_classes) derived from annotation tensor.
The function returns tensor that is of a size (width, height, num_classes) which
is derived from annotation tensor with sizes (width, height) where value at
each position represents a class. The functions requires a list with class
values like [0, 1, 2 ,3] -- they are used to derive labels. Derived values will
be ordered in the same way as the class numbers were provided in the list. Last
value in the aforementioned list represents a value that indicate that the pixel
should be masked out. So, the size of num_classes := len(class_labels) - 1.
Parameters
----------
annotation_tensor : Tensor of size (width, height)
Tensor with class labels for each element
class_labels : list of ints
List that contains the numbers that represent classes. Last
value in the list should represent the number that was used
for masking out.
Returns
-------
labels_2d_stacked : Tensor of size (width, height, num_classes).
Tensor with labels for each pixel.
"""
# Last value in the classes list should show
# which number was used in the annotation to mask out
# the ambigious regions or regions that should not be
# used for training.
# TODO: probably replace class_labels list with some custom object
valid_entries_class_labels = class_labels[:-1]
# Stack the binary masks for each class
labels_2d = list(map(lambda x: tf.equal(annotation_tensor, x),
valid_entries_class_labels))
# Perform the merging of all of the binary masks into one matrix
labels_2d_stacked = tf.stack(labels_2d, axis=2)
# Convert tf.bool to tf.float
# Later on in the labels and logits will be used
# in tf.softmax_cross_entropy_with_logits() function
# where they have to be of the float type.
labels_2d_stacked_float = tf.to_float(labels_2d_stacked)
return labels_2d_stacked_float
def get_valid_logits_and_labels(annotation_batch_tensor,
logits_batch_tensor,
class_labels):
"""Returns two tensors of size (num_valid_entries, num_classes).
The function converts annotation batch tensor input of the size
(batch_size, height, width) into label tensor (batch_size, height,
width, num_classes) and then selects only valid entries, resulting
in tensor of the size (num_valid_entries, num_classes). The function
also returns the tensor with corresponding valid entries in the logits
tensor. Overall, two tensors of the same sizes are returned and later on
can be used as an input into tf.softmax_cross_entropy_with_logits() to
get the cross entropy error for each entry.
Parameters
----------
annotation_batch_tensor : Tensor of size (batch_size, width, height)
Tensor with class labels for each batch
logits_batch_tensor : Tensor of size (batch_size, width, height, num_classes)
Tensor with logits. Usually can be achived after inference of fcn network.
class_labels : list of ints
List that contains the numbers that represent classes. Last
value in the list should represent the number that was used
for masking out.
Returns
-------
(valid_labels_batch_tensor, valid_logits_batch_tensor) : Two Tensors of size (num_valid_eintries, num_classes).
Tensors that represent valid labels and logits.
"""
annotation_batch_tensor = tf.py_func(sample, [annotation_batch_tensor], tf.int32)
labels_batch_tensor = get_labels_from_annotation_batch(annotation_batch_tensor=annotation_batch_tensor,
class_labels=class_labels)
valid_batch_indices = get_valid_entries_indices_from_annotation_batch(annotation_batch_tensor=annotation_batch_tensor,
class_labels=class_labels)
valid_labels_batch_tensor = tf.gather_nd(params=labels_batch_tensor, indices=valid_batch_indices)
valid_logits_batch_tensor = tf.gather_nd(params=logits_batch_tensor, indices=valid_batch_indices)
return valid_labels_batch_tensor, valid_logits_batch_tensor
def classifier(x, dropout):
"""
AlexNet fully connected layers definition
Args:
x: tensor of shape [batch_size, width, height, channels]
dropout: probability of non dropping out units
Returns:
fc3: 1000 linear tensor taken just before applying the softmax operation
it is needed to feed it to tf.softmax_cross_entropy_with_logits()
softmax: 1000 linear tensor representing the output probabilities of the image to classify
"""
pool5 = cnn(x)
dim = pool5.get_shape().as_list()
flat_dim = dim[1] * dim[2] * dim[3] # 6 * 6 * 256
flat = tf.reshape(pool5, [-1, flat_dim])
with tf.name_scope('alexnet_classifier') as scope:
with tf.name_scope('alexnet_classifier_fc1') as inner_scope:
wfc1 = tu.weight([flat_dim, 4096], name='wfc1')
bfc1 = tu.bias(0.0, [4096], name='bfc1')
fc1 = tf.add(tf.matmul(flat, wfc1), bfc1)
#fc1 = tu.batch_norm(fc1)
fc1 = tu.relu(fc1)
fc1 = tf.nn.dropout(fc1, dropout)
with tf.name_scope('alexnet_classifier_fc2') as inner_scope:
wfc2 = tu.weight([4096, 4096], name='wfc2')
bfc2 = tu.bias(0.0, [4096], name='bfc2')
fc2 = tf.add(tf.matmul(fc1, wfc2), bfc2)
#fc2 = tu.batch_norm(fc2)
fc2 = tu.relu(fc2)
fc2 = tf.nn.dropout(fc2, dropout)
with tf.name_scope('alexnet_classifier_output') as inner_scope:
wfc3 = tu.weight([4096, 1000], name='wfc3')
bfc3 = tu.bias(0.0, [1000], name='bfc3')
fc3 = tf.add(tf.matmul(fc2, wfc3), bfc3)
softmax = tf.nn.softmax(fc3)
return fc3, softmax
def get_labels_from_annotation(annotation_tensor, class_labels):
"""Returns tensor of size (width, height, num_classes) derived from annotation tensor.
The function returns tensor that is of a size (width, height, num_classes) which
is derived from annotation tensor with sizes (width, height) where value at
each position represents a class. The functions requires a list with class
values like [0, 1, 2 ,3] -- they are used to derive labels. Derived values will
be ordered in the same way as the class numbers were provided in the list. Last
value in the aforementioned list represents a value that indicate that the pixel
should be masked out. So, the size of num_classes := len(class_labels) - 1.
Parameters
----------
annotation_tensor : Tensor of size (width, height)
Tensor with class labels for each element
class_labels : list of ints
List that contains the numbers that represent classes. Last
value in the list should represent the number that was used
for masking out.
Returns
-------
labels_2d_stacked : Tensor of size (width, height, num_classes).
Tensor with labels for each pixel.
"""
# Last value in the classes list should show
# which number was used in the annotation to mask out
# the ambigious regions or regions that should not be
# used for training.
# TODO: probably replace class_labels list with some custom object
valid_entries_class_labels = class_labels[:-1]
# Stack the binary masks for each class
labels_2d = map(lambda x: tf.equal(annotation_tensor, x),
valid_entries_class_labels)
# Perform the merging of all of the binary masks into one matrix
labels_2d_stacked = tf.stack(labels_2d, axis=2)
# Convert tf.bool to tf.float
# Later on in the labels and logits will be used
# in tf.softmax_cross_entropy_with_logits() function
# where they have to be of the float type.
labels_2d_stacked_float = tf.to_float(labels_2d_stacked)
return labels_2d_stacked_float
def get_valid_logits_and_labels(annotation_batch_tensor,
logits_batch_tensor,
class_labels):
"""Returns two tensors of size (num_valid_entries, num_classes).
The function converts annotation batch tensor input of the size
(batch_size, height, width) into label tensor (batch_size, height,
width, num_classes) and then selects only valid entries, resulting
in tensor of the size (num_valid_entries, num_classes). The function
also returns the tensor with corresponding valid entries in the logits
tensor. Overall, two tensors of the same sizes are returned and later on
can be used as an input into tf.softmax_cross_entropy_with_logits() to
get the cross entropy error for each entry.
Parameters
----------
annotation_batch_tensor : Tensor of size (batch_size, width, height)
Tensor with class labels for each batch
logits_batch_tensor : Tensor of size (batch_size, width, height, num_classes)
Tensor with logits. Usually can be achived after inference of fcn network.
class_labels : list of ints
List that contains the numbers that represent classes. Last
value in the list should represent the number that was used
for masking out.
Returns
-------
(valid_labels_batch_tensor, valid_logits_batch_tensor) : Two Tensors of size (num_valid_eintries, num_classes).
Tensors that represent valid labels and logits.
"""
labels_batch_tensor = get_labels_from_annotation_batch(annotation_batch_tensor=annotation_batch_tensor,
class_labels=class_labels)
valid_batch_indices = get_valid_entries_indices_from_annotation_batch(annotation_batch_tensor=annotation_batch_tensor,
class_labels=class_labels)
valid_labels_batch_tensor = tf.gather_nd(params=labels_batch_tensor, indices=valid_batch_indices)
valid_logits_batch_tensor = tf.gather_nd(params=logits_batch_tensor, indices=valid_batch_indices)
return valid_labels_batch_tensor, valid_logits_batch_tensor
