The structure of axial-free attention.

مواد مشابهة

• Performance comparison of the number of different axial-free attention layers.
  حسب: Hua Wang (13530)
  منشور في: (2025)
• Different aggregation ways. We have designed four aggregation approaches to fuse the outputs of axial-free attention layers, which are common in other network architectures. “**" denote significance levels of , from a two-sided paired t-test when comparing our best model with others. The best results are shown in bold.
  حسب: Hua Wang (13530)
  منشور في: (2025)
• Performance with different loss functions. “ ” and “**” denote significance levels of and , respectively, from a two-sided paired t-test when comparing our best model with others. The best results are shown in bold.
  حسب: Hua Wang (13530)
  منشور في: (2025)
• Dice and Jaccard index change with/without supervision. “**" denote significance levels of , from a two-sided paired t-test when comparing our best model with others. The best results are shown in bold.
  حسب: Hua Wang (13530)
  منشور في: (2025)
• The effect of different types of position encoding. We compared the performance of BTFormer with position encoding and without position encoding. “ ” and “**" denote significance levels of and , respectively, from a two-sided paired t-test when comparing our best model with others. The best results are shown in bold.
  حسب: Hua Wang (13530)
  منشور في: (2025)