src/models/resnet.py

import torch
import torch.nn as nn
import torchvision.models as models

class ResNetClassifier(nn.Module):
    def __init__(self, model_name='resnet18', num_classes=5, pretrained=True, dropout=0.2):
        super().__init__()
        if model_name == 'resnet18':
            self.backbone = models.resnet18(pretrained=pretrained)
            self.feature_dim = 512
        elif model_name == 'resnet34':
            self.backbone = models.resnet34(pretrained=pretrained)
            self.feature_dim = 512
        elif model_name == 'resnet50':
            self.backbone = models.resnet50(pretrained=pretrained)
            self.feature_dim = 2048
        else:
            raise ValueError(f'不支持的ResNet模型: {model_name}')
        self.backbone = nn.Sequential(*list(self.backbone.children())[:-1])
        self.classifier = nn.Sequential(
            nn.AdaptiveAvgPool2d(1),
            nn.Flatten(),
            nn.Dropout(dropout),
            nn.Linear(self.feature_dim, 512),
            nn.ReLU(inplace=True),
            nn.Dropout(dropout * 0.5),
            nn.Linear(512, num_classes)
        )
        self._initialize_weights()
    def _initialize_weights(self):
        for m in self.classifier:
            if isinstance(m, nn.Linear):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
                nn.init.constant_(m.bias, 0)
    def forward(self, x):
        features = self.backbone(x)
        return self.classifier(features)
    def get_features(self, x):
        features = self.backbone(x)
        return features.view(features.size(0), -1)