models/mobilenetv2.py

import torch.nn as nn
from torchvision import models

class MobileNetV2Classifier(nn.Module):
    def __init__(self, train_base=False):
        super().__init__()
        self.base_model = models.mobilenet_v2(weights=models.MobileNet_V2_Weights.DEFAULT)
        for param in self.base_model.features.parameters():
            param.requires_grad = train_base

        in_features = self.base_model.classifier[1].in_features
        self.base_model.classifier = nn.Sequential(
            nn.BatchNorm1d(in_features),
            nn.Dropout(0.5),
            nn.Linear(in_features, 128),
            nn.ReLU(),
            nn.BatchNorm1d(128),
            nn.Dropout(0.5),
            nn.Linear(128, 1),
            nn.Sigmoid()
        )

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