cnn.py

import torch
import torch.nn as nn

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")


class CNN(nn.Module):
    def __init__(self, name, input_channels, output_channels, output_dim):
        super(CNN, self).__init__()
        self.name = name
        self.conv1 = nn.Conv2d(input_channels, output_channels, kernel_size=11, padding=1).to(device)
        self.conv2 = nn.Conv2d(output_channels, output_channels, kernel_size=3, padding=1).to(device)
        self.conv3 = nn.Conv2d(output_channels, output_channels, kernel_size=3, padding=1).to(device)
        self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
        self.relu = nn.ReLU(inplace=True)
        self.fc = nn.Linear(output_channels * 30 * 30, output_dim)
        self.sm = nn.Softmax(dim=1)

        self.initialize_weights()

    def initialize_weights(self):
        for module in self.modules():
            if isinstance(module, nn.Conv2d) or isinstance(module, nn.Linear):
                # init random weights
                nn.init.kaiming_normal_(module.weight)

                # init random bias
                if module.bias is not None:
                    module.bias.data.zero_()

    def __str__(self):
        return f"{self.name} CNN model"

    def forward(self, x):
        x = self.conv1(x)
        x = self.relu(x)
        x = self.pool(x)

        x = self.conv2(x)
        x = self.relu(x)
        x = self.pool(x)

        x = self.conv3(x)  # Applying the third Conv2d layer
        x = self.relu(x)
        x = self.pool(x)

        output = self.fc(torch.flatten(x, start_dim=1))
        return self.sm(output)