Table 14 Analysis of test data and features generated at the GAP and Output layers for clustering using Teacher and Student models (with and without knowledge distillation) for binary classification on the APTOS 2019 dataset.
From: Knowledge distillation-based lightweight MobileNet model for diabetic retinopathy classification
Data/Layer | Model | Silhouette Score | Davies-Bouldin Index |
|---|---|---|---|
Test Data | - | 0.0373 | 5.71 |
GAP Layer | Teacher | 0.3777 | 1.02 |
Student (KD) | 0.3305 | 1.23 | |
Student (w/o KD) | 0.5702 | 0.59 | |
Output Layer | Teacher | 0.5713 | 0.62 |
Student (KD) | 0.6118 | 0.51 | |
Student (w/o KD) | 0.6887 | 0.38 |
