Table 2 Analysis of the different proposed components on the DRISHTI-GS dataset
From: Self-supervised pre-training for joint optic disc and cup segmentation via attention-aware network
Method | OD | OC | ||
|---|---|---|---|---|
F1 | BLE | F1 | BLE | |
Baseline | 0.9334 | 9.07 | 0.8265 | 19.97 |
+ Standard Attention | 0.9388 | 8.98 | 0.8314 | 19.19 |
+ Multi-Scale Attention | 0.9488 | 8.17 | 0.8486 | 18.24 |
+ Aggregation Attention Module | 0.9433 | 8.21 | 0.8432 | 18.47 |
+ Self-supervised Pretrain | 0.9517 | 7.83 | 0.8636 | 15.72 |
+ Multi-Scale Attention + Aggregation Attention Module | 0.9652 | 7.01 | 0.8879 | 13.98 |
+ Multi-Scale Attention + Self-supervised Pretrain | 0.9788 | 6.28 | 0.9001 | 11.03 |
+ Aggregation Attention Module + Self-supervised Pretrain | 0.9763 | 6.34 | 0.8978 | 11.54 |
Ours (full) | 0.9801 | 6.21 | 0.9087 | 10.07 |