kaira.losses.image.FocalLoss

Inheritance diagram for FocalLoss

class kaira.losses.image.FocalLoss(alpha=None, gamma=2.0, reduction='mean')

Bases: BaseLoss

Focal Loss Module for dealing with class imbalance.

This implementation works for both binary and multi-class problems. Focal loss addresses class imbalance by down-weighting well-classified examples, focusing training on difficult examples [Lin et al., 2017].

Methods

__init__	Initialize the FocalLoss module.
forward	Forward pass through the FocalLoss module.

__init__(alpha=None, gamma=2.0, reduction='mean')

Initialize the FocalLoss module.

Parameters:

• alpha (float or tensor) – Weighting factor for the rare class. Default is None.

• gamma (float) – Focusing parameter. Default is 2.0.

• reduction (str) – Specifies the reduction to apply to the output. Default is ‘mean’.

forward(inputs: Tensor, targets: Tensor) → Tensor

Forward pass through the FocalLoss module.

Parameters:

• inputs (torch.Tensor) – The input logits tensor.

• targets (torch.Tensor) – The target tensor.

Returns:

The focal loss between the input and the target.

Return type:

torch.Tensor