Table 4 Comparison of Teacher models selection on ternary classification on APTOS 2019 dataset.
From: Knowledge distillation-based lightweight MobileNet model for diabetic retinopathy classification
Model | Alpha (\(\alpha\)) | Trainable Parameter | Accuracy (%) | Precision (%) | Recall (%) | F1 Score (%) |
|---|---|---|---|---|---|---|
MobileNet | 0.25 | 279,635 | 94 | 89 | 88 | 89 |
0.50 | 950, 691 | 92 | 84 | 79 | 81 | |
0.75 | 2, 014, 195 | 92 | 87 | 76 | 79 | |
1 | 3, 470, 147 | 93 | 86 | 80 | 82 | |
MobileNetV2 | 0.35 | 724835 | 93 | 86 | 82 | 84 |
0.50 | 1, 016, 387 | 91 | 83 | 78 | 80 | |
0.75 | 1, 684, 131 | 93 | 86 | 82 | 84 | |
1 | 2, 552, 579 | 92 | 85 | 82 | 83 |
