def loss(self, logits, labels, images):
'''
Loss to be minimised by the neural network
:param logits: The output of the neural network before the softmax
:param labels: The ground truth labels in standard (i.e. not one-hot) format
:param images: The input
:return: The total loss including weight decay, the loss without weight decay, only the weight decay
'''
nlabels = self.config.nlabels
loss_type = self.config.loss_type
weight_decay = self.config.weight_decay
loss_hyper_params = self.config.loss_hyper_params if hasattr(
self.config, 'loss_hyper_param') else [1, 0.2]
if nlabels > 2:
oh_labels = tf.one_hot(labels, depth=nlabels)
else:
oh_labels = tf.cast(labels, tf.float32)
with tf.variable_scope('weights_norm'):
weights_norm = tf.reduce_sum(input_tensor=weight_decay * tf.stack([
tf.nn.l2_loss(ii)
for ii in tf.get_collection('weight_variables')
]),
name='weights_norm')
if loss_type == 'weighted_crossentropy':
segmentation_loss = losses.pixel_wise_cross_entropy_loss_weighted(
logits, oh_labels, class_weights=[0.1, 0.3, 0.3, 0.3])
elif loss_type == 'crossentropy':
segmentation_loss = losses.pixel_wise_cross_entropy_loss(
logits, oh_labels)
elif loss_type == 'dice':
segmentation_loss = losses.dice_loss(logits,
oh_labels,
only_foreground=False)
elif loss_type == 'dice_onlyfg':
segmentation_loss = losses.dice_loss(logits,
oh_labels,
only_foreground=True)
elif loss_type == 'crossentropy_and_dice':
segmentation_loss = loss_hyper_params[0] * losses.pixel_wise_cross_entropy_loss(logits, oh_labels) \
+ loss_hyper_params[1] * losses.dice_loss(logits, oh_labels)
elif loss_type == 'dice_cc':
segmentation_loss = loss_hyper_params[0] * losses.dice_loss(logits, oh_labels) \
+ loss_hyper_params[1] * losses.connected_component_loss(logits, oh_labels)
else:
raise ValueError('Unknown loss: %s' % loss_type)
total_loss = tf.add(segmentation_loss, weights_norm)
return total_loss, segmentation_loss, weights_norm
def loss(logits, labels, nlabels, loss_type, weight_decay=0.0):
'''
Loss to be minimised by the neural network
:param logits: The output of the neural network before the softmax
:param labels: The ground truth labels in standard (i.e. not one-hot) format
:param nlabels: The number of GT labels
:param loss_type: Can be 'weighted_crossentropy'/'crossentropy'/'dice'/'dice_onlyfg'/'crossentropy_and_dice'
:param weight_decay: The weight for the L2 regularisation of the network paramters
:return: The total loss including weight decay, the loss without weight decay, only the weight decay
'''
labels = tf.one_hot(labels, depth=nlabels)
with tf.compat.v1.variable_scope('weights_norm'):
weights_norm = tf.reduce_sum(input_tensor=weight_decay * tf.stack([
tf.nn.l2_loss(ii)
for ii in tf.compat.v1.get_collection('weight_variables')
]),
name='weights_norm')
if loss_type == 'weighted_crossentropy':
segmentation_loss = losses.pixel_wise_cross_entropy_loss_weighted(
logits, labels, class_weights=[0.076, 0.308, 0.308, 0.308])
elif loss_type == 'crossentropy':
segmentation_loss = losses.pixel_wise_cross_entropy_loss(
logits, labels)
elif loss_type == 'dice':
segmentation_loss = losses.dice_loss(logits,
labels,
only_foreground=False)
elif loss_type == 'dice_onlyfg':
segmentation_loss = losses.dice_loss(logits,
labels,
only_foreground=True)
elif loss_type == 'crossentropy_and_dice':
segmentation_loss = config.alfa * losses.pixel_wise_cross_entropy_loss_weighted(
logits, labels, class_weights=[0.076, 0.308, 0.308, 0.308]
) + config.beta * losses.dice_loss(
logits, labels, only_foreground=True)
else:
raise ValueError('Unknown loss: %s' % loss_type)
total_loss = tf.add(segmentation_loss, weights_norm)
return total_loss, segmentation_loss, weights_norm
def loss(self):
y_for_loss = tf.one_hot(self.y_pl, depth=self.nlabels)
if self.exp_config.loss_type == 'crossentropy':
task_loss = losses.cross_entropy_loss(logits=self.l_pl_,
labels=y_for_loss)
elif self.exp_config.loss_type == 'dice_micro':
task_loss = losses.dice_loss(logits=self.l_pl_,
labels=y_for_loss,
mode='micro')
elif self.exp_config.loss_type == 'dice_macro':
task_loss = losses.dice_loss(logits=self.l_pl_,
labels=y_for_loss,
mode='macro')
elif self.exp_config.loss_type == 'dice_macro_robust':
task_loss = losses.dice_loss(logits=self.l_pl_,
labels=y_for_loss,
mode='macro_robust')
else:
raise ValueError("Unknown loss_type in exp_config: '%s'" %
self.exp_config.loss_type)
return task_loss
def loss(logits, labels, nlabels, loss_type):
'''
Loss to be minimised by the neural network
:param logits: The output of the neural network before the softmax
:param labels: The ground truth labels in standard (i.e. not one-hot) format
:param nlabels: The number of GT labels
:param loss_type: Can be 'crossentropy'/'dice'/'crossentropy_and_dice'
:return: The segmentation
'''
labels = tf.one_hot(labels, depth=nlabels)
if loss_type == 'crossentropy':
segmentation_loss = losses.pixel_wise_cross_entropy_loss(logits, labels)
elif loss_type == 'dice':
segmentation_loss = losses.dice_loss(logits, labels)
elif loss_type == 'crossentropy_and_dice':
segmentation_loss = losses.pixel_wise_cross_entropy_loss(logits, labels) + 0.2*losses.dice_loss(logits, labels)
else:
raise ValueError('Unknown loss: %s' % loss_type)
return segmentation_loss
def loss(logits,
labels,
nlabels,
loss_type,
mask_for_loss_within_mask=None,
are_labels_1hot=False):
'''
Loss to be minimised by the neural network
:param logits: The output of the neural network before the softmax
:param labels: The ground truth labels in standard (i.e. not one-hot) format
:param nlabels: The number of GT labels
:param loss_type: Can be 'crossentropy'/'dice'/
:return: The segmentation
'''
if are_labels_1hot is False:
labels = tf.one_hot(labels, depth=nlabels)
if loss_type == 'crossentropy':
segmentation_loss = losses.pixel_wise_cross_entropy_loss(
logits, labels)
elif loss_type == 'crossentropy_reverse':
predicted_probabilities = tf.nn.softmax(logits)
segmentation_loss = losses.pixel_wise_cross_entropy_loss_using_probs(
predicted_probabilities, labels)
elif loss_type == 'dice':
segmentation_loss = losses.dice_loss(logits, labels)
elif loss_type == 'dice_within_mask':
if mask_for_loss_within_mask is not None:
segmentation_loss = losses.dice_loss_within_mask(
logits, labels, mask_for_loss_within_mask)
else:
raise ValueError('Unknown loss: %s' % loss_type)
return segmentation_loss
