Binary and multiclass loss function for image segmentation with one-hot encoded masks of shape=(<BATCH_SIZE>, <IMAGE_HEIGHT>, <IMAGE_WIDTH>, <N_CLASSES>). Implemented in Keras.
All loss functions are implemented using Keras callback structure:
def example_loss() -> Callable[[tf.Tensor, tf.Tensor], tf.Tensor]:
def loss(y_true: tf.Tensor, y_pred: tf.Tensor) -> tf.Tensor:
pass
return loss
Loss function | Implementation |
---|---|
Weighted Tanimoto loss | https://github.com/maxvfischer/keras-image-segmentation-loss-functions/blob/master/losses/multiclass_losses.py#L8 |
Weighted Dice's coefficient loss | https://github.com/maxvfischer/keras-image-segmentation-loss-functions/blob/master/losses/multiclass_losses.py#L42 |
Weighted squared Dice's coefficient loss | https://github.com/maxvfischer/keras-image-segmentation-loss-functions/blob/master/losses/multiclass_losses.py#L74 |
Weighted cross entropy | https://github.com/maxvfischer/keras-image-segmentation-loss-functions/blob/master/losses/multiclass_losses.py#L107 |
Focal loss | https://github.com/maxvfischer/keras-image-segmentation-loss-functions/blob/master/losses/multiclass_losses.py#L150 |