resnet_mlp.py

import torch.nn as nn
from torchvision.models import resnet50, ResNet50_Weights

class MLPHead(nn.Module):
    def __init__(self, in_features, hidden_dim, num_classes):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(in_features, hidden_dim),
            nn.BatchNorm1d(hidden_dim),
            nn.ReLU(inplace=True),
            nn.Linear(hidden_dim, hidden_dim),
            nn.BatchNorm1d(hidden_dim),
            nn.ReLU(inplace=True),
            nn.Linear(hidden_dim, num_classes),
        )

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

class ResNetMLP(nn.Module):
    def __init__(self, num_classes=10, freeze_backbone=True, hidden_dim=512):
        super().__init__()
        weights = ResNet50_Weights.DEFAULT
        self.resnet = resnet50(weights=weights)
        if freeze_backbone:
            for p in self.resnet.parameters():
                p.requires_grad = False
        for p in self.resnet.layer3.parameters():
            p.requires_grad = True
        for p in self.resnet.layer4.parameters():
            p.requires_grad = True
        num_features = self.resnet.fc.in_features
        self.resnet.fc = nn.Identity()
        self.mlp_head = MLPHead(
            in_features=num_features,
            hidden_dim=hidden_dim,
            num_classes=num_classes,
        )

    def forward(self, x):
        x = self.resnet(x)
        x = x.view(x.size(0), -1)
        x = self.mlp_head(x)
        return x