Table 5 Ablation study results for different components, evaluating on Robust-MIS 2019 with different labelled and unlabelled data amount, on binary segmentation.
From: SegMatch: semi-supervised surgical instrument segmentation
Method | Labelled: 1794 (30%) | Labelled: 5983 (+17K unlabelled) | ||
|---|---|---|---|---|
Mean Dice | NSD | Mean dice | NSD | |
SegMatch | 84.3 ± 2.3 | 80.2 ± 1.9 | 93.7 ± 1.02 | 93.6 ± 0.99 |
No adversarial | 80.6 ± 1.67 | 77.9 ± 1.74 | 90.6 ± 1.87 | 89.9 ± 1.96 |
No adversarial & No weak aug | 75.6 ± 1.62 | 73.7 ± 1.76 | 88.7 ± 1.59 | 87.6 ± 1.69 |
No adversarial & No strong aug | 71.5 ± 1.45 | 68.2 ± 1.37 | 85.5 ± 1.39 | 86.9 ± 1.42 |
