Table 3.
Segmentation model performance comparison using mean IoU scores.
| Model (W, H, C) | OpenEDS | HUSHH | ||||
|---|---|---|---|---|---|---|
| Sclera | Iris | Pupil | Sclera | Iris | Pupil | |
| Deep VOG (240, 320, 3) | - | - | 0.6806 | - | - | 0.8171 |
| EllSeg (240, 320, 1) | - | 0.8278 | 0.9157 | - | 0.6607 | 0.7632 |
| RITnet (640, 400, 1) | 0.0290 | 0.4663 | 0.7142 | 0.0674 | 0.4157 | 0.5404 |
| Our model (100, 100, 1) | 0.9743 | 0.9601 | 0.9281 | 0.9500 | 0.9352 | 0.9073 |
(W, H, C), optimal image size for each model. The model with the highest mean IoU scores, denoted in bold, indicates superior performance (higher scores represent better performance).