Figure 2.

Architectures of CNN and ResNet18 models. (A) Schematic showing the typical architecture of a CNN which involves the input being passed through multiple hidden layers, including convolutional and pooling layers. The weights from the hidden layers are flattened into a single vector and passed through the fully connected layer for classification. (B) Schematic showing the architecture of the ResNet18 model. The first convolutional layer (conv1) with 64 kernels sized 7 × 7 sliding with stride 2. Subsequent convolutional layers are divided into 4 blocks with 2 layers each. The layers employ kernels of size 3 × 3, with layers in the second block (conv2_x) having 64 kernels each, the third (conv3_x) having 128 per layer, the fourth (conv4_x) having 256 per layer, and the last (conv5_x) having 512 per layer. Weights from the last convolutional layer are average-pooled and passed through the fully connected layer for classification.