def evaluation(logits, labels, images, nlabels, loss_type):
'''
A function for evaluating the performance of the netwrok on a minibatch. This function returns the loss and the
current foreground Dice score, and also writes example segmentations and imges to to tensorboard.
:param logits: Output of network before softmax
:param labels: Ground-truth label mask
:param images: Input image mini batch
:param nlabels: Number of labels in the dataset
:param loss_type: Which loss should be evaluated
:return: The loss without weight decay, the foreground dice of a minibatch
'''
mask = tf.arg_max(tf.nn.softmax(logits, dim=-1), dimension=-1) # was 3
mask_gt = labels
tf.summary.image('example_gt', prepare_tensor_for_summary(mask_gt, mode='mask', nlabels=nlabels))
tf.summary.image('example_pred', prepare_tensor_for_summary(mask, mode='mask', nlabels=nlabels))
tf.summary.image('example_zimg', prepare_tensor_for_summary(images, mode='image'))
total_loss, nowd_loss, weights_norm = loss(logits, labels, nlabels=nlabels, loss_type=loss_type)
cdice_structures = losses.per_structure_dice(logits, tf.one_hot(labels, depth=nlabels))
cdice_foreground = cdice_structures[:,1:]
cdice = tf.reduce_mean(cdice_foreground)
return nowd_loss, cdice
def evaluate_losses(logits, labels, nlabels, loss_type):
'''
A function to compute various loss measures to compare the predicted and ground truth annotations
'''
nowd_loss = loss(logits, labels, nlabels=nlabels, loss_type=loss_type)
cdice_structures = losses.per_structure_dice(logits, tf.one_hot(labels, depth=nlabels))
cdice_foreground = tf.slice(cdice_structures, (0,1), (-1,-1))
cdice = tf.reduce_mean(cdice_foreground)
return nowd_loss, cdice, cdice_structures, cdice_foreground
def evaluation(logits, labels, images, nlabels, loss_type, weak_supervision=False, cnn_threshold=None, include_bg=True):
'''
A function for evaluating the performance of the netwrok on a minibatch. This function returns the loss and the
current foreground Dice score, and also writes example segmentations and imges to to tensorboard.
:param logits: Output of network before softmax
:param labels: Ground-truth label mask
:param images: Input image mini batch
:param nlabels: Number of labels in the dataset
:param loss_type: Which loss should be evaluated
:param include_bg: Whether or not to include the definitely_background label
:return: The loss without weight decay, the foreground dice of a minibatch
'''
softmax = tf.nn.softmax(logits, dim=-1)
mask = tf.arg_max(softmax, dimension=-1) # was 3
mask_gt = labels
if weak_supervision:
if not cnn_threshold is None:
threshold = tf.constant(cnn_threshold, dtype=tf.float32)
####
s = tf.multiply(tf.ones(shape=[4, 212, 212, 1]), threshold)
softmax = tf.concat([s, softmax[..., 1:]], axis=-1)
mask_thresholded = tf.arg_max(softmax, dimension=-1)
if include_bg:
# include_bg typically true for training and false for validation
# this involves creating two tensorflow variables
# which will result in two copies of the images appearing on tensorboard
# which is very annoying
tf.summary.image('example_gt', prepare_tensor_for_summary(mask_gt, mode='mask', nlabels=nlabels))
tf.summary.image('example_pred', prepare_tensor_for_summary(mask, mode='mask', nlabels=nlabels))
if not cnn_threshold is None:
tf.summary.image('example_thresholded_pred', prepare_tensor_for_summary(mask_thresholded, mode='mask', nlabels=nlabels))
tf.summary.image('example_zimg', prepare_tensor_for_summary(images, mode='image'))
else:
tf.summary.image('example_gt', prepare_tensor_for_summary(mask_gt, mode='mask', nlabels=nlabels))
tf.summary.image('example_pred', prepare_tensor_for_summary(mask, mode='mask', nlabels=nlabels))
tf.summary.image('example_zimg', prepare_tensor_for_summary(images, mode='image'))
total_loss, nowd_loss, weights_norm = loss(logits, labels, nlabels=nlabels, loss_type=loss_type)
cdice_structures,hard_pred, labels,l,r = losses.per_structure_dice(logits, tf.one_hot(labels, depth=nlabels))
if weak_supervision and not include_bg:
####FIX THIS: Effectively want it to be -2 but
cdice_foreground = tf.slice(cdice_structures, [0, 1], [-1, nlabels - 2])
else:
cdice_foreground = tf.slice(cdice_structures, [0, 1], [-1,-1])
# cdice = tf.reduce_mean(cdice_foreground)
return nowd_loss, cdice_foreground
def evaluation(self, logits, labels, images):
'''
A function for evaluating the performance of the netwrok on a minibatch. This function returns the loss and the
current foreground Dice score, and also writes example segmentations and imges to to tensorboard.
:param logits: Output of network before softmax
:param labels: Ground-truth label mask
:param images: Input image mini batch
:return: The loss without weight decay, the foreground dice of a minibatch
'''
nlabels = self.config.nlabels
if self.params > 2:
output = logits[0]
else:
output = logits
mask = tf.argmax(tf.nn.softmax(output, axis=-1),
axis=-1) # reduce dimensionality
mask_gt = labels
tf.summary.image(
'example_gt',
self.prepare_tensor_for_summary(mask_gt,
mode='mask',
nlabels=nlabels))
tf.summary.image(
'example_pred',
self.prepare_tensor_for_summary(mask, mode='mask',
nlabels=nlabels))
tf.summary.image('example_zimg',
self.prepare_tensor_for_summary(images, mode='image'))
if self.params > 2:
mask = tf.clip_by_value(tf.cast(labels, tf.float32), 0, 1)
mask = tf.expand_dims(mask, -1)
tf.summary.image(
'example_masked_input',
self.prepare_tensor_for_summary(images * mask, mode='image'))
_, nowd_loss, _ = self.loss(logits, labels, images)
cdice_structures = losses.per_structure_dice(
output, tf.one_hot(labels, depth=nlabels))
cdice_foreground = cdice_structures[:, 1:]
cdice = tf.reduce_mean(cdice_foreground)
return nowd_loss, cdice