Table 6.
Segmentation performance comparison of the proposed method before and after pruning noisy and mildly noisy GT labels from ISIC 2017 train and test sets
| Train pruned (Ntrain) | Test pruned (Ntest) | Jac | Dsc | Acc |
|---|---|---|---|---|
| None (2150) | None (600) | 0.807 | 0.880 | 94.779 |
| Noisy (580) | 0.817 | 0.889 | 95.536 | |
| Noisy + Mildly noisy (493) | 0.832 | 0.900 | 96.393 | |
| Noisy (2131) | None (600) | 0.806 | 0.878 | 94.723 |
| Noisy (580) | 0.815 | 0.887 | 95.518 | |
| Noisy + Mildly noisy (493) | 0.827 | 0.895 | 96.210 | |
| Noisy + Mildly noisy (1982) | None (600) | 0.802 | 0.875 | 94.658 |
| Noisy (580) | 0.812 | 0.885 | 95.398 | |
| Noisy + Mildly noisy (493) | 0.824 | 0.893 | 96.073 |
The values in bold are the highest scores