Table 3 Architectural configurations and training hyperparameters across various trained models, including a custom baseline-CNN model and pretrained architectures (VGG16, ResNet101, DenseNet121, InceptionV3, and ResNet50).
From: Artificial intelligence derived grading of mustard gas induced corneal injury and opacity
Parameters | Baseline Model | VGG16 | ResNet101 | DenseNet121 | InceptionV3 | ResNet50 |
|---|---|---|---|---|---|---|
Convolutional Blocks | 3 (with 2 convolutional layers) | Predefined | Predefined | Predefined | Predefined | Predefined |
Filters | First block: 32 | |||||
Second block: 64 | ||||||
Third block: 128 | ||||||
Kernel size | 3 × 3 | |||||
MaxPooling | 2 × 2 | |||||
Zero Padding | No | Yes | Yes | Yes | Yes | Yes |
Dense Layers | 2 | 2 | 2 | 2 | 2 | 2 |
Neurons in Dense Layers | 1024,4 | 1024,4 | 1024,4 | 1024,4 | 1024,4 | 1024,4 |
Dropout Ratio | 0.5 | 0.5 | 0.5 | 0.5 | 0.5 | 0.5 |
Learning Rate | 1.00E-03 | 1.00E-02 | 1.00E-04 | 1.00E-04 | 2.00E-03 | 1.00E-04 |
Epochs | 100 | 100 | 100 | 100 | 100 | 100 |
Batch Size | 64 | 64 | 64 | 64 | 64 | 64 |
Optimizer | Adam (weight decay: 1e-5) | Adam (weight decay: 1e-5) | Adam (weight decay: 1e-5) | Adam (weight decay: 1e-5) | Adam (weight decay: 1e-5) | Adam (weight decay: 1e-5) |
