Skip to main content
. Author manuscript; available in PMC: 2022 Nov 14.
Published in final edited form as: Biomed Image Regist (2022). 2022 Jul 9;13386:103–115. doi: 10.1007/978-3-031-11203-4_12

Fig. 2.

Fig. 2.

Validation registration accuracy. In both self-supervised (MSE) and supervised (Dice) cases, SuperWarping the U-Net yields better mean Dice and endpoint error than the baseline (similar to VoxelMorph) and trains faster, requiring only 40 epochs to reach the final accuracy, which are 0.954, 0.152 (Ours–Dice), 0.906, 0.711 (Baseline–Dice).