Table 12 Ternary classification performance on APTOS 2019 datasets.
From: Knowledge distillation-based lightweight MobileNet model for diabetic retinopathy classification
Author and Year | Models | Trainable Parameters | Accuracy | Precision | Recall | F1-Score |
|---|---|---|---|---|---|---|
Rao et al., 202073 | InceptionResNet | 55, 900, 000 | 88% | 88% | 88% | 0.88 |
Kobat et al., 202274 | DenseNet + Cubic SVM | - | 93.85% | 90.90% | 80.60% | 83.78% |
Butt et al., 202241 | GoogleNet + ResNet-18 + SVM | - | 89% | 89% | 89% | 0.89 |
Athira et al., 202340 | ResNet50 | 25, 600, 000 | 94% | 94% | 94% | 0.93 |
Teacher Model | MobileNet | 279,378 | 94.18% | 94.23% | 94.18% | 94.18% |
Student without KD | Reduced parameter MobileNet | 71,491 | 79.09% | 85.09% | 79.09% | 81.03% |
Student with KD (Proposed Model) | Reduced parameter MobileNet | 71,491 | 93.09% | 93.03% | 93.09% | 93.07% |
