Losses (`empanada.losses`)#

Confidence weighting of loss that’s efficient.

class empanada.losses.BCLoss(*args: Any, **kwargs: Any)[source]#

Defines the overall loss for a boundary contour prediction model.

Parameters

pr_weight – Float, weight to apply to the point rend semantic segmentation loss. Only applies if using a Point Rend enabled model.
top_k_percent – Float, fraction of largest semantic segmentation loss values to consider in BootstrapCE.

class empanada.losses.PanopticLoss(*args: Any, **kwargs: Any)[source]#

Defines the overall panoptic loss function which combines semantic segmentation, instance centers and offsets.

Parameters

ce_weight – Float, weight to apply to the semantic segmentation loss.
mse_weight – Float, weight to apply to the centers heatmap loss.
l1_weight – Float, weight to apply to the center offsets loss.
pr_weight – Float, weight to apply to the point rend semantic segmentation loss. Only applies if using a Point Rend enabled model.
top_k_percent – Float, fraction of largest semantic segmentation loss values to consider in BootstrapCE.

Losses (empanada.losses)#