batch_kpis

Description

Loss and metric functions and classes that can be calculated per batch.

A good overview and collection can be found e.g. here: https://lars76.github.io/neural-networks/object-detection/losses-for-segmentation/ (the contained code samples are quite instructive but in tensorflow, thus not used here).

Classes

AbstractIoULike	General functions for intersection over union like calculation on binarized in- and output.
AbstractIoULoss	Shared settings for intersection over union based losses.
AbstractIoUMetric	Common properties of IoU calculation.
BalancedBCELoss	Balanced binary cross entropy loss.
BalancedPenaltyReducedFocalLoss	Balanced version of the penalty reduced focal loss from CenterNet.
BatchReduction	Aggregation types to reduce the 0th (meaning the batch) dimension of a tensor.
IoU	Calc sample-wise intersection over union (IoU) values output batch.
MaskRCNNLoss	Loss and associated data for a standard Mask R-CNN model.
Net2VecLoss	Simplified intersection over union as loss.
TverskyLoss	Calc Tversky loss (balanced Dice loss) for given outputs amd targets.
WeightedLossSum	Weighted sum of loss results.