The study evaluates the performance of the TM-CNN algorithm against human-reviewed standards for detecting junctions and terminals in magnetic structures. The absence of benchmarks necessitated a novel approach, utilizing independent image annotations to establish detection accuracy. In examining Bi:YIG films, the ideal stripe pattern, indicative of high spatial coherence, contrasts with the often labyrinthine patterns observed experimentally, which reflect lower coherence due to directional variability in stripe alignment. This highlights the complexities in defect formations inherent within the studied structures.
Evaluating TM-CNN for junction and terminal detections posed challenges due to the lack of benchmarks, leading us to use human-reviewed annotations as the gold standard.
In Bi:YIG films, the perfect stripe pattern is energetically stable, leading to large spatial coherence, while actual structures often display less organized labyrinthine patterns.
Collection
[
|
...
]