tf.keras.losses.circle

Computes the Circle loss.

tf.keras.losses.circle(
    y_true,
    y_pred,
    ref_labels=None,
    ref_embeddings=None,
    remove_diagonal=True,
    gamma=80,
    margin=0.4
)

It is designed to minimize within-class distances and maximize between-class distances in L2 normalized embedding space.

Args
y_true	Tensor with ground truth labels in integer format.
y_pred	Tensor with predicted L2 normalized embeddings.
ref_labels	Optional integer tensor with labels for reference embeddings. If None, defaults to y_true.
ref_embeddings	Optional tensor with L2 normalized reference embeddings. If None, defaults to y_pred.
remove_diagonal	Boolean, whether to remove self-similarities from positive mask. Defaults to True.
gamma	Float, scaling factor for the loss. Defaults to 80.
margin	Float, relaxation factor for the loss. Defaults to 0.4.

Returns
Circle loss value.