Table 2 Specifics of the CNN architectures applied and tested in this study.
From: Diagnostic performance of convolutional neural networks for dental sexual dimorphism
Model | Size (MB) | Parameters (M) | Depth | Image size | Hyperparameters | ||||
|---|---|---|---|---|---|---|---|---|---|
Optimization algorithm | Batch size | Momentum | Weight decay | Learning rate | |||||
DenseNet121 | 33 | 8.1 | 121 | 224 × 224 | SGD | 32 | 0.9 | 1e-4 ~ 1e-6 | Base Ir = 0.001 Max Ir = 0.00006 Step size = 100 Mode: triangular |
ResNet50 | 98 | 25.6 | 107 | 224 × 224 | |||||
ResNet101 | 171 | 44.7 | 209 | 224 × 224 | |||||
Xception | 88 | 22.9 | 81 | 299 × 299 | |||||
InceptionV3 | 92 | 23.9 | 189 | 299 × 299 | |||||
InceptionResNetV2 | 215 | 55.9 | 449 | 299 × 299 | |||||
VGG16 | 526 | 138.4 | 16 | 224 × 224 | |||||
MobileNetV2 | 14 | 3.5 | 105 | 224 × 224 | |||||
