Table 1.

Performance comparison of the three trained models on training, validation, and test datasets with respect to the Dice coefficient (higher is better), reliability (higher is better), and mean deviation from crack tip ground truth in millimeters (lower is better).