Table 3 Comparative performance of the proposed models using ANN classifier versus CNN/ViT baselines.
From: Intelligent retinal disease detection using deep learning
Model | Trainable Parameters | Epochs | Validation Accuracy (%) | Training Time (min) |
|---|---|---|---|---|
DenseNet121 (Fully fine-tuned CNN) | ~6.9 M | 35 | 97.4 | 30.8 |
MobileNetV2 (Fully fine-tuned CNN) | ~2.3M | 30 | 98.0 | 11.9 |
ViT | ~85 M | 40 | 98.4 | 39.6 |
Model A (DenseNet121 with PCA) | ~0.157 M | 20 | 97.4 | 1.32 |
Model B (MobileNetV2 with PCA and DWT) | ~0.16 M | 20 | 96.6% | 1.35 |
Model C (MobileNetV2, DenseNet121 with PCA and DWT) | ~0.28 M | 20 | 98.2 | 2.32 |
Model D (MobileNetV2 with PCA) | ~0.16 M | 20 | 98.1 | 1.32 |
