highly significantp < 0.0001
Based on 20 independent runs, a paired t-test on accuracy values confirms a highly significant difference (t(19) = 55.37, p < 0.0001), providing strong evidence against the null hypothesis.
Based on 20 independent runs, a paired t-test on accuracy values confirms a highly significant difference (t(19) = 55.37, p < 0.0001), providing strong evidence against the null hypothesis.
As the ATT module is introduced into deeper stages (e.g., Stage 1 + 2, Stage 1 + 2 + 3, and all stages), a clear trend of diminishing returns is observed: while the performance remains superior to the baseline Swin-Tiny without ATT (84.83% accuracy), it gradually decreases.