Losses

The losses submodule implements loss function to be used in segmentation tasks.

segmentation.losses.BoundaryCELoss(numClasses, use_3D=True)

Wrapper function for boundary_crossentropy. Args:

numClasses: number of classes

Returns:

boundary_crossentropy function

segmentation.losses.DistancedBoundaryCE_Loss(numClasses, use_3D=True)

Wrapper function for dist_boundary_crossentropy. Args:

numClasses: number of classes

Returns: dist_boundary_crossentropy function

segmentation.losses.DistancedCELoss(numClasses, alpha, use_3D=True)

Wrapper function for dice_categorical_cross_entropy. Arguments:

numClasses: number of classes alpha: parameter to weight contribution of dice and distance-weighted categorical crossentropy loss

Returns:

categorical_cross_entropy function

segmentation.losses.DistancedRegionCELoss(numClasses, use_3D=True)

Wrapper function for dist_region_crossentropy_loss.

Args:

numClasses: number of classes

Returns:

dist_region_crossentropy_loss function.

segmentation.losses.ExpLogLoss(numClasses, gamma=1, use_3D=True)

Wrapper function for exp_log. Arguments:

numClasses: number of classes alpha: parameter to weight contribution of dice and distance-weighted categorical crossentropy loss gamma: exponential of logaritmic dice and CE

Returns:

categorical_cross_entropy function

segmentation.losses.FocalLoss(numClasses, alpha, use_3D=True)

Wrapper function for dice_focal. Arguments:

num_classes: number of classes alpha: parameter to weight contribution of dice and distance-weighted categorical crossentropy loss

Returns:

dice_focal function

segmentation.losses.JaccardContour_Loss(numClasses, use_3D=True)

Wrapper function for Jaccard Index. Args:

numClasses: number of classes

Returns:

mean jaccard weigthed by class

segmentation.losses.MeanDiceLoss(numClasses, use_3D=True)

Wrapper function for mean_dice. Args:

numClasses: number of classes

Returns:

mean dice weigthed by class

segmentation.losses.RegionCELoss(numClasses, use_3D=True)

Wrapper function for region_crossentropy_loss Args:

numClasses: number of classes

Returns:

region_crossentropy_loss function

segmentation.losses.WeightedDiceBoundaryLoss(num_classes, alpha, use_3D=True)

Wrapper function for multiclass_weighted_dice_boundary_loss. Args:

num_classes: number of classes alpha: parameter to weight contribution of dice and boundary loss

Returns: multiclass_3D_weighted_dice_boundary_loss