Table 3 Evaluations on the various meta-learning methods and loss functions to the baseline and proposed method.

From: Bidirectional meta-Kronecker factored optimizer and Hausdorff distance loss for few-shot medical image segmentation

Method

Training time (s)

Method—baseline

Training time (ms/1iter)

SSL-ALPNet + CE

809

–

40.01

SSL-ALPNet + CE + Boundary loss

1114

305

55.71

SSL-ALPNet + CE + Boundary loss + HDLoss

1416

607

70.81

SSL-ALPNet + MAML

899

90

44.95

Proposed method

1522

713

76.10

  1. The required memory usage during training (min–max), the required training time, and the time required for 1 iteration are compared. It is based on 20 epochs.
  2. Significant values are in [bold].
Back to article page