Table 8.
Performance comparison between a W-Net and a U-Net configured to have a comparable amount of weights.
| # Params | DRIVE | CHASE-DB | HRF | ||||
|---|---|---|---|---|---|---|---|
| AUC | DICE | AUC | DICE | AUC | DICE | ||
| “Big” U-Net | 76,213 | 98.00 | 82.53 | 98.29 | 81.09 | 98.15 | 80.73 |
| Little W-Net | 68,482 | 98.09 | 82.78 | 98.44 | 81.52 | 98.24 | 81.05 |
| W-Net vs U-Net | 7731 |
+0.09 p<0.05 |
+0.25 p<0.05 |
+0.15 p<0.05 |
+0.43 p<0.05 |
+0.09 p<0.05 |
+0.32 p<0.05 |
W-Net achieves higher performance, despite having slightly less parameters. Statistically significant results marked bold.