Tables 1 and 3 both report the performance metrics (Dice, IoU, Accuracy, Precision, Recall, F1) for the exact same set of existing methods (U-Net, Mask R-CNN, DeepLab, FCN, SegNet, UNet++, etc.) used for comparison.
However, the numerical values for these baselines are completely different between the two tables, with no explanation provided for the discrepancy.
For example, looking at U-Net:
Table 1: Dice = 0.92, IoU = 0.89, Accuracy = 0.94, Precision = 0.91, Recall = 0.95, F1 = 0.93.
Table 3: Dice = 0.87, IoU = 0.88, Accuracy = 0.91, Precision = 0.89, Recall = 0.92, F1 = 0.90.
Every single metric for U-Net (and all other models like Mask R-CNN, DeepLab, etc.) differs significantly between the two tables (e.g., Dice drops from 0.92 to 0.87; Accuracy drops from 0.94 to 0.91).
Since both tables serve as the benchmark comparison for the proposed method, which set of baseline numbers is correct? If Table 3 is the correct baseline, then Table 1 is fabricated. If Table 1 is correct, then the proposed method’s reported improvements in Table 3 (e.g., accuracy 0.95 vs U-Net’s 0.91) are artificially inflated because the baselines are incorrectly underreported. Please clarify which table contains the actual experimental results for the baseline models, as this inconsistency invalidates the performance comparison and the conclusion that the proposed method outperforms existing ones.