model.py

import torch.nn as nn
from torchvision.models import efficientnet_b2, EfficientNet_B2_Weights


class EffnetB2(nn.Module):
    def __init__(self, num_classes=3):
        super().__init__()
        self.model = efficientnet_b2(weights=EfficientNet_B2_Weights.DEFAULT)
        for param in self.model.parameters():
            param.requires_grad = False
        in_features = self.model.classifier.get_submodule("1").in_features
        self.model.classifier = nn.Sequential(
            nn.Linear(in_features=in_features, out_features=num_classes)
        )

    def forward(self, x):
        return self.model(x)