optax.losses.binary_dice_loss

optax.losses.binary_dice_loss#

optax.losses.binary_dice_loss(
predictions: jax.typing.ArrayLike,
targets: jax.typing.ArrayLike,
*,
smooth: jax.typing.ArrayLike = 1.0,
apply_sigmoid: bool = True,
) → Array[source]#

Binary Dice Loss convenience function.

Parameters:

  • predictions – Logits of shape […] or […, 1].

  • targets – Binary targets of shape […] or […, 1].

  • smooth – Smoothing parameter.

  • apply_sigmoid – Whether to apply sigmoid to predictions.

Returns:

Loss values of shape […] (batch dimensions only).